<TITLE: System Components for Video on Demand: Architecture and Implementation
ACADEMIC DOMAIN: technology
DISCIPLINE: information technology
EVENT TYPE: doctoral defence discussion
FILE ID: UDEFD080
NOTES: continuation of UDEFP080

RECORDING DURATION: 140 min 30 sec

RECORDING DATE: 3.6.2004

NUMBER OF PARTICIPANTS: unknown

NUMBER OF SPEAKERS: 4

S1: NATIVE-SPEAKER STATUS: Romanian; ACADEMIC ROLE: research student; GENDER: male; AGE: 24-30

S2: NATIVE-SPEAKER STATUS: Swedish; ACADEMIC ROLE: senior staff; GENDER: male; AGE: 31-50

S3: NATIVE-SPEAKER STATUS: Norwegian; ACADEMIC ROLE: senior staff; GENDER: male; AGE: 31-50

S4: NATIVE-SPEAKER STATUS: unknown; ACADEMIC ROLE: senior staff; GENDER: male; AGE: 51-over

SU: unidentified speaker

SS: several simultaneous speakers>



<PRESENTATION UDEFP080 by S1>

<S2> yes thank you er , <NAME S1>'s thesis deals with the architecture and and the implementation of video on demand so-called V-O-D system and er as you heard there with a with such a system the television viewer can get a a video programme typically a movie but it could can be a game also or another content delivered to his television from a central video server over the telecommunication line , and er and er i think it's the topic is is highly relevant , at er at this moment because as as as we know er the the broadband infrastructure is starting to be there there is a heavy roll-out of ADSL , er connections it's it's one of the only things that is really profitable now in the telecommunication industry and er and also there is a set-top boxes er start also to be there and as a as a current finnish example we heard of a foreign example in the in the speech earlier but a current finnish expa- example is the song networks that is , has started now to recently be to provide internet television over internet er er or or by using seven megabyte per second ADSL connections in the helsinki region the subscribers at that have their set-top boxes , so so this is is er is very interesting and very timely topic , and er but however we of course have these er video demand systems er in in in internal systems so we have them in hotels everybody is is surely know about that also if you are are flying the new the new er modern big airplanes to over oversees you and er longer flights you can have have er and if you are especially if you are in the business class you can have have er the the movies there delivered to your to your set to your seat and you can select what what movie movie you want to to to watch so there there there are special systems but er as mr <NAME S1> points out in in in his thesis here er it's er , it's er because of these current technical er mostly interoperability problems it's it's er hard to to find really usable business model for for offering video on demand service to large number of customers , and er and as we heard there the main focus of the work is here here is to devise solutions for just for improving that for er improving the interoperability and compatibi- compatibility and in the working of the various components that we have in a in a system and er and these these solutions may get possible to combine different server and and client environments better than than er what is now the case and er and it ma- makes it also possible for old and new generations of of equipment to work together er because we have to live with the fact that er that we have a bigger server (xx) er some of them are bit older some are bit newer and er they don't have they shouldn't be abl- have to replace the older ones , and er , this er makes it possible to , also the the calculation in this thesis makes it possible to dimension the the video servers in in the right way so it's so we know know how how many , s- er machines we have to put in there to serve a certain amount of users , of course there are are others other problems also that are not really attacked er touched upon in this thesis that are slowing slowing down now the roll-out of the of the video on demand services and er one such problem is is er that the current I-P networks don't really guarantee a qual- a level of quality of service that er is needing from for streaming in a stabile fashion say four megabyte per second to the sub- subscriber because 'cause er as many of you know you you are going through many er switch switches and serv- servers before you come from the serv- from the starting server to the end user and er there are problems to keep up this er stabile er stabile bit rate , and and another that might might be even even more important now is the the fright for a fear for for for that makes the current content owners doubtful if if they dare really to pa- put their their content their movies games available online , er just to end up these opening remarks i i would point out like to point out that that er <NAME S1> has put in a very substantial amount of work in this thesis and and er because they are really based on 12 12 internationally refereed scientific papers and er of which three are journals and three er nine conference manuscripts and and this cle- clearly is more than what what is is normally the case in doc- doctoral dissertations , so th- with this i i close my opening remarks and we can go on to the more detailed @@ review of the thesis <P:09> and er i in the start there the custom is that that er there are more general remarks here in the beginning and then we are go going more more say page by page into into the into a more detailed review and er , the my first question is about the how how do you think that this relate to er to to the kind of er systems that that you can now already buy from the shops where you can store er the the the television broadcast li- like you are you are recording on on the video tape but there there are digital er digital disks in in in the set-top boxes and and in that way you can have a kind of a video on demand of course you have to re- restrict yourself to the broadcasted programmes but anyhow you can in a way convenient way record them and er and then go back to the recording on their hard-disk and and you have had these p- these systems like like T-I-V-O in in united states for for s- several years and now they are coming here or you can get them from for example nokia has such a set-top box where you have have this lo- local storage capacity er do you think that thi- this er will in a way in a restricted way er offer the same services to the customer that you can get from a real video on demand system </S2>
<S1> er in a way yes in a way no the we can say that the same functionality exists today in er VCRs so er of course the er these er boxes with er hard-disks are er more convenient way of er doing you know in the same thing just put video tape to record the show at a certain hour and then watch it later but i would say that this functionality today does not stop movie rental shops to exist so in the same way er set-top boxes will with hard-disks will not stop video on demand <S2> [mhm mhm mhm] </S2> [(xx)] and er it it can also be some the the service can also be in a way (xx) so it can be a price differentiation so there can be some maybe some er dedicated movie channels that will send movies overnight and something like this so you can if you can er let's say er choose from today what movie you would like to see er tomorrow night you can make this <S2> [mhm] </S2> [choice] and maybe there is capacity overnight to send this movie [to you] <S2> [mhm] mhm </S2> so in a way the the the functionality of er these set-top boxes and hard-disks is greater than er VCRs today but we (xx) <S2> mhm yeah </S2> video on demand </S1>
<S2> mhm , yeah i i think you are correct in that my second mo- more er question more general question is about now if you want to scale up really er the video on demand system that you have devised here in the , describing your thesis er to serve say hundreds of thousands of people or even million of peo- millions of people er i don't really find here ideas how how this topology of of servers should be constructed er there is for example there has been proposals about content delivery networks special networks where you have er er so-called edge servers close to the user to the users and and you will copy content from the central servers to these edge servers and then then in that way er re-distribute the load er could you comment on on this or a way of scaling up really to be a the the same kind of server that you have that the broadcasting companies now that can cover ha- millions of of of of of viewers </S2>
<S1> i think that to create such a system can be in a way split in two parts , er the basic er the basic er servers the basic network infrastructures and by basic servers i would say er in a way (dumb) video on demand server which is able to to send a movie based on some request and then the the network infrastructure that to it's able to get such a (xx) , and then the second thing would be some kind of application (network) intelligence that would gather data from this system and er give the commands which movie has to be transferred from the central servers to edge [(user servers)] </S1>
<S2> [mhm] mhm yeah yeah , but er have you scenario how fast this will happen really in in fact and and where are we now and and how long will it take before it is rea- rea- really reality for example in a country like finland </S2>
<S1> i think in a way this this would be a a big step forward for the from the let's say the beginning scenarios of V-O-D so in i think that in the beginning there will be local companies who will who will er have video on demand servers and these local companies will probably be telephone operators and cable operators which now (xx) content and then we then have these V-O-D servers in their network in a way and you will get the the movie from the same provider from where get you TV <S2> mhm </S2> and er go going in a way globally to that you as the user can also (set) V-O-D service from some from some (xx) i think it's a it's a huge step forward <S2> mhm </S2> in a way <S2> mhm </S2> because the the local company would would need to to have er agreements with the global distributors and then they also need to have the the network infrastructure <S2>  mhm </S2> and er in a way the local er the global distributor can be seen as a competitor for the global <S2> mhm </S2> for the local distributor so then it's maybe this also needs to be regulated <S2> mhm yeah </S2> but i think that that future looks very nice because i think there are many not so known movies that will be not available from local <S2> [mhm] </S2> [operators] in the beginning </S1>
<S2> yeah yeah , yeah that was my general remarks if you want to go on here @@ </S2>
<S3> thank you very much and er dear mr <NAME S1> ladies and gentlemen we have decided to er split our comments so i will now have some some general comments and questions on your methods and on the overall and based a little bit on your presentation also er then we will be going through your thesis er chapter by chapter and er we will interact as opponents also so you'll get questions from both of us before we sum up at the end , now this er this work covers very important topic of video on demand you have er shown a complete system you've also identified problems and you have solved them and er for this you have made a validation implementation i'll just refer to that as your validation system er you have also identified problems that are outside the scope of the thesis er which is good it's being clear and your contributions have also been clearly identified , er i'll have some general comments on the er thesis and also i will have a small discussion on the topics that you have not covered and you chose not to cover just to have the discussion so er i really look forward to the discussion and and your answers , first of all i i think you gave a very good general definition of er video on demand and digital TV and interactive TV er i'd like to ask a a line of questions more on the definition of video on demand content , now from er from my perspective er there is a variety of different content that you can get on demand and of course video on demand assumes that you'll get some kind of video , what would you say if you wants to have other types of new digital media is this possible or is it limited to video this video on demand concept </S3>
<S1> i think that er other types of media are also er also important that can have in a way the video on demand framework can cover also other medias but in in some er in i think that the problems that are in in the video video on demand are in a way little bit more difficult from er problems (haunting) some other media for example er i think that in a way we can say that the current er web browsing is a media on demand <S3> mhm-hm </S3> and then there are also problems but i think that due to to the nature of those problems and to the nature of that media they are in a way easier to solve than the video on demand </S1>
<S3> in what way for instance if you take a a radio station er which have been very popular to stream over the internet they have streaming media for radio stations why is that simpler than a video on demand system </S3>
<S1> er i think that it's simpler er in a way because of er the the bandwidth required for that is much much lower </S1>
<S3> okay , to continue on the new digital media you were er mentioned digital TV interactive TV and video on demand and you put it nicely into place , what is interactivity more than an application do you see any futuristic possibilities for using interactivity while we are resting in our couch </S3>
<S1> er . so to to give a let's say very er futuristic scenario er it would be probably something that interacts in the real time with the real TV content <S3> mhm-hm </S3> where for example several users would choose in a way the how a let's say a big show or something er the ways to go on </S1>
<S3> mhm , if in er if you have a video recorder and er a remote control DVD player you are used to functionalities like start stop er fast forward rewind er DVDs even jumping from a chapter to a chapter or er is this functionality going to be present in video on demand or is there a special category of functionalities that's important for video on demand </S3>
<S1> er i think that this functionality will be present in video on demand maybe not from the beginning and that it might be that this functionality will be in a way associated with the price <S3> mhm </S3> so this can be either that er user is somehow er limited towards certain amount of this fast forward <S3> mhm </S3> and er basically i don't see jumping from from chapter to another as very let's say very difficult to obtain functionality but let's say more or less fast forward and (xx) could be something that requires extra resources </S1>
<S3> mhm , er resources where where where does it require these [resources] </S3>
<S1> [er] depending on how this is implemented it can require resources in er network <S3> [mhm] </S3> [set-top box] and (so) <S3> mhm </S3> it's i think that maybe all cases it requires extra resources (in the set-top box) </S1>
<S3> so er you mentioned VCRs and and TIVO system was mentioned , what about this personal video recorder or the PVR is that actually a video on demand system with a storage so you can stop and it will actually continue would you can play it back later and do you foresee things like this </S3>
<S1> er yes if the content is local on a hard-disk then you should have the same functionalities that are on DVD </S1>
<S3> this is maybe one of the problems with er video on demand and the fear of piracy because in in traditional video on demand thinking er the content is not actually stored it's just played and then it disappears right <S1> yeah </S1> or you think er it's possible to er hack the system and and store it </S3>
<S1> er yes i think that as a general opinion from (xx) i think that everything can be hacked <S3> mhm-hm </S3> the and the most important thing is to to limit the consequences , i think that broadcasters and everybody the the content producing industries are willing to accept certain degrees of piracy <S3> mhm-hm </S3> but then er i think that for them limiting is is very very important and er the i think that it's not so important to to make a system completely er hack-proof but i think that maybe with a system where er hacking it's not er is not so cost-effective anymore i think that that would be a very good solution so if to to have and get the movie would would have a such a let's say tremendous cost and then possibilities of distributing this would be almost none then it doesn't make sense (xx) </S1>
<S3> is this maybe the same as er video recording a TV programme and and letting your mother-in-law look at it also that's not really hacking is it </S3>
<S1> er , yeah it that's that's not hacking <S3> mhm </S3> i think based on the law that's not hacking </S1>
<S3> @so but of course@ it's it's phenomenal when it's digital because er what you would do is you would of course hack it and then put it on a peer-to-peer network </S3>
<S1> yeah i mean you cannot put a VCR tape <S3> [no] </S3> [on a] peer-to-peer <S3> [no] </S3> [network] otherwise <S3> @@ </S3> maybe taping would be forbidden </S1>
<S3> it's much more difficult , er in the interactivity er you touched on your er brief talk the user interface what is the user interface that you foresee </S3>
<S1> er something based on pushing buttons or maybe something based on voice </S1>
<S3> so a traditional remote control that maybe would be added functionality or <S1> yeah </S1> voice </S3>
<S1> so the er the i don't see much er a remote control with the kind of (xx) device and then i think that in a way the traditional buttons on remote control with the four directions the okay and cancel <S3> mhm </S3> that should be <S3> that [should be enough] </S3> [enough for] inter- interface at least in the beginning </S1>
<S3> mhm mhm i probably agree , and er you gave a very nice discussion on on game consoles or media PC or or television but er isn't it true that most households have all three and will always have all three of these er terminals </S3>
<S1> er , i do not think that so many households have also game console and media centre PC and also digital TV </S1>
<S3> but do you agree that most have TVs and most have PCs , but they don't use their PC as a TV and they don't use their TV as a PC <S1> yes </S1> don't you think if you put these together this would change radically </S3>
<S1> er i think it it might be i think that this interactive TV in a way is er a concept difficult to sell in market alone so i think that it's much is that this comes from from something that already exists for example as an extra feature from game console that that would be in a way that would force the interactive TV into houses <S3> mhm </S3> i think that that would be very nice and it would be very very good to to do something (xx) game with er same with the the PC technology <S3> mhm </S3> the the problem on the PC might be this piracy and hacking but it would be probably very nice to allow people with a PC in a way to (they) use some of the features of interactive TV <S3>  mhm-hm </S3> and then maybe they get hooked and then they go to the (xx) </S1>
<S3> i think you might be right , i'll i'll slightly change a little bit i i would like to talk er in general about your er validation system , your validation system proves that you have solved the problems and shortcomings of the video on demand system and you actually have a a complete system working , right , now the problems that you solve seem in some way to be er the need for extensions so extensibility is one of the solutions , but isn't the danger that your solutions which are extensions and partial solutions of interfaces isn't it a problem that this might just restart the problem that when you have this system and you go one step further and want to extend it you need to do all the work again </S3>
<S1> er there is some work to be done if we want want to extend the the system and the solution er that i proposed tried to in a way to minimise that thing , er to to say that er my extensions solve all the problems would be not true because i don't think that the research in extensions stops here but i think for th- the extensions that i made solve all the problems that i foresee <S3>  mhm-hm </S3> (at this point) and er about the (way) extensions to be to be RTSP libraries the the applications in which this library is used in a web prove that these extensions are quite general </S1>
<S3> so you would say that er you've actually made er not only extensions and s- partial solutions but maybe a generic framework of which it's possible to build further extensions for instance if you wanted to add er one new extension or one new functionality what impact would that have on your complete architecture is it possible or would you have to redo the architecture </S3>
<S1> er the i would refer to the the figure on page <P:10> on page 26 , er in a way the (xx) on the extension [so] </S1>
<S3> [that's] that's figure 2.1 right </S3>
<S1> yes <S3> yeah </S3> er if for example we have to (xx) an extension to RTSP that has to be communicated to the multimedia application <S3>  mhm-hm </S3> then we we need to to modify the additional library layer and we also need to modify the high level interface and take care of this extension in the multimedia application <S3> mhm-hm </S3> because in the multimedia application need the information there is actually no other way , to send it <S3> mhm </S3> but er the integrated library does not need to be modified <S3> mhm-hm </S3> and this is for er for the extensions that is allowed but they kind of have this (this kind of) <S3> mhm </S3> which only allows only extension (clients) it does not allow new commands <S3> okay </S3> to add new commands in the integrated library would be , needed to be slightly modified <S3> okay [that's good] </S3> [but i would] say that for actually for any extension that needs to be done the additional library layer may need to be modified (to take care about) </S1>
<S3> mhm-hm so it's it's a partial generic and and of course i i understand it's difficult to answer because we don't know what these extensions or new functionalities are , any of the these ideas in the in the complete validation system er have they been commercialised or are they actually in use </S3>
<S1> er i think that maybe the the most advanced er er the most advanced ideas have not been commercialised </S1>
<S3> mhm , would you say that er in the validation which of course is a result of all your work and is the integration of all your work can you make any distinctions of what parts of this is actually theoretical work and what part is applied and only dealing with interfaces and maybe er to be a little bit rude black boxes </S3>
<S1> yeah er i think that the the theoretical work is er mostly about finding a solution to a certain set of requirements and this solution has to be in a way designed and er and thought about and then the the implementation work comes to validate this solution so after the design process er people had a an id- an idea about how this would be in practice <S3> mhm </S3> and then then with the implementation we tried to to check if that idea about how this (is going to work) is correct </S1>
<S3> mhm , if you split the er the er work load between the two what would you say the the split would be </S3>
<S1> i would say that maybe that er the initial design and the whole let's say design process would be maybe 25 per cent of all and the rest would be implementation </S1>
<S3> mhm mhm good that's good to know </S3>
<S1> but i would say that there were also quite many ideas that came up in the implementation <S3> [mhm] </S3> [on the] in a way revalidating like like feedback <S3> [feedback] </S3> [revalidating] our ideas based on our implementation </S1>
<S3> okay , er in your thesis you you very clearly state that er the the piracy issues or the issues with intellectual property management protection or digital rights systems digital rights man- management er is not in the scope of the thesis but anyhow you do say that er one reason or motivation for this is that the technical solutions for these things are really solved er can you comment on that i think that's a little bit optimistic </S3>
<S1> er to a certain degree there is it is indeed optimistic but then er there is er currently encryption in digital TV and er many er TV let's say premium digital TV channels are encrypted and i think that works quite well and i think that that solution could be also in a way er moved and adapted to video on demand </S1>
<S3> so this this you think this is even viable for set-top boxes which now come with storage capabilities </S3>
<S1> er yes , you can actually store the content encrypted and decrypt it in let's say to to have er one solution would be that the content is er decrypted er just before the decoding and decryption module and the decoding decoding module is actually in the same functional unit <S3> [okay] </S3> [in this] way the the content would travel in encrypted form (xx) i think that that's very (xx) </S1>
<S3> okay i'd like to turn to er a question on page 12 , page 12 you have er a sentence saying that <READING ALOUD> the current technical problems of video on demand make very difficult to find a valid business model for offering V-O-D services to mass markets </READING ALOUD> er i'd like you to try to explain to me what is a valid business model for video on demand </S3>
<S1> er , er the er er valid business model would be er something where the the users would like to have the service and they would think that the the price is right and they would see the benefits of it and the service provider would er see this and it would actually make sense for the service provider to provide the service , and er in particular in er video on demand i think that today there are technical solutions to all problems <S3> mhm-hm </S3> but the the price is it's in a way very very high so of course we can have video on demand to pay if you are prepared to pay let's say one thousand for the set-top box and thirty euros for the for each movie but that's not that doesn't make a valid business model because there would be so few people to to really want to to who have such a thing , i would say that a valid business model would come when offering , when the service provider could offer video on demand to let's say many people above a er certain per cent so there is in a away mass market and when the and when the price of of of video on demand movie would be er comparable with er rental shop movie rental shops </S1>
<S3> okay so i i'd i sort of agree with your answers but i'd like to be even more concrete if i come to you and i have a deal with the warner brothers i have 500 movie titles which i have the intellectual right er property rights for , based on your validation model your implementation what do i have to do to be able to make money that's a valid business model <SU> @@ </SU> i wanna make a lot of money on my 500 movies [is it possible] </S3>
<S1> [er] er if you have a network </S1>
<S3> okay so i need a network </S3>
<S1> yes er you would need er to to have a some kind of encryption to have er some kind of protection <S3> mhm-hm </S3> to this [er] </S1>
<S3> [that's] a piece of software i can buy </S3>
<S1> er i think it's er (xx) box <S3> okay </S3> which inside and er then you need a a number of users with access , able to pay and they need access to the network and probably these users would er need a set-top box . and </S1>
<P:06>
<S3> so the question is it possible or is it not possible today </S3>
<S1> er i think it is possible but it's very [difficult] </S1>
<S3> [mhm] that's er that's what i thought 'cause there aren't many existing businesses in the world do there are there <S1> yeah </S1> are there there's no business in in finland that offers video on demand over broadband networks </S3>
<S1> as far as i know there is not </S1>
<S3> in france the example from france are they actually offering video on demand or just TV channels </S3>
<S1> i think they are only T it's only TV channels </S1>
<S3> mhm </S3>
<S2> i think it's it's the same as as this song networks in helsinki so they are just staying (with) the <S3> mhm </S3> TV <S3> mhm </S3> more or less the real-time </S2>
<S3> mhm so not on demand , okay erm are you aware of or do you know of a company called er envivio </S3>
<S1> er yes i <SIC> heared </SIC> about it </S1>
<S3> you heard about it do you er do you know about their solution for video on demand </S3>
<S1> erm i have brief idea </S1>
<S3> okay then i i'll i'll drop that line of questions , i'd like to turn now to er metadata what role does metadata play in in video on demand </S3>
<S1> i think that metadata is er has a very important or (xx) demand because people are going to select the movie mainly on metadata and maybe the the other factor would be the trailer what the first (xx) people would have about a movie would be this metadata which i would say that includes (xx) actors and (xx) maybe director (xx) this description of the movie </S1>
<S3> and in your validation implementation where would this be how how would you make this metadata known to the potential customers </S3>
<S1> i think of a good way to do it is (to go to) web pages <S3> mhm-hm </S3> so this would be a method of us doing this would be is described in er the last chapter <S3> mhm-hm </S3> and er where you actually see , this metadata comes from database and from the database the web pages are generated which are explained to <S3> mhm </S3> (to have this) </S1>
<S3> does this mean you actually do have to have a PC also in in addition to the TV </S3>
<S1> er not necessarily if the set-top box that you are using is able to display web pages </S1>
<S3> so you would have a channel dedicated to advertising your buyer metadata </S3>
<S1> er not necessarily it it can be that you have a you start from web page and then you you go to some page where you can see a list of movies then you can have some selection and then you can check movies of this (really this) metadata and when the user selects a movie he just can press the button and then the the playing system starts automatically <S3> mhm </S3> so the the selected movie the U-R-L is somehow sent to the to the player to the (xx) playing part V-O-D player that's that </S1>
<S3> mhm </S3>
<S2> if you mind i have just a follow follow-up question there what do you think because you are watching this video on demand on the same television as you are watching normal programmes and er there you have have this as you told us and as we know the electronic programme guide the the E- EPG how would this shouldn't this metadata this programme information be in a some way included in the same electronic programme guide that is known to the to the watchers or or or do you think they are totally separate </S2>
<S1> i think it's totally the solution to this is totally separated it's much better because this er EPG is actually sent continuously and the the whole concept is based on the fact that the current digital TV boxes don't have er so big memory <S3> mhm  </S3> and er total bandwidth to actually send this er this metadata over er this er DVD (xx) cables which make a <S2> [mhm] </S2> [EPG] would create so huge bandwidth that i think it it's not a , it's not (working as a) technical solution </S1>
<S2> yeah okay that's the transmission but i thought about the user interface so would (it present itself) to the user er by the same format it's it's another thing that it can come come over the over the the the internet </S2>
<S1> the the format that in which the EPG is is presented to the user i would say that it's more or less the problem of the set-top box as system is designed today and if you have a selection of five hundred movies i don't think it would be a very good solution to be only able to display them all in a way so i think that some selection and some filtering should be should be possible <S2> mhm </S2> [for] </S1>
<S2> [but] but you have a same situation when in in in you can have hundred of TV channels also they they must also be managed over over the E-P or the programme guide so and you can have er categorisation in ge- genres for example sport programmes and culture and so on and that makes it easy </S2>
<S1> mhm , i think it depends so the er i'm let's say i'm using the EPG in in my set-top box and then i get the list of channels and then i can scroll up and down and left to right in the future (above it) and er personally as a user i would not like to be to only have a list of five hundred movies that i can just scroll up and down and maybe select something and see more information i would like to be able to to filter that list if i want to <S2> mhm mhm </S2> so if if filtering would be possible in the display of EPG <S2> [yeah] </S2> [that] would be a good [solution] </S1>
<S2> [(it's)] already there you can (load) it in some DVDs </S2>
<S3> you really need some way of search you would need some way of profiling you need some personalisation <S1> yeah </S1> i think that's very important , now (collective) service is also an an issue that's you er is out of the scope of the thesis but of course it's it's very important quality of service and if you introduce er er what i call quality of experience which you might call end-to-end quality of service er to what extent does your er validated system support quality of service and how does it support quality of experience more more importantly </S3>
<S1> er the the system was er designed to to (go over) let's say good methods so from network point of view there is no quality of service and er this assumption is based on on the fact that er at least in the beginning the the video on demand will will come from the local operator <S3> mhm-hm </S3> so the the example with the song network in helsinki the the company providing the TV channels is called <COMPANY NAME> and er they do not have any quality of service <S3> mhm-hm </S3> they just have their er (xx) which is (goes through) only one er switch which is theirs and then it enters into the to the network and their network is designed in such a way that the basically everything is works perfectly there's no packet loss <S3> okay </S3> and er and er the let's say end-to-end user experience is is good </S1>
<S3> is this necessary for a video on demand system or could you think that er you could be less stringent and strict and introduce quality of service and maybe introduce things like er error correction error concealment for packet loss and things like this </S3>
<S1> er i would say er see this as another advanced feature if we take the let's say the future scenarios where global company offering V-O-D could er er come to the service provider network <S3> mhm-hm </S3> (xx) to the user i see there are means for this quality of service and for measuring this quality of service <S3> mhm-hm </S3> of of several layers <S3> mhm </S3> so in a way the the provider of V-O-D the global provider would like to make sure that the quality of experience that the user is experiencing is er it's very good and it's er to its standard otherwise its image <S3> mhm </S3> would be <S3> [mhm] </S3> [damaged] the <S3> [mhm mhm] </S3> [user doesn't care] whose (spot) it is <S3> mhm no </S3> <S3> [@@] </S3> [they have no reputation of that kind] </S1>
<S3> it's the money-maker so you have to make sure it's er <S1> yeah </S1> good yeah , okay i would just to end my my rather lengthy general comments er you have er published your work er widely and er thoroughly i'd like you to er try to explain to to us er what your main contribution in the joint publications would be in general not not specifically for each but what would your contribution in the team be for such a joint publication and how did this change from er the first one until you became the first author </S3>
<S1> er , the , my contribution in a way remained constant over the over the duration of the of the work and er it was er mainly in the area of er system design so mainly the published components were let's say designed by me and er most of the implementation of the er systems and components designed by me was still (xx) a lot of done by me and there are papers where the er i am sometimes the first author and sometimes not the first author but the work is in a way continued and er the the reason for this are in a way political because er the work was done on two fronts for example i was in control but somebody else was in the transport part and being from two different organisation we decided okay one paper one organisation's questions other paper the other organisation's questions </S1>
<S3> your co-authors are they er fellow PhD students or or other colleagues </S3>
<S1> er , both cases </S1>
<S3> but they're from er this institution </S3>
<S1> er not er not in all publications </S1>
<S3> which other institutions are represented </S3>
<S1> er there is er nokia there is the (ATNT) er well different (xx) and er i think that (will be everything) </S1>
<S3> okay good </S3>
<S1> there is some person who (i do not think) (xx) but at the moment of the publication (xx) </S1>
<S3> okay , well i think that er ends my overall comments so we should er continue by by going through er chapter by chapter if you'd like to start </S3>
<S2> er you know this introduction chapter that is partly already covered by the general comments but i (would) want to want to elaborate on these once again on these reasons problems in current current systems that are listed on page 13 and er especially this lack of players or that would be one pro- would be one reason one problem it's not really a problem anymore i mean there are er a lot of of media players available on on both windows and linux platforms er it i understand that when this was au- authored it was a bit of a problem on on in the linux but if you are looking now (what really are the problems) that it's slowing down the the deployment er i would say that this mi- might not be especially important <S1> [er] </S1> [or] yeah </S2>
<S1> the this does not this doesn't (xx) <S2> yeah </S2> be the focus , but this work was er actually done was started in er 2000 in early 2000 when the the things were really looking very different than today <S2> mhm mhm </S2> we were searching for a nice player for the , capable of playing M-PEG programmes such (xx) not existent at that time </S1>
<S2> yes , and then of course there are the networks i also that er , they release kind of four four five megabyte a second it's not really available (on) fairly ma- er many places in finland yet if you are thinking now about for example finnish domestic conditions er but you have not really y- your okay your obviously it's it's behind there what you're talking about er like qua- quality of absence of quality of service but i think this networking thing is is someth- it's a that's a really big thing that you don't really have no have hi- high highband access to really and they're (xx) yet </S2>
<S1> er it can be that the the the bandwidth is that which be available from the end-user until the er service provider is er actually bigger than the current services but the current services are limited because of commercial reasons and because the aggregate bandwidth that the the service provider is capable of connecting to internet is limited <S2> mhm </S2> so if they would be offering four megabytes to everybody then this four megabytes times the number of users that that would be unsustainable for the </S1>
<S2> yeah yeah yeah , do you think that there there should be some some public er strategy there like in sweden they are really putting in a lot of governmental money in in in for offering the bo- broadband access <S1> er </S1> on a large scale </S2>
<S1> i think it it can be done both ways so what the <COMPANY NAME> is offering in er in helsinki area as far as i know it's at the level of three four megabytes </S1>
<S2> no er it's seven seven megabytes (xx) <S1> [yeah] </S1> [seven] megabyte yeah </S2>
<S1> and that the each er video each TV channel has okay probably less than seven megabytes <S2> yeah yeah [it's about five five] </S2> [and the] <S2> five [four to five] </S2> [yeah] and so in a way this doesn't have much to do with the with the data connection which is limited to one or two megabyte for example , so if the total line capacity is seven megabytes you can split it in let's say one or two megabytes for the for the data and then the rest for the video <S2> mhm </S2> and the video can be er completely separated <S2> mhm </S2> so it's the data which makes the application to internet (xx) from one side and what remains can be can be <S2> [mhm] </S2> [(put)] on video <S2> yeah </S2> because the video traffic does not in a way it's not created with a connection to the internet backbone [this is] </S1>
<S2> [yeah er i think] the big picture to 'cause when will the the typical what you call in finland <FOREIGN> pihtiputaan mummo </FOREIGN> in some village er the northeast of finland when she will she get the the really the needed access er will it will it will the market produce it or should the government go in and it's okay it's outside this it's it's interesting question i think and i think it should be in a way addressed also in in the when you're thinking in the introduction that that if you really want this to be a very widely available </S2>
<S1> yeah er i think it it depends on the the (district) infrastructure i would say there are good phone lines and there is er nice cable (xx) which no matter if it's analogue or digital i think there is no need for the government to put money in it <S2> mhm </S2> so if these two networks are good enough then i think they they are just good enough and there is no no <S2> mhm </S2> no need for <S2> mhm </S2> at least for the next (xx) <S2> mhm mhm </S2> but but at least we can can cope quite well </S1>
<S2> mhm mhm okay , so . yeah i think i don't have so much more to say say about the introduction because it has already already been touched on here and er so maybe we can go to the to the section two then about chapter two about integration of communication protocols and er . er i've one question there about the the the speed because 'cause when you are doing this integration there is some cost (xx) and er and er could you comment on this er . er for example there is er here discussion here on on page er 35 and 36 is that er that this encapsulation of data in the RTP real time protocol packet packets er will be er one that will slow down the the , the this and and er i would just could you just say how serious how critical is this this problem with the encapsulation of the RTP </S2>
<S1> er i think that encapsulation of RTP has to be implemented in a in a very careful way because if we are talking about the the server able to stream four hundred megabytes per second i think if you are doing er twice memory copy from one memory location to the other i think this is not possible anymore so in a way there there should be some some techniques of avoiding er let's say what i would call useless resources because at four hundred megabytes we are we are approaching the limits of the P-C-I bus maybe the memory speed (xx) too much data going let's say (xx) <S2> mhm </S2> so in i think in this way the the approach would be should be quite careful </S1>
<S2> yeah . er but you have not really clear i don't find any clear measurement of exact impact of of using the RT the RT the en- encapsulation here in the in the thesis </S2>
<S1> er . i don't think that would er , i don't see such a how such a measurement could be could be done in a let's say very consistent way so i would say that doing it try one of the <DISC CHANGE> bandwidth that can be achieved and then this this depends on so many factors that it it is actually very difficult to to measure and say that okay the driver (xx) of the network card is <S2> [mhm mhm] </S2> [the fault and not] the let's say the bandwidth (xx) <S2> mhm </S2> so if if there is a limitation in the speed it is quite difficult to find from where it comes <S2> mhm </S2> so if we er for example try to to simulate the the sending of the packets over network so that we are not sending over network we are just throwing packets away in the process we we do not use we do not send let's say four hundred megabytes per second over the P-C-I bus to the network card we are not using the network driver so it would be difficult to to split this group [(for one and the same) this is (xx)] </S1>
<S2> [yeah yeah] no it's not , method- <SIC> methodologic </SIC> problem there and er in that way you you're just giving here the total efficiency <S1> (xx) </S1> yeah </S2>
<S1> yeah there are probably expensive tools to to try to measure these but we didn't have anything available </S1>
<S2> yeah yeah okay , thank you i think that was er my questions er according to the second chapter two here do you have some </S2>
<S3> no </S3>
<S2> okay , so then we can go on with this chapter three transport of M-PEG-4 over I-P and er these are one of the many very heavy s- topics in this dissertation and er , and er i have only okay some very sm- detailed thing there are some , pages for example page 40 where the acronyms are er they are opened er on a later stage they are al- they are have been used already before and and not been op- opened in in in the first time they were used so that there are some some small things here but @that@ that is not of course any serious , internet task force and D- D-M-I-F are two examples here , and er <P:11> so er <P:08> yeah then i have one question on already on on page 70 about the M-PEG-4 and and and the the there's qu- you say here in the , bel- er here at 3.5 that it's similar to quicktime this M this the M-PEG-4 the MP MP-4 really and er , er could you com- comment on that a bit more quicktime it's on page 70 </S2>
<S1> yes well i'm not an an expert in the MP-4 file format but er the MP-4 file format is based on the same kind of er (xx) description descriptors as the quicktime <S2> mhm </S2> four <S2> mhm </S2> and er as far as i know er <COMPANY NAME> is very much involved in creating this MP-4 file format this is the main reason why it's it is very similar with quicktime </S1>
<S2> yeah but of course quicktime is a mu- much older started already in the mid middle of er in the in the start of the 90s so it's a much older format in that way then MP-4 is a more much more advanced of course it has been done the M-PE- M-PEG-4 standardisation <S1> (adaptation) </S1> mhm <P:16> yeah then on page 75 in the in the end of this er this er section you say that the er in the (xx) <READING ALOUD> the performance data show that while the RTSP D-M-I-F based server is able to accept around five connections per second </READING ALOUD> how , how do you from where have you have you got this five connections <S1> er </S1> do you see it's in one two three four five six seven eight eight page 75 on the eighth </S2>
<S1> er there is the table on page 73 which er shows how much time in seconds it is required for er each (command) <S2> mhm </S2> each (xx) so the the data described is based on that so it is probably er 200 milliseconds required from the the selection of the D-MIF plug-in until the the let's say (xx) interface user command because (the plug-in) <S2> mhm mhm </S2> which would be er (xx) <S2> mhm </S2> so it's based on real measurement <S2> based on real measurement yes </S2> (xx) </S1>
<S2> yes , er but the could you bring out now the the clea- clearer the difference er what is the cost that you we have to do to pay now er for doing this co- this this mapping compared to to don't to don't having this mapping er could i had a bit difficulties to find really the a clear a clear er description of what was really the the difference here </S2>
<S1> er in a way the , there are there are several choices that one can do (when getting) to the M-PEG-4 if you're not client of the server the the first choice would be to to use what the M-PEG-4 standard specifies which is er D-MIF and er default D-MIF signal and er this is something er quite complicated this D-MIF default signalling and i think it was implemented by two or three companies (i'm thinking) late 99 2000 around the year 2000 but these implementations (xx) was not that (product value) and then er some people decided that er or thought that using RTSP instead of er D-MIF default signalling for certain applications like V-O-D it er makes more sense <S2> mhm </S2> er other er ways to to transport M-PEG-4 content would be to completely skip D-MIF and use something else <S2> yeah </S2> for example RTSP bus <S2> mhm mhm </S2> er this er this approach of implementing er the (xx) D-MIF default signalling with er RTSP er has er several advantages so in a way it combines both (xx) so er the company envivio was mentioned earlier as far as i know their product they're not using D-MIF , so one of the application of this mapping of er that makes er let's say non- (lenient) s- er system several client working with er big base <S2> mhm </S2> would would actually make a link between two <S2> mhm </S2> so you could use the reference player MP-4 M-PEG-4 the I-M one which is D-MIF-based <S2> mhm </S2> the player's play from a envivio product </S1>
<S2> mhm , yeah but how what what's the (target) in time the cost in time for for doing this mapping of of of er compared to just using har- RTSP er i know we are you are using (xx) D-M-I-F using the (xx) and then converting that mapping that into RTSP and and i'm just look looking for the the time that the time that goes into this mapping er compared to don't doing the er don't having to do the mapping </S2>
<S1> er so er how much time the application is spending to [this] </S1>
<S2> [yeah] so (as) i mean count for example how how how many connection per second you would have without this mapping that that could be you have five se- you count five connections now er that we have s- saw here five connections per second but if you have (disk drive) without this mapping how many connection would you then </S2>
<S1> i would say that it would be maybe at most six </S1>
<S2> six <S1> yeah </S1> so yeah yeah so it's it's in that way it it doesn't slow it down very much </S2>
<S1> no i think er quite quite much of of this time goes into the opening of the MP-4 file <S2> mhm mhm </S2> and then the the library for reading MP-4 files that i was use- using (were) all these not (xx) optimised so i think that things can be optimised from many directions <S2> mhm </S2> and in the end er this er file connections per second are the what the server is would be in a way capable of accepting <S2> mhm </S2> so therefore if a server is only capable of streaming to streams to one hundred clients i would say that accepting five commands per second is a very decent amount <S2> mhm </S2> because in practice you probably don't have that much <S2> mhm </S2> then you can delay let's say the sixth can delay if there is a (xx) maybe one second which [(xx)] </S1>
<S2> [yeah] yeah yeah okay thank you i think that that are my my comments to this section chapter three and er then we can go on to chapter four the </S2>
<S3> <S2> [yeah] </S2> [i have] some chapter three <S2> okay fine @@ </S2> not finished yet @@ for your for chapter three <COUGH> if you ju- turn to to page er 39 three one at the er last sentence of the paragraph there are talks about having er elementary streams of varying quality so i'm assuming that elementary streams of varying quality you are referring to some form of scalability , so the questions are er how would you er (deploy) scalability in your system er is there a possibility for any feedback channel in your validation system to be able to signal scalability and how would you control these elementary streams of varying quality in a scalable way </S3>
<S1> er the the answer would be that the this the system and this mapping accepts all these things and that in a way that D-MIF does not provide the intelligence for managing scalability for control so this would be needed to be to be made at the application level so the application would need to to send the proper commands </S1>
<S3> and the application layer this er this discussion being the player <S1>  yes </S1> so this is an issue to deal with the player the player will simply receive a number of elementary streams and it has to be intelligent enough to choose </S3>
<S1> er that yes the the initialisation the the player will (see) the the description of the er M-PEG-4 and it will realise that there are several er (layers) several elementary streams [which (xx)] </S1>
<S3> [so so there's no way] there's no impact on the the transport at all the transport really doesn't care it's <S1> [yes] </S1> [unaware] of scalability <S1> yes </S1> okay </S3>
<S1> what can what will create problem would be maybe the the command channel the feedback channel <S3> mhm-hm </S3> because over the the D-MIF that will , specific command would go probably as a er user commands <S3>  mhm </S3> and they would be needed er these user commands will only play pause fast forward so there there would be needed some kind of let's say mapping <S3> mhm  </S3> additional mapping from er something to some (xx) and the RTSP could provide support for such commands set parameter gate parameter and option commands </S1>
<S3> do you [know if er] <S1> [(xx)] extension (xx) </S1> do you know if any of the commercial er systems that i'm thinking of windows or real or darwin do they are they able to use this RTSP information <S1> er </S1> control it do they use any control information </S3>
<S1> i do not know </S1>
<S3> oh . you turn to page er 42 . at the beginning of the of the paragraph er you specifically mention that the work presented <READING ALOUD> deals with remote retrieval control </READING ALOUD> <S1> yes </S1> er this is specifically for video on demand applications , but my question would be is this work really application specific or is it not more generic and can provide a framework for er media delivery in general not only video on demand </S3>
<S1> er it it has a let's say a degree of generality but it's not always applicable so er years ago when i <SIC> heared </SIC> that M-PEG-4 is coming i was dreaming to see the first game (on) M-PEG-4 <S3> mhm </S3> and let's say this architecture would not fit there so you because this (D-MIF) player is so it's so general i mean the (love) of many many commands like you have to move this monster that way <S3> mhm </S3> and er i mean there is now RTSP equivalent in there <S3> mhm </S3> because RTSP was designed with a specific purpose and the the the purpose of D-MIF is much more general than the purpose of RTSP </S1>
<S3> so you would simply have to modify D-MIF to be able to play games , er i'm sure you you're aware of the M-PEG-4 initiative called olga which is M-PEG-4 for MP-4 used for games you're not aware of this activity <S1>  no i'm not </S1> it was an exploration activity under M-PEG-4 for many years where they deployed the multi-user worlds for gaming fashion using M-PEG-4 er unfortunately er the reason why you haven't heard of it was that it was not a success so er i'm sure that er they would need someone to come up with er extensions to the D-MIF framework to be able to to do this so i i think you're quite right , er if you turn to page 43 . er you are at the top of the page of 43 you you deal with a crucial problem that's also been mentioned here which is the time delay if you wanted to use er M-PEG-4 as a video conferencing system or if you want to build a video conferencing system based on M-PEG-4 what is what is your main problem so can you discuss a little what the time delays of M-PEG-4 would do to a video conferencing setting </S3>
<S1> er i would say that the the time delays in M-PEG-4 are maybe only different from time delays in M-PEG-2 as regarding the coding and not the transmission i'm not aware of anything that would make a difference of transmission er to M-PEG-4 and would not make a difference to transmission (xx) in M-PEG-2 , it can be that the initial set-up of the link between er the participants in the conference would take longer in M-PEG-4 in the way that it's implemented in this mapping of of D-MIF and RTSP <S3> mhm </S3> as opposed to let's say I-E-T-F based protocol or others but but transport there and not (go there) (xx) <S3> [okay] </S3> [(xx)] (xx) </S1>
<S3> so the table on page 73 is the is the actual measured transport delays </S3>
<S1> er , yes but the , the so this is er yeah they are they are actual measured delays <S3> mhm </S3> but then this is er they are delays of the commands and it's maybe not the delay of the transport stream itself </S1>
<S3> so what would the delay of the transport stream be in in the order and what's the delay of the codec in the order </S3>
<S1> er i don't know but i would say that in video conferencing the the codec should be should be er a litt- i'd say litt- little bit different i'm not very familiar with M-PEG-4 but in M-PEG-2 er maybe it's not so good to use (displays) because of the extra delay in (xx) yeah that is <S3> mhm </S3> so to to have very very short delays between er frame capture on one side then frame display on the other side i think there are several things to (work at) , er the the measurements here were done in two computers in the same sub <S3> [mhm] </S3> [(xx)] so the delay between the rounded delay i think was below one second </S1>
<S3> okay what would be the permittable delay in a video on demand system from your choice till you actually get your presentation </S3>
<S1> i would say few seconds </S1>
<S3> you would allow a few seconds <S1> yeah </S1> mhm </S3>
<S1> i mean it's , talking about less than one second i don't think it makes much sense so optimising below one second , but then i think above three four five seconds delays are not (xx) </S1>
<S3> mhm and that would start to be annoying <S1> yeah </S1> so two three seconds you're probably probably right , okay that was chapter three thank you </S3>
<P:05>
<S2> (if that your) case so chapter four , i have only some more general questions and er <NAME S3> you have have a you go more into into detail in this (xx) is your speciality , but er i have one see one problem with the with the data security when you are er here using in in your player very heavily the plug-ins that you download from o- o- over the net and and now now there are in in for example in many companies er there are now er firewalls that makes it er not so easy to to download execute a plug-ins and and i understand because this virus this huge problem of virus has has emerged of the @you@ you (author) this but but would you like to comment on this the the about the da- data security programming and the the plug-ins </S2>
<S1> yes the downloading of plug-ins from internet in this architecture is a a general concept which was not implemented <S2> mhm </S2> and er , er in this case in the scope of this player the er the initial thought was that er plug-ins are downloaded from the service provider so that the plug-in would be downloaded from a trusted secure place <S2> mhm </S2> and er mhm internet today i would say that downloading some plug-in from some source may not sound as a very good idea but er i think that there are methods and encryption and er er ways of securing and verifying plug-ins today that should be quite okay you can download something from microsoft <S2> okay yeah yeah </S2> there are some questions there also but at least the user (xx) i think in a very high degree that that plug-in comes from microsoft and it was and it the user gets exactly what microsoft (is telling) </S1>
<S2> exactly yeah yeah , and er what is the now the status of of this player can it be can can you now (xx) is it from the net so you can download it or </S2>
<S1> er no the the research to this player was discontinued <S2> [mhm mhm] </S2> [(for example)] and the and the project was finished </S1>
<S2> that's er one business that you have a so much of this competing players <S1> [er er] </S1> [or or] or players that that that that do the same </S2>
<S1> yes er in a way the the concept of heavy (qua-) underlined architecture that would work both in PC world where you don't have M-PEG-2 (xx) the code and in set-top box in it it didn't catch up <S2>  mhm </S2> so you you have either er player for er for linux PC or players for set-top box <S2> mhm </S2> which is such as based on <S2> mhm </S2> on low (xx) (box demands) for that </S1>
<S2> yeah this general general framework is not really adaptable </S2>
<S1> er the general framework it is adaptable but it can be that the having a special player today for set-top box for example it it makes little bit more sense than to have the same player also for the for the PC <S2> mhm </S2> because in today's world the if you get something it is very very probable that that's M-PEG-2 and you can't have the M-PEG-2 decoder but maybe in the future when there will be many codecs used and er there are the set-top box will may not have hardware based decoders for everything it can be that the set-top box has also some let's say general linux player little bit modified and little bit (adaptive) <S2> [mhm mhm] </S2> [(for example)] the the first thing that i would say that would need to be modifying with the (xx) <S2> [mhm] </S2> [so] set-top box really want to (xx) control <S2> mhm </S2> and in normal use with basic commands and so this is the reason why the this architecture supports several input plug-ins and it's it's not does not stop to (xx) </S1>
<S2> yeah , thanks for that and no more questions </S2>
<S3> okay i'd like to make some some general comments first er this is an implementation of a player i realise and the implementation of a player was done er let's put it in in another era er there weren't as many players around so er i'd like to discuss at a more general level er interesting for me is how the er specifications were made and which choices were made in er designing the architecture so er can you tell me how how this work progressed through er specification to i- implementation what was important when you did this work </S3>
<S1> so er when the design was started it was , important that er it would work on a set-top box with a hardware M-PEG-2 decoder and then it was also important that it works on a PC with software decoder and then it was also important that er other codecs would be added in the future in a very easy manner so in in a sense that the the software can be upgraded more or less independent of hardware , it was also important that not everything is er loaded into memory at one time because of set-top box's memory is an important thing and that because of this er let's say (duality) each having different er input interfaces this was also important and er , er basically this was er this player was thought to in a way to have er multiple use so that , yeah we also wanted to make it a netscape plug-in </S1>
<S3> mhm-hm erm you sort of partially answered the question but the the application in focus was of course video on demand <S1> yes so [was] </S1> [er the] i'm assuming that these specifications are very video on demand specific <S1> yes </S1> especially with the with the set-top box requirements is it er or did it become more general can it be used for other purposes like a complete PC player </S3>
<S1> er no , i think that that the choices of today it's are much better than the (xx) on this player </S1>
<S3> mhm , in what way </S3>
<S1> er this er this player was er limited to to only M-PEG-2 <S3> mhm </S3> and today er the the player would have more much more plug-in , so the the main idea when we designed this was that we do not make the plug-ins by ourselves to the er you know to the best (xx) so that we take for example M-PEG-2 decoding library and we just put that the interface to it okay this is also the the concept of today so maybe quite many concepts of these can be found in players today </S1>
<S3> mhm , i'd like to show you a figure so i'll use this , over here <PREPARING OVERHEAD, P:36> i don't know if you can see this figure you have seen it before i that i know , er i'd like to bring this into to play as one way of designing players this is an overall by of designing a player important for this case is that the player will receive an I-P packet it doesn't matter how there'll be some network interface more importantly is there'll be metadata we've discussed metadata it's important there'll be a control panel the control panel might er well be a remote control and there will be possibilities for (metrics) because you will measure something , how does your specifications from previous fit in to the thought that you can (devise) out of this figure </S3>
<S1> er so the the input plug-in would be actually the the one controlling the er these basic play pause commands and er as the (metrics) it can be that er data is er is actually extracted at the different er level and er sending back these (metrics) to the server is not implemented but it is possible , the metadata when the all this let's say extra data about the content (xx) actually er there in this arc- architecture in the movie descriptors i mean media descriptors and er this can be available to to all the other plug-ins so it would be for the input plug-ins for rendering plug-ins to display such data to the user , so you in a way i don't want to open er a different er window if you're on this (xx) you just provide some layer (xx) on set-top box </S1>
<S3> so you would use the metadata to configure only at the user's side or do you see the metadata as playing a broader role in being included in the control signal and actually requesting actions from the server in a video on demand system </S3>
<S1> er , how the metadata can , request </S1>
<S3> metadata if you have a personalisation profiling you might want to feed metadata back because you want to update your profile which is maintained by your server's provider , which again might give you another experience the next time you go in it might give you a different flavour of advertising might be personalised advertising if you think this is part of a player or is it more part of the system level </S3>
<S1> i would say that it's more part of the system level er in this architecture if somebody would like to add such a thing i would see it more like a tool that it will be in a way separate from anything else then like part of the media (xx) system </S1>
<S3> so you'd like to keep it totally and , delivery-unaware if you like or or i'm sorry player-unaware </S3>
<S1> er erm to the maximal possible extent yes </S1>
<S3> okay i think that requires a thought with the er measures , well on page 76 . which is your first you of course mention the very important aspect which you also discussed a little bit already which is that of content types and formats and and schemes and er as far as i know there is an overwhelming confusion on content representation in the media industry today there seems to be battles the only ones that are comfortable are the broadcasters M-PEG-2 i'm sure that's a a compression standard that will serve the broadcasters for long time 'cause it's a good one , but for video on demand for other types of er players or or service providers er M-PEG-4 er is a potential candidate you mentioned yourself er windows video coding nine or windows media series nine and also H dot 2-6-4 , could you give your comments on on how these and other formats should live together and how do we tidy up the confusion in the industry how can we as experts tell the media industry which content representation to choose </S3>
<S1> i think this requires quite much thought <S3> mhm-hm </S3> and er i'd i think that the the first thing to to start from would be the requirements <S3>  mhm-hm </S3> so er H dot 2-6-4 and windows media nine are in terms of benefits quite the same and er they present the advantage of er M-PEG-2 that they require much less (xx) for the for the same (requirements) so in a way that would they would be the natural choice for er for video on demand today so if somebody would like to build a video on demand today i think that maybe very few people would choose M-PEG-2 , and er the the co-existence of M-PEG-2 with windows media nine and H dot 2-6-4 , i think we should take it as a given fact and try to design solutions that would , that would er (suppose) that all of all these three are in the same place and er they may come to the same set-top box you need a service operation </S1>
<S3> okay so if you were a set-top manufacturer and you were going to serve the video on demand or digital TV I-TV industry you would actually advice that you have to have all three <S1> er [(xx)] </S1> [or would] you make special purpose , single </S3>
<S1> in the beginning maybe not three but two of them which would be M-PEG-2 and (one) here but i would definitely try to go to this solution (xx) with set-top box (xx) </S1>
<S3> a duel , okay i will give you the question of the year which er which codec is best video coding nine or H dot 2-6-4 </S3>
<S1> er personally i don't think much difference between them i think that the this question does not not have a technical answer but who is going to predominate it (xx) it's (xx) er political business </S1>
<S3> i think i agree it's just a it's either political or er religious probably one of the two , if you look at page 81 at figure 4.2 i would like you to er to help me find the interface to the I-P packets in this figure </S3>
<S1> er transport media plug-in </S1>
<S3> so the transport media plug-in has the depacketiser this will accept an I-P packet <S1> yes </S1> and it will er unwrap it , it it does this in your implementation only allow for RTP or can you have er your own packetiser and depacketiser </S3>
<S1> er . it only allows for RTP (xx) so having a bigger number to transport these packets (into it play this packet) </S1>
<S3> what would you have to change in your player or in your system er to be able to introduce your own packetiser and depacketiser </S3>
<S1> er </S1>
<S3> you still put them into RTP packets they're valid RTP packets but you decide another and a different way of putting the bits into the packets , what has to change </S3>
<S1> er , things can be changed at different levels so you can change this media plug-in or you can change the demand (perspective) so the the media plug-in is taking the data out from RTP and it feeds it somewhere (xx) and er , er this the er let's say the M-PEG-2 transport stream is er multiplexed with something else where it's packetising something else then so that some other (player) this can be demultiplexed , so it can be that there is in the case of M-PEG-2 transport stream it has to be within the er transport plug-in and the demand (xx) plug-in </S1>
<S3> which is the transport media plug-in and the other one is </S3>
<S1> er it's the demand (xx) plug-in it's not on this figure <S2> mhm ah </S2> if that if it will be supposed that in the the multiplexing function is in the same transport media plug-in </S1>
<S3> so it's it's an integrated plug-in <S1> yeah </S1> okay , so if you wanted to introduce mechanisms like priorisation layered coding scalable coding the er this is the only plug-in that would need to change the transport media plug-in </S3>
<S1> er . for er scaling the scalable , i think that decoder plug-in should accept this or the demand (xx) plug-in should somehow know about it </S1>
<S3> if it's an M-PEG-2 decoder it it is capable of decoding <S1> yeah </S1> so you need some somewhere else you need to be able to unpack the er I-P packets <S1> yeah </S1> in the [correct way] </S3>
<S1> [and then] it will be the transport media plug-in </S1>
<S3> so it's the only thing that has to change <S1> yes </S1> mhm , in er page 87 , figure 4.5 , how is the rendering done </S3>
<S1> er because this is the , this is how it's (on) the the set-top box <S3> mhm-hm </S3> so the set-top box had integrated integrated the code er the multiplex (xx) so it was let's say outside totally outside this player <S3> [okay] </S3> [so once] once the the M-PEG-2 transports (xx) was (stated) to er certain device everything's done <S3> okay  </S3> by the set-top box <S3> okay </S3> </S1>
<S3> is there any way in er this media playback to include er event reporting and event reporting in the sense that you want to give the user feedback on the plug-ins in operation missing plug-ins how to get plug-ins et cetera er I E how do you make your configuration and your choices </S3>
<S1> er in this is not not done </S1>
<S3> so this is just a given , scenario </S3>
<S1> er basically er this scenario was envisioned but the design of this were to be somewhere in the distant future <S3> okay </S3> and er this could be possible to be done (within two years) </S1>
<S3> mhm could it it's not , as of such it's not possible on the set-top box that you have used </S3>
<S1> er i don't know if it can i think that a solution could be designed at some point because it was a linux based set-top box and basically what (xx) application is available to linux that could be (imported) there <S3> mhm-hm </S3> so the solution i think it is possible <S3> mhm-hm </S3> but nothing was done to this direction , practical </S1>
<S3> okay thank you very much </S3>
<S2> yeah i have one one last question about on this is er do you see foresee some possibilities that er you could have have the player as a applet really (xx) level er it mean in java because i asked before because we have one one er research activity at in our institute er this motion vector coding that is has been developed by by VTT over a a long range of years and there are the (xx) symmetrical scheme where where where the decoding is and rendering is is is just performed by w- a- a single applet that is downloaded at the execution time and that that m- m- that er makes it very very easy to use and er but the would you just like to comment on on this </S2>
<S1> (borderline) in a way far from being an expert in java but in java there is such a thing as er java media framework which er i guess has a decoding capabilities and it specifically deals with er media decoding and transport and (xx) , to implement er with decoder in pure java i would say would be quite slow i do not know how slow it would be but it would be slow </S1>
<S2> yeah we have er total different scheme it's er it's er very much based on table (xx) and er but that can be done in real time and it's er it's [er] </S2>
<S1> [so] (thing) could be (valuable) very much on the on compiler so as a (questionable) example er at some point in i think in 98 we were having an M-PEG-2 decoder for some and with er compiled with all the optimalisation it was running in real time with no optimalisation it was decoding one frame in two seconds <S2> mhm-hm </S2> (xx) </S1>
<S2> okay er but nowadays you have of course on a on a PC lot of resources <S1> yeah </S1> yeah okay thank you <P:05> would you like to start </S2>
<S3> okay we'll go to er chapter five for the the architecture of a video on demand server er i'd like to ask first if if er in general terms are there any specific er video on demand-like servers available commercially on the marketplace today </S3>
<S1> if there are video on demand servers available </S1>
<S3> not video on demand servers but ser- media servers that could be used for video on demand </S3>
<S1> i think that a number of companies provide video on demand service commercial </S1>
<S3> and who who </S3>
<S1> er kasenna <COMPANY NAME> </S1>
<S3> mhm do these servers differ from your choice or are they along the same line </S3>
<S1> er i don't know their terminals is (xx) so i don't know what methods they use <S3> mhm </S3> how how in what way they are optimising the er (disk) (xx) <S3> mhm </S3> are they (xx) </S1>
<S3> er in your thesis you talk about speeds of er it's up to five hundred megabytes per second er where is this number today do you know </S3>
<S1> i think there are few of these servers today that are capable of streaming eight hundred megabytes from single <S3> mhm </S3> from single hardware </S1>
<S3> mhm , yes i chek- right we , we have one we have a er server from (A-V-I-D) if you know the company it's er capable in the brochure of @er@ streaming one gigabyte per second and we have managed to get a measure about eight hundred er but even (A-V-I-D) are going away from that solution because they would rather have (clouds) of smaller servers and put them together in a group instead of have these single powerful servers , in your work of designing er this architecture were you influenced at all about er from er were you influenced er on work in media management and and work flow management from the er broadcast and or film industry </S3>
<S1> no </S1>
<S3> so you haven't looked at that at all 'cause that's very similar type of of work from a video on demand </S3>
<S1> er yes <S3> yeah </S3> i today i do know but when this work was started that was <S3> [mhm] </S3> [completely] different thing <S3> okay </S3> (another thing) <S3> okay </S3> er this started from er in a way from the same project as the player with the set-top box because on er on the set-top box we er we requested to achieve (xx) that the player could work with video server <S3> mhm-hm </S3> <S2> mhm </S2> and (xx) video server at that time was capable of handling about one hundred streams M-PEG-2 point four megabytes <S3> mhm </S3> and the cost was around in one hundred thousand <SS> mhm </SS> and we were (xx) we can <S3> [@@] </S3> [do the same] much [less] <S2> [mhm] mhm </S2> so this is how this server came out </S1>
<S3> mhm , on page 94 . er you op- you have a a er it's the last s- paragraph you have er some comments about the er seek time and the estimates that you use er how much do these seek times actually differ from manufacturer to manufacturer </S3>
<S1> i think that the seek time is between eight to 13 seconds today in let's say low and hard-disk </S1>
<S3> and this is fairly stable there's no manufacturer that really goes lower or a lot higher </S3>
<S1> no but the technology is the same <S3> okay </S3> er there are in a way two two aspects of this there are the so-called professional hard drives which are er which are i think about five times more expensive <S3> mhm </S3> than this normal and there the the figures go lower so in what we have used was this 7200 rotation per minute hard disks and there are 10,000 rotation per minute and 15,000 rotation per minute but the seek time there can drop up to three point something <S3> mhm </S3> so er given this data there are some formulas in these chapters and these can't be computed <SS> mhm-hm </SS> erm what i think it changed since the this work is mainly the total (xx) that er hard disk is capable of </S1>
<S3> mhm-hm , this er i think leads into my next question because on on page 96 and 97 you have examples with actual numbers are these numbers realistic er today or were they realistic yesterday </S3>
<S1> i think these numbers are realistic today </S1>
<S3> today okay then i'd like to say that to gain high , the what what is the gain of high efficiency and and er what is the cost function of such a gain for instance you say er that one of the examples gives 256 megabytes or the other two gigabytes of memory for buffering does this really matter , in the overall system </S3>
<S1> er i think it matters in a way and er there are er several , in a way several er drawbacks what you see more and more so er these computations were made using , er with M-PEG-2 in mind <S3> mhm </S3> M-PEG-2 has four megabytes but if we are going lower to M-PEG-4 with one megabyte <S3>  mhm </S3> then er er there because there are more streams and to achieve the same overall speed of let's say 400 megabytes per second we would need to buffer much more <S3> mhm </S3> so it might not be an issue that okay maybe there is 256 per one gigabyte (xx) but it might be an issue there if is say one gigabyte or five gigabyte </S1>
<S3> mhm-hm mhm-hm yeah , this er performance model that you have er made is this a generic tool or is it specialised for your validation video on demand system </S3>
<S1> er , it was certainly designed to (evolve) this video on demand but it to a certain degree this can be regarded as a general tool <S3> mhm </S3> so er in in many in different aspects er , in most of the applications data is not (frequent) (xx) so you have a number of seeks per second <S3> mhm </S3> so you can use this model in a way to to measure er the performance of the (video) application <S3> mhm </S3> based on the number of seeks per second that that application requires from a hard-disk system </S1>
<S3> mhm-hm , er when you say that if you turn to 101 and page 5.5 , can you just help me explain the relationship between continuous read which is much higher and and the disk pump , why is the continuous read so much higher have i </S3>
<S1> er because there is no not seek involved in the continuous read </S1>
<S3> so what would be required to get the the disk pump closer to the contin- i i'm assuming that continuous read is is the er limit <S1> yeah </S1> so what would be required to to get it even higher </S3>
<S1> er lex less numbers of seeks per second </S1>
<S3> that's the only way of getting <S1> yes </S1> it higher </S3>
<S1> because with each seek you actually lose a fraction of the time (xx) <S3> okay </S3> and the more of that time you lose the less time for the actual reading </S1>
<S3> and how do you reduce the number of seeks </S3>
<S1> er by er buffering so to , if you have let's say one hundred streams with four megabytes and er you need once every second basically you have one hundred seeks because you assume that each file would be different one hundred would be quite high number <S3> mhm </S3> but if you target to have ten seeks per second then each stream needs to read every ten seconds <S3> mhm </S3> for them to read every ten seconds and then you have to read for the whole this period of ten seconds so there you need buffering <S3> mhm-hm </S3> if you want five seeks per second then you have to read every every (xx) reads every ten seconds <S3> right right </S3> mhm and then you need the the buffering <S3> [(xx) okay] <S2> [mhm mhm] </S2> </S3> </S1>
<S3> okay erm <PREPARING OVERHEAD, P:16> i hope you can see this er this figure , on page 102 you you work (xx) or talk a little bit about packetising , as far as i can see you have er (somewhere) you have your content up here and you have your media er encoder and this is an M-PEG-2 encoder that's fine so the M-PEG-2 encoder provides you with a (xx) screen er down here you get er I-P packets and you put them into er RTP or (UVB) in general and you (use) (xx) the RTB , this which i've labelled middleware with a question mark is your actual packetising do you think it's possible to pack the bits into the I-P packets smarter than is done in standard systems to be able to increase the quality of servers or quality of experience </S3>
<S1> i think that . er one tested way to er properly packetising this to to make it fast enough so there you can make a difference in a way <S3>  mhm-hm </S3> but it can be that there are schemes of and ways of packetising er data in such a way that er the impact of one possible packet loss is diminished </S1>
<S3> 'cause in your validation system this packetising is done unaware of the codec right so it's done in the standard way </S3>
<S1> it's assumed that the codec is M-PEG-2 transport stream <S3> okay </S3> so this is the only assumption which is made </S1>
<S3> so if you think again that it's possible to optimise the packetiser given the information that you know these are bits and you know that this is a video stream you know that this is a video on demand system </S3>
<S1> er yes i would say that it's quite possible especially because the server has so large buffers so it could have let's say five megabytes worth of data i think this leaves quite nice room for several packetisation schemes </S1>
<S3> what's the size of the packets in your system </S3>
<S1> er the size of the (UGB) packet tries to be less than the maximum transfer unit per different which is one thousand five hundred packets so this would mean i think five transport stream packets in one (UGB) plus (address) </S1>
<P:10>
<S3> okay thank you that's er , please </S3>
<S2> yeah i have er on page 89 i have a short question about the er when you're talking about cost requirement in the second paragraph there er and then on the other ha- side you're talking about deployment requirements and you say that they are er they are i can't really see the the the difference here because you are talking about cost effectiveness in in both cases could you just elaborate on that </S2>
<S1> yes so er i think that er scalability is different from cost in the way that er single server can be very cost effective but it there is no way of adding the second server to the system then it's not much use for [(xx)] <S2> [mhm mhm] yeah </S2> also cost effective <S2> yeah </S2> and then er the reliability is again somewhat similar factor that the server would be only unable to work let's say nine per cent of the time it doesn't matter much if it is very cost effective </S1>
<S2> mhm yeah yeah , yeah that's clear and er i also have one remark that was already dealt with about the the buffer sizes because the memory price is so much lo- lower now i think than that it maybe er in if you can save say one gigabyte or two gigabyte the memory that doesn't so much make a sense anymore but of course you had had the argument there if if if you have very much many streams and there'll be you have to have have large buffers and then it might starts to start to have have a certain significance </S2>
<S1> in the in the server implementation in the the measurements er it it is talked about this (xx) plan and i think that that one is reducing memory to half basically <S2> mhm </S2> and i think that having ten (captors) per programme would be using only half the space <S2> [yeah] </S2> [(because) (xx)] influence it </S1>
<S2> that's clear yeah , and on on the page 101 the fig- figure 5.5 i have a question about the the theore- the theoretical disk pump the this how how this curve is calculated you you have have those squares </S2>
<S1> er so the theoretical disk pump is er half way bet- it's 85 per cent of continuous (read) <S2> mhm </S2> because in theory we established that one complete seek operation takes 15 milliseconds we have er ten seeks per seconds so this means that er 15 times ten which is 150 milliseconds is lost during the seek operation <S2> mhm </S2> so basically this is just 85 per cent of the continuous [read] <S2> [okay] okay </S2> because this 85 is what's remaining if we subtract the <S2> yeah [okay] </S2> [(150)] milliseconds from one second </S1>
<S2> well that makes sense yes , and er then there are here say on page 102 that er er <READING ALOUD> the conclusion of the test  </READING ALOUD> in the first paragraph <READING ALOUD> of the disk pump in the V-O-D was capable to serve all 100 clients and it was displaying the same behaviour as the testing programme </READING ALOUD> i i would just i would have maybe liked to have some more expressive data on on the test er it it's only here a phrase er but er some some tables or some numbers </S2>
<S1> er in yeah in the in the real , in the real server the amount of measurement that can be done it's in a way limited so you have let's say if we have if we mhm establish that we have ten seeks per second then the disks er in a way each ten second the the server must make a complete cycle of seeks and read so in the end we can say that okay the server was able to complete its cycle every time <S2> mhm </S2> but then it er this does not let much space for the the real measurement <S2> mhm </S2> and er the the way the tests were designed was er to in a way to to get give us best (xx) <S2> mhm </S2> one for example was that er we tried to get in the the files system because it can introduce some (overhead) dependent on file system and and er some er some files can er can be fragmented <S2> [mhm] </S2> [so] if the file is fragmented then you actually get one extra seek [(xx)] </S1>
<S2> [mhm mhm] yeah yeah yeah , so there is a method methodological problems again here to make ex- make exact measurements you know </S2>
<S1> so in a server you in the real server you need to have some provision for er for the file system fragmentation and also for the the user seeking in different part of the (xx) <S2> mhm </S2> that that's something to be eliminated <S2> mhm </S2> that can be completely [(xx)] </S1>
<S2> [yeah yeah] so that that's why you have a more of a overall statement here , yeah i think that was my comment in about this </S2>
<P:13>
<S3> so for chapter six er for the management platform and multimedia systems what is new or left in your work compared to commercial systems and management up-to-day because i i'm assuming that this is a field where er the progress has been enormous in the recent years of how you actually manage and maintain commercial systems </S3>
<S1> i think that commercial systems today have probably much more features than this management system here and er they are probably much more complex <S3> [mhm] </S3> [and] much more complicated than <S3> [mhm mhm] </S3> [(this here)] i think that the price for such (an expensive) start from about one hundred thousand <S3> mhm </S3> <S2> [mhm mhm mhm] </S2> [(xx)] so er let's say maybe 99 per cent of this management system was implemented by one server so this is in a way quite nice and (light) system which can be let's say deployed in in places where the complexity of of big systems is not needed for example small (xx) </S1>
<S3> mhm-hm </S3>
<S2> mhm mhm mhm </S2>
<S3> is it er worth commercialising </S3>
<S1> er theoretically yes </S1>
<S3> in practice @@ </S3>
<S1> in practice er there are er many questions to be to be solved </S1>
<S3> so you have there has been no efforts of commercialising it after it was made </S3>
<S1> sorry </S1>
<S3> er it ha- it hasn't been commercialised yet </S3>
<S1> er no it has not been commercialised yet <S3> okay </S3> er in in commercialising this er there is a problem of er bundling it into another </S1>
<S3> mhm , exactly , okay </S3>
<S2> yeah i have just thought about here the the way of talking about what was the the the management side that is this er this er function for like like here on page one hundred 100- 111 six six dot two figure that er i have thought that i mean the the the view of the user normal user is fairly different from the the one of the administrator so do you think that there are common elements in (xx) kind this kind of a unified </S2>
<S1> yes i think that the the metadata for example is the same <S2> mhm </S2> the goals i mean the title of the movie is the same <S2> [mhm] </S2> [for] administrator and <S2> yeah </S2> and that the basically if everything stays in the same database and the same application server is generating the web pages for the administrator and for the user <S2> mhm </S2> i think that in a way much of the publication server code can be used for both </S1>
<S2> yeah yeah , (xx) user interface as you'd er pointed out in the beginning using interface for the user should be very very easy and and er tied to the to the to the to the hand hand hand held remote control and and that's of course the the the interaction mode is different than than for administrator <S1> yeah </S1> yeah </S2>
<S1> er also i was supposing that basically the the web pages are HTML for both the administrator and the user <S2> mhm </S2> so i think that i mean the the system is designed that the administrator is er (xx) modifications to the web browser , in the same sense on the client side er the client can get an HTML page over HTTP <S2> mhm </S2> but the system can be easily modified that the client actually gets an XML <S2> mhm mhm </S2> information <S2> mhm </S2> and there is some easier or some little bit different user interface <S2> mhm </S2> too <S2> yeah yeah </S2> but in in the in the end the there is the same query that er is searching the database <S2> yeah [yeah] </S2> [for] movie [(xx)] </S1>
<S2> [yeah] yeah , okay that was my questions on this and then we have have a summary in the . i don't think i have have very much in the summary , do you have </S2>
<S3> yes <S2> yeah </S2> first before i do the summary i'd like to er ask two very general questions er first of all what's do you feel are the strongest points of this thesis </S3>
<S1> i think that the the strongest points rely on er on combining the theoretical research and design with the implementation where a very nice er feedback can be can be obtained from the from the actual implementation </S1>
<S3> mhm mhm , where and what as of today and as of your work is the weakest link in a video on demand system where do you need to put the most work to be able to deploy fully video on demand systems </S3>
<S1> i think that , the , in my opinion the the set-top box of the future will dictate much of the this kind of network <S3>  mhm </S3> (xx) in video on demand , the capabilities of set-top box and the the way they are introduced to to people's houses is going to dictate much </S1>
<S3> mhm </S3>
<P:13>
<S3> okay so mr <NAME S1> and the audience you have given us er a great presentation putting video on demand systems in relation to digital TV and interactive TV and er we were able to use our reading of the thesis and also your presentation to er form our questions and er i must say that er the presentation and the discussion has been very good you have given some thorough insight and er valid answers to our questions both within the scope of the thesis and also way beyond the scope of the thesis 'cause we have touched upon er materials that that go far beyond the work that you have done and also that is scope and and interest of the general public also of today because video on demand systems and digital TV interactive TV really is something that goes into all our lives because that's the entertainment form that we really use the most and i think it will continue to be the entertainment form that most families and households will use , so to to sum up you have done an enormous amount of work i think it's been very valuable it's been very complicated to understand all the interfaces that you have had to go through and how you have solved these crucial problems that you identified in the video on demand system that must have been an enormous amount of programming after the complicated task of doing your designs and er making a validation system i think it's very valuable and as you said in your own words it's very good to get the feedback from implementing a validation system on your theoretical work , i also think that you have er proven to the research community that this er work is valuable by your publications they have been widely spread and also give insight into these solutions , so the results presented contribute er towards the solutions and identified problems for video on demand systems er i'm sure that er they have been recognised internationally and i'm sure that your institution has had great joy in supervising and following this work throughout your thesis so in conclusion we would like to recommend that your thesis be approved thank you very much </S3>
<S1> <READING ALOUD> if anyone present still has some remarks to make or questions to ask concerning my dissertation i respectfully request that they ask the <FOREIGN> kustos </FOREIGN> for the floor </READING ALOUD> </S1>
<P:15>
<S4> as the <FOREIGN> kustos </FOREIGN> of this dissertation i declare the discussion concluded thank you </S4>
