Internet Engineering Task Force 93, 23 Jul 2015

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF93-TSVAREA-20150723-1740

Description

TSVAREA meeting session at IETF93
2015/07/23 1740

A

B

Or change request to the agenda as its shown.

C

B

So, apparently, not okay, so after bashing the agenda them spends on. I am just short. We talk about the transport area, directors and this non comm cycle. After this, we have a presentation by email about the impact media transport protocol with afterwards a small discussion about.

B

We are we're actually trying to work on the evolution of transport, and then we have, towards the end, enables for transport layer, protocol, evolution of sylvester and hopefully enough time towards the end, to have a bit of open mic discussion about evolution of transfer protocols in the chance for the area.

B

So before going ahead, we have so many people probably have seen is that I am NOT standing again for a third term, and my term ends in April next year at the ITF meeting in kronos itís, and so this is like a car for winning the nominees for the time and for the air position transferred it. If you'd like to see more information, something like desired expertise or in other words a job description, you can go to the nam com, page and I.

B

Think on this page you will see maybe in a month or so, I'm not sure the official job description. But I think if you look back to the mailing list, did we post it there while rushing? Yes, they are beautiful people. Sorry, most of the draft plans for comments right. You can see it on the chance for the area. Maybe this also, let's we're transparent yeah. So in case you don't have enough nominees. um Read this and look at the picture. That's nice guy! I called golem and he's waiting outside.

B

I think the banquet hall on the lower level- and you see he has already lel, yellow dot, means he's part of the isg.

B

D

Good choices, yeah.

B

And I think he has to a simple rule: everything it doesn't like he just marches away so and over it and he's big and thanks on Spenser for making the slide in the picture. I would never seen this guy's to blind okay, then some heads up or clarifications you like to say this once.

D

E

B

That's the signpost: okay, okay, then we go sorry for the confusion and then we can hand over to email. Maybe you want to use your own laptop present presentation. I should have to write clock for you. Yeah use this one yeah we have been nice and no queues offers maybe but her blood, so that should work.

B

But you have to go and then but you have to stand in the yellow box.

B

There now we have to this morning for them.

B

C

C

Okay, yeah, thank you. So this is a short presentation about the MPEG media transport protocol, but I think what my purpose of being here is to initiate this discussion on the future of media distribution and get the IETF to be involved in this and get their valuable feedback. So next slide please! So we nothing thank you.

C

So we are looking at an evolved landscape of media distribution, our starting point and the MPEG was to address like broadcast use cases, but those have evolved so far that it's it's not the traditional broadcast that we are used to. We, the industry, is also moving away from the mpeg-2 transport system, as does the transport layer for media delivery. We are looking at an all IP network, we're looking at a variety or very rich media content that is being presented, so TV doesn't look the same way as accused to look.

C

We are looking at like a very high definition: content with you, HD so with high dynamic range, very high rates or very high birth rates, but we're also looking at the very rich scenes where you have multiple video streams being displayed.

C

We have companion devices that display parts of the service, additional content like new stickers and some additional like descriptive contact and you're looking at targeted advertisements, so it's quite different to what we were used to and those were actually the drivers behind like working MPEG to phase out and back to TS and replace it by a more modern like transport layer, and this was done in MPEG as part of an th.

C

It's published standard 2308 part one, and it provides like the system layer for run for the MPEG media, so mmt defines a transport protocol and that's what I'd be talking about here. It's it has also like a signaling layer, fec framework, but also like looks into presentation layer as well so looking at html5 and enabling seen updates in broadcast environments, for instance. So next slide piece. So the requirements we had so we need. It is like something that is generic, so it should be like similar to mpeg-2 TS.

C

So it should work with any media type. Especially I mean the MPEG media types, but it's not limited to that and it should support both real-time. More real by media distribution should support multiplexing, so easy multiplexing is the requirement there. So mpeg-2 TS was quite hard to read multiple X. So if you added like new components to the stream, if you want to do like targeted ads so replacing portions of the content with content that is coming of a broadband channel, so it needs to be like easy to.

C

Multiplex has to be self-contained, so like can back to TS itself descriptive. So once you get access to the stream, you'll find they're signaling information that describe all the components that exist there, maybe also like neighbor services and so on. So that's why we had to define this signaling layer as part of the of this of mmt, so focus of the transport layer should be on the delivery.

C

So we decouple presentation, layer from transport, layer and support for multiple sources or said content might be coming from different sources and needs to be be multiplexed of the sea beside and resynchronize, also for playback, and also for playback on different devices, for instance in the home environment. So next, so why we did not opt for our TV at that time, so a lack of multiplexing. So we believe that multiplexing in RTP is quite limited, so I thought I think only support for RTP rtcp multiplexing, but not multiple service components, server, maintenance.

C

We wanted like something that supports generic format. Instead of having to develop like special payload format and having to update your streaming service each time you have to add support for new codex.

C

We didn't appreciate the coupling of presentation and delivery for different reasons. So having like a PP time, stamps or presentation, time stamps and the protocol is not flexible enough.

C

So when we think about redistribution of content when we think about consuming content coming from different sources and when we think also about like rich presentations, where you have like targeted advertisements or contemporary pieces over content in the presentation based on some customized or some user profile- and things like that, so we thought it's very important to separate and the couple like the presentation layer from the from the delivery layer.

C

Also the limited support for non real-time. So I said the scenes are quite rich, so you might want to present images or companion content at the same time as you presenting the audio video and that might also go to the other devices in the end the year and the whole network, for instance. So we had of this transport protocol had to provide support for that next.

C

So this is like an architecture that we use, so we have like content providers which could be like multiple sources. We have sending entity. So that's the server and receiving MTP and I said there is some signaling in both directions. So receiver.

F

C

Report back about reception conditions to the server and that can be controlled like how much reporting is required by the server and the media delivery is from the server to the receivers next.

C

So the protocol itself is like a basic protocol, but you are looking at so puts we thought I mean we like the idea of lct, for instance, as a building block. It lacked a few features and had like some some features that we didn't see you need for. So the starting point was something like like a CT, but of course our target was to optimize for media streaming, so very or low delay support for low delay.

C

The protocol itself is quite simple, so it has like that type field which would identify the type of the payload and we define a limited set of pillows so for for payload types, then an identifier for sub clothes, so that's feature that we don't actually have a net CT, but we need here. So we have to support like multiple sub clothes or sub streams that correspond, for instance, to the different service components or your video up, titling or static content, and so on.

B

C

Have a time stamp, which is like a delivery time stamp? It's something that, for instance, a city ever.

F

C

An optional header field RTP has a presentation time stamp there, but we opted for a delivery time step, so that would enable like pectins, that's measuring, like delay, jitter and so on. We have a sequence counter: a packet sequence number, which is like scoped by the packet ID. So if there are losses in a particular service component, those can be detected to the packet sequence number and you have an optional packet counter, which is like a global packet, no number for the whole lemon TP flow and it's optional because it makes likely multiplexing quite difficult.

C

So if you wanted to multiplex without the packet counter, you don't need to rewrite any of the packet headers. If you had the packet counter, then you need to really sequence: Thea, the the flow we have Heather extensions and an optional our FEC payloads identified at the end of the packet. So that's pretty much.

C

What the packet header contains next slide is about the procedures. So we define like a session which consists of exactly one mm PP blow its memory. Mpp flow is identified by the destination IP address and port number, so it can support like so specific and multicast and any source multicast and particular. We are interested in the case where receiver gets contact from different sources at the same time. So it consists named. Mvp flow consists of multiple subfloors.

C

Each has its own packed ID, Adam identifier and each sub flow would carry like service components which could be audio. Video subtitles in genetic data like regular files.

F

C

So on, and it supports FEC at the flow or sub flow levels, so you could decide to protect the whole flow or only like the most important parts of it like audio, for instance, or you could have like multiple FEC protection, where you use different code rates for the different service components are next payload formats. I think it's not relevant for the draft.

C

What we have here, but I think it's like one of the core idea there is to support a very limited set of three loads, which are generic enough to address all needs that we foresee. So we start with the will support of a container formats or payload format for the mp4 files. So the idea is that whatever you can put in an mp4 file, you captain stream they using an ftp. We support the payload format for genetic data, so that's similar to what you could do, for instance, with the ALC and flute and F cast.

C

So a similarity to F cast is that those files can be self-contained, so they have their own metadata as part of the transport object and we support like signaling. We have a payload format for da signal information which could be XML or binary representations of signaling messages.

C

The last payload format is for repaired data, so we decouple like repair data from the actual sauce packets. By having them on. There is a separate on a separate payload format and fragmentation aggregation are performed at the payload format, so we don't support that at the protocol itself.

C

C

I want to waste like much time on the discussion of the payload formats, but the idea that, if you support like streaming of a particular container, I've seen, for instance, see and the IETF talk about defining matroska as a container format and IETF, so that could be like an additional payload format that we can. We can adopt there and whatever you can store in matroska with an optimized payload format. For streaming. Like my poster files, we could address the needs of certain applications.

C

So far, the only format you are supporting and when it comes to media stream is the mp4 file. Formal I said that touch the decision Don in MPEG, and the priority was given to the MPEG media formats.

C

Next talks about generic fun delivery, which is like a more data, lost the application to send any type of file self-contained. So it provides in the payload data to carry the meter date of the file and the delivery is like based on the offset of the payload into that file. So we have a transport object, identifier, similar to LCP, for instance, and an offset which places.

C

The payload inside that transport object. The difference here is we're not using like I am identifier or a sequence number using the offset. So that's that's more flexible when it comes to locating or handling the packetization of the occupied.

C

Next is about singing messages, so we have like a generic format for signaling messages, both in XML and binary format, representations and the message payload. Can it's flexible so become a sign like a message ID for that message and define like the message contacts?

C

We can't defy like a new message content there. So it's in a way very similar or bottles like concepts from mpeg-2, TS signaling and the last one is the FEC framework.

C

Oh, maybe I don't have the FEC framework, so we have a built-in buffer model, so a hypothetical receiver buffer model which we can use to like Pepsi beside at the center side, to make sure that, given like some restrictions on the receiver side, restrictions on buffer capacity and on end-to-end delay, but you can ensure like perfect reception at the receiver, so no buffer, overflows or under flows and those the end-to-end delay, but he's singling there is meant to cover, like jitter digital ink instance.

C

If you follow like different path, transmission of a different path, it covers also like FEC decoding reorder england's form. So that's pretty much all I had to say I think to do this. We are here to talk about congestion, sore atf, I. Think as like more expertise on this topic. So so father said we designed for provision networks or broadcast networks, but would like to address, like hybrid youth, k, use pieces which are like a very crucial for next generation media.

C

So we assume that people would be consuming content both of our broadcast and broadband simultaneously, and we need like congestion control for at least for the internet, part of it.

C

So we think that maybe you can reuse like existing components here, for instance, broadcast like web rcmp fncc thinks that we're in our mt, but also over broadband sore, the regular internet, maybe the workloads that is done, RM card, that's one mic.

C

Relevant to enable like an MTP streaming over the internet, but also I mean for next slide piece. So we are also looking at ways of developing like this all IP media delivery, so that other like a CEOs like MPEG, ATSC, 3gpp, DVD and so on, find something that he's suitable for them. So we are I said we are looking also at like the developments and the media streaming of the internet, so well dash, for instance, or streaming over.

C

Http is playing a role, but also looking at improving that so by looking at a different like passport protocols that might reduce the and it lay the steaming delay the Union time and choice which time and so on, and for that to believe that, having like a development of a suitable protocol here might be quite beneficial for rod, industry and so I would like to hear. I mean the opinions of people here and the ITF.

C

We thought that this is very similar to what rmt has done, but also enhances that with like new use cases, namely steaming use cases, hybrid delivery and so on. So this could be maybe potentially done in by reviving are empty and working there. On on this we could start maybe new, ad hoc group or buff, maybe, and look at the use cases that we are considering and clear.

C

The last option could be like, but we first, you were like an informational RFC track, so ripped open will also open to giving control to the IETF about this protocol or the development of this protocol and sync up basically with what the ATF does in our future releases of the mmt peoples of colon MPEG. So that's all I had to say about this and ok.

F

Thank you. If, bit of time for questions from the community members, um one of the things I didn't hear, you say it was anything about what you envision sort of the outer layer being or you imagine yourself over UDP or um yes, are directly.

C

F

C

So the initial at least the initial consideration is that UDP would be the suitable protocol, but we've seen also like thoughts and contributions. The words like the doing mm TP / web sockets or delivering like an MPP content of WebSockets for better integration, for instance with CDN. So TCP could also be like another alternative yeah. But UDP is the mean so.

F

One of the problems you have here is embedded in this or some sort of unsolved problems that have been hard problems that have been unsolved for a long time and these problems are replicated, so one of them is sort of the version of how do you deal with specifying congestion control for something looks, looks like a tunnel when you don't know anything about what you're carrying and there's sort of a recursive congestion control problem there hug sorry I. It sought to understand you. How do we.

C

Deal with, what's there.

F

C

F

Protocol that looks like a tunnel which can't own responsibility for when data arrives, but is sort of responsible for congestion control, in some sense to delayers other people who are sharing its channel.

F

This is a hard problem that has not been solved. Part of it is I think because they're actually asking the wrong question and I don't know what the right question is. I, dont still don't know what the answer is and I don't think this work is the right place to solve that question, but this work would be a lot easier. If that question was answered, yeah.

C

I'm not sure I agree there, because we I believe we control what is delivered there through defining a very limited set of payload types. So so far we have only for payload pipes, which we support. So we know quite well what we are delivering so in a sense, I slightly disagree, so.

B

We just closed the lines of people at them. I can just work through, but nobody else, but stars malin hit just in time.

G

Mei kuen I don't want to be worth picking, but you said like a natok group, so this seems like a pretty big effort. Don't think, there's anything we can do a talk here, because we do this for long time already and- and the interesting part you said also for me, was that you said you have different go. You have additional requirements for ITP and RTP can't solve what you need.

G

So maybe it would be an easier solution to actually figure out how we can change our GP to make it fit to your use cases, because I think there are a lot of things you can. You can ask you, can reuse, basically or which already fulfilled what you need and maybe there's not much to add. I, don't know I mean it was a very short presentation for this whole big thing, but that's something to think of.

C

This yeah, I think we thought about that. Actually, so we thought about, for instance, defining payload format for delivering guy, so BM FF or mp4 files, aza a container of course. That effort has.

C

It has been known, for instance, with the mpeg-2 transport stream over RTP, which was at least reading the RFC itself, saying that it's a bad design like overwriting, for instance, the RTP synchronization procedures with em back to TS, sand colonization procedures, so I think there are like fun fundamental, like differences which led us to like the decision of going like more towards something that is different than our teepee.

C

But, of course we are here, I mean to raise this type of questions and see if we can do something about that using FTP or maybe reusing a components, but aren t has defined like it see it see, ends one. So we.

H

Are open to the so gory first and I'm pleased you've come here if you're going to be delivering 30 megabit I saw bit streams into my house and over my wires I would love to know that they were somehow doing the right thing. So thank you for coming here and is the chance to change your specs as you as you evolve, I mean, or are you asking just to standardize what you already have.

C

I think we are all so this could be like starting point, but what we have done there could be just starting point. We are open to changing that to evolving that ok way the group sees here. Yes, so 30 megabits per second I think I'm, not like exaggerating, I think we're looking at a 4k streaming already started in the u.s. like Oh a while ago, you HD streaming looking at the traits that are like really around 3040 megabits per second and that being delivered over like best effort internet. Of course, I mean the content.

C

Providers would usually have some agreements with ISPs, but nevertheless I think it's. If it's done wrong, it could could result in disastrous impact on the Internet. Well,.

I

I sigerson I'm, not that worried, because your starting point looks an awful lot like a Finnish system design and um so in the beginning. You said you met me. I said earlier that RTP doesn't work for you, but I. Don't understand why and and without understanding what your motivations are for doing something very different. It's gonna be hard for us to understand how we can help you um as an individual I would think the chances are slim that that what you have here will be what what comes out of an IDF effort.

I

If, if one were to be started um and- and there might be quite a bit of um you- know dissecting and refactoring going on. If we do something about, I think that understanding what the what the issues are with RTP, which in your we are very biased towards IDP, obviously because it's our baby and we think it can do anything- um would be- would be I think a good starting point right, a CVT.

I

What does it not do for you right and and um a lot of things you have on here, I think we have solutions for that. That could work in RTP based architecture, but if there's some fundamental thing that that is not working and that's why you need all this other things that that will be I, think a good starting point um and and I think, maybe that the next time you come, which is, of course, that it's great that you're here and I hope you come. um I would really like to know about that part.

I

So start in the beginning, but I don't need to see packet header formats in this presentation. I would.

C

Have liked to see RTP.

I

Doesn't work for us because you know we want to do this and it it's inefficient or it just can't do it for reasons I.

C

Agree I mean I tried to touch on that. A.

I

C

I think in few slides earlier maybe I didn't take enough time to to talk about that, but I agree. We need to look at case-by-case point by point and see. Can we do that so can I do self committee.

I

Really one big reason that is very simple for me to understand, because God I'm, not a media expert, yeah,.

B

Like a sake of time, I would propose to take this off. I know maybe onion on the list so that we can continue so yeah. If you had to pick one thing.

I

What would you pick once.

C

I one thing I would pick like self-contained multiplexed stream, but carries all components of one media session. I'm, not sure you can do that with that.

B

Yeah, so okay, I, would say so we're running a bit of time out of time. I think, there's interesting. I think it's probably wise to continue this on the aerialist or I think you're talking anyway with Gauri and maybe even Corey, you and me manage to meet tomorrow so for some minutes to figure out what's going on there. Okay, thank thank you very much so.

D

I think no hint over to Spencer would explain. Signpost is.

B

D

Now we do now, we do signpost.

E

So um we've had a lot of conversations, probably starting about taps off era- probably a little over a year ago with the revolution of transport and things like that, and this has gotten a lot of stuff going on and we were just wanted to help people figure out where to participate in this.

E

So this would be a Mia culpa, which means, through my own fault, but I, had help so glass of hell. So, like I said, we've been pushing very hard on protocol evolution in the ITF in the IAB and now in the eye RTF, and we want to involve people who need to be involved and we haven't really made it clear how to be involved. This is the this is the part where I am falling on my own sword and they'll be plenty of sword, sticking out the back to for the brother folks.

E

So here's where to go the course we're doing transport services in the taps working group. That's an active, open, mailing list that you all may very well know about.

E

Aaron falk has been doing work on how ossified is the protocol stack or hops and trying to do the kind of measures of ossification and what what protocols?

E

What what the roadblocks in existing network has look like. So that's the mailing list for that there's a new one. That's sponsored by the IV, which is a open discussion list for IP stack evolution, its associated with the IP stack evolution IB program.

E

You all would have seen the spud mailing list, which was used for discussion of the spuds buff and Dallas, and things like that so kind of anything. That's follow on to to that and then tsv area for general evolution. Discussions in probably appropriate for discussion of the presentation that we're about to have you guys may see references to a stack evo at ikea, an IAB dot org, which is a private program, mailing list for stack evolution. You may see references to the semi workshop I be mailing list, that's dormant! For now.

E

If they ever do another workshop, you would pop back up. Some of you may have seen references to something called UDP 35, which was kind of how very early in the conversation that turned into the IP stack evolution, IB program and it's obsolete. We're going to go ahead and close. That can.

F

You explain you d be 35 and semi I'm.

E

F

Are the references of those words? What are the expansions of semi.

E

Help get register rule romulus about the years yeah.

B

We could get listen to what was a question. Ok good! Go ahead,.

J

Hi Brian Trammell I be semi was stack evolution in a middle box internet. That was a workshop that was held in Zurich January of this year. I was basically the workshop discussion list for workshop attendees and then UDP UDP 35 stood for UDP as layer. Three point: five of the of the internet protocol stack, which was we had no idea what we were doing at the time and we're like that.

J

We we'd like to move forward and we are not sure what, but we have this kind of suspicion that um UDP might be part of the answer. So that was the thing that stuck to it and then you know. Basically, the only discussion on that list was more or less okay. This this list name is terrible. um No I mean referring to it as we're going to fix the internet with UDP is kind of is kind of terrible. We need to. We need to think about this, a little bit more, which is right, yeah.

J

We were naming it by what we thought the solution might be. You know at the end of the day, which is which is an.

C

Xl way to do.

J

Unintentional engineering.

J

So I got that answered. The question I would like to say this is this: is the public announcement of the fact that there is a stock Evo discuss at IAB, org and I.

E

Will send an Allison a note to that about that to the tsv area mailing list as soon as it, we have anything that looks like a network.

J

I've got network. I can send that right now. Actually, if that's okay I'll, do that. Thank.

E

F

K

Can you hear me hey I'm here to present our ideas enables for transport they prescribe it can be. We really try to follow all things happening in transport, layer, protocol evolution and the next slide. There is some background information. What how we perceive that please go to the next slide, so we have seen experimentation with transports using gives a space implementation. That is really good, because there is an app speed, evolution, comparative standardization or or being in just the browser's I mean some of these are and browsers, but so it's much much faster.

K

This fast deployment, worthless anization. That is happening mainly over UDP, for example, quick, but also, if spot whatever. That will be happens. That will enable similar things to happen. It also seen that the middle box issues are being addressed, so vindovix is SCM a given tcp bio formats and- and that is really putting that is really awesome. Eyeing at work bought and to enact encryption and sample applications like gaming really provide the hope and that that is enforcing to let UDP boss, boss, also to left probably encrypted EVP pass.

K

The latter for efforts here tops or the stack evolution problem, take tcp maintenance is evolution as well. Also quick is happening, so this is some kind of background and what's the, what is the scope of this presentation that song? That's a next slide, please this so based on that we put requirements on a DP framework.

E

K

To achieve healthy ecosystem and also to keep keep up the face face of like faster evolution, we wanted to also investigate what is the effect of an accelerated deterioration. What happens if many app developer start developing their own own TPS? What can we do to to still keep things going? Also yeah? How is it stable to keep the stability of in this case? So you have this requirement and- and we have some ideas- we have some ideas in draft and also I modified some of these ideas.

K

These draft was originally submitted to the more springy idea, but what is definitely not in scope is expected features of transport protocols, so this is about the framework hmm yeah this. So we group requirements. There is some control requirements, and this might sound strange, so we say that we should be able to enforce expected. Pp behavior implementations might be buggy or malicious on purpose, for example, condition control might be too aggressive, I mean today it's not happening.

K

We are having extremely friendly implementations in most of the cases, but if you allow any application to have a transport protocol, it can easily happen by mistake or on us, and we would like to protect other flames of the same user. We would also like to protect the user in a sense d, not overly do not send too much data by an application which which he wouldn't like this and too much data.

K

We would like to protect other users sharing some kind of resource with this slow, and we would like to have some behaviors to enforce, for example, condition control aggressiveness or a given condition, control MTU packet facing whatever we be whatever else we can imagine something to be enforced.

K

Also, we would like to an option to have to allow the path influencing transport protocol selection. Why is that? Because the pause, my offer enhancement it might offer some cooperation with some call transfer protocols, or that is the opposite of cooperation. It might be blocking some transport protocols and on purpose- and if we select such a protocol, then it will not go through the path.

K

We would also like to ensure that the user or operating system has control over things. For example, what transport protocol is selected for a given application, or what is the preferred resource sharing between applications of this user of DC, OS and application streams, and also, if there is any communication from from middleboxes, especially on behalf of the user, then the user would probably like to have the final word in that.

K

So, even even if the application is saying something and the user would like to have something different, he might want to change what the application is saying. So please go to next slide.

K

Also the requirements for for accessibility, so application should be able to access available transport protocols, so any transport protocol, or at least any general purpose transport protocol shall be possible to select by applications and it shall be possible to be inserted to transport. The conceptual framework I've listened to do a quick Bob of yesterday and, and that was the time I realized that quick is only implemented for for google chrome and it's it's.

K

If I understand it, right is not implemented for for the google youtube and if, if we, if there is general purpose transport of it, could very nice if that could be somehow inserted into the system, even if it's, if that is not in kernel and an be accessed by by other applications, and also part of the accessibility, as there will be more and more transport protocols, it will be more harder and harder to have consistent transport protocol selection, which means selecting the selecting, cooperating transport particular instances on both end hosts and in any middle box involved in in transport protocol layer.

K

Yet it might be in a way in some cases it's really simple. When you develop both the application and the server, you can develop your own method. But if you look for web- and if you imagine more than one or more than a few transport protocols for web, it becomes much more complex. So please do go to the next slide.

K

We do believe in privacy and security and and these are really really really important requirements. It's important to ensure confident e of entry and communications.

K

The we think that in some cases it may be justified that the middle box excesses and modifies the transport protocol headers or the transfer protocol itself, but in that case the engine content should be protected somehow differently and also. We would like to ensure security of internal communications for us to take reasonable us F for that to avoid that any third parties could exploit implementation of loads in transport protocol so, for example, encrypting or end or authentic, a ting transport protocol fields, including user space transport protocol, can help that.

K

Because that's if you make the encryption authentication ports able enough and you tap the end to end implementation, then it at least has to be the end hosts. Inserting something nasty and many in the middle cannot insert anything because that is not authenticated and can be discovered by by the authentication layer. So you don't have to be deaf prepared for for testing. It's not saying you shouldn't test your transport protocol. It's just saying that you should have minimal attack surface on iron, not to test it transferable. To go.

K

Please go trackside and we believe that there is a huge amount of value to allow the access provider to be part of the value chain. Then that is exactly accepted by one of the endpoints and- and we see two levels here- wall every selection between different trade-offs in local domain, quite a service or icing, which is most fit for the transport protocol and or the application. So there are examples you are selecting between lower latency and higher utilization, selecting between highest you put versus more stable throughput.

K

It doesn't normally mean that the amount of resources allocated to that given flow changes. It just means selecting different the free source right. We like to.

J

Take my place, mine, okay,.

K

But we believe in a in a higher layer of cooperation, then the importance and the quality of experience can be improved further by increasing the resource share of critical applications, which we believe may be made fair in the longer run, though, that need incentives and further consequences, and we, we actually wrote a draft about such examples.

K

Any of these shall be explicit, cooperative and externally, ball, which improve performance but which again might have consequences, for example, economic consequences, but other other consequence system not misuse, especially their increasing resource for, and we believe that it shall be possible for the end hosts to opt out and get reasonable default default handling, which means, at the same time, that any middle box has to provide demonstrated value accepted by the users to actually do something.

K

Also, and also, we believe, the different travel levels of trust shall be possible and which, which means different solutions from basically encrypting everything over UDP, hiding everything accepting the default treatment to even accessing content. If there is enough trust between the endpoints and the middle box, please go to next slide and any of these requirements shouldn't result in significant degradation of performance characteristics when achieving that the requirements I mean later loli lilo set up latency reasonable reboot, especially in long signaling, console conversation shot, be avoided, especially do switch, which delay packets.

K

This is right for the common case V. We imagine some exceptional cases, for example, when a transport protocol to be used can be downloaded to whatever system can store transport protocol and and then use the transfer protocol. But this is really exceptional drew please go to next slide, so we have some ideas. I think that we believe some of our ideas are covered quite developed by by our understanding of this spread media initiative, so that provides an in-band channel, a protocol for media box communication.

K

There is an explicit communication and behavior brother, and that provides the potentially authenticated under encrypted messages to the middle of xiamen. Spot might allow separate signal channel, it might be villains, but it doesn't really matter from this perspective, but this encryption to the amino boxes shall be different from them and encryption and we believe that is a very important piece of the puzzle to achieve a healthy ecosystem and I think I.

K

I think the discussions in sport are very interesting to us in general, so the details are are not about spot in this slide set, but please go to next I'd rather about trust and enforcement issues, and that is where we, we think that, within the device, the end user should control resource sharing and congestion control aggressiveness, which, which might, for example, require that congestion detection is visible for for control functions. So that means that congestion should be a should be possible to detect outside of the application somewhere below the application.

K

For example, if the transport protocol is implemented in the applications, it might require policing solutions in the end and first we might want to communicate the congestion control paper used to, for example, middleboxes, and also.

K

We believe that the end user should have the final word in what is communicated to media boxes, with what authentication keys he might have, and also there is some trust issues between understand window boxes, so what he can be used for a given communication. How can how can we get the the the right ease into the middle boxes to have an encrypted communication but at the same time to be able to have meaningful discussion with the middle box?

K

Also, how can the who can be kill different parts of the communication method, at the content, transported or and and what is the possible consequence of the middle box communication hope that can be communicated back? It's also an interesting question and I think autothrust question so go to next pile again. So again, how should control these? Should it be the operating system? Should it be some poet of the operating system like application store, should shoot or could it be sometimes than a friend or should it be the user?

K

Should it be something like I've community database on behalf of the user? Similarly, for example, adblock plus filter use all have all of these has reasons to control, and- and there are some examples in the following slides for this control- please go to exercise there's an example here. It's this API looks very similar attitude to the pep.

K

Hanukkah going down to something, we call the trust and policy controller, and in this architecture we believe that the trust and policy controller may receive rich metadata, because that is controlled by the user and the OS, and that controller might in a way translate this metadata to something like treatment and network information and at the same time it that should include removing privacy, sensitive port and it could deter my preferred treatment and other other metadata to communicate through something.

K

We call me double cooperation protocol using a database and using user configuration, and he may also influence transport protocol selection and, in a way, a beauty I. Seen in this concept is that the same application same operating system same database, but a user with different preferences might result in different treatment and network information communication to the middle boxes, but at the same time, for the user, it's it not into music. So it's not! This decision is not made when he is using applications.

K

The decision is made when he is configuration configuring, this trust and host controller. If he wants to do that, there should be meaningful default there, but I mean as an example, I could choose the reasonable database and the reasonable setting for my parents, for example, who are completely unaware of any kind of technology.

K

So please go to next slide. This is this is a different example. I I put some random boxes here, trust and place. The controller is included here. There is something called protocol selection protocol parameter, setting, policing quality of experience, erection some API similar to tops some API is different because the transfer goes including applications. There are many things here like George joined, batala detection, Polly, seeing something like spa dormido bus cooperation, protocol different middle box signaling, which it determines what to to have. We have transport protocols there and there are some different levels of trust.

K

So we do trust some of this code, but we don't trust or trust other codes last, and there is an example here. So if you, if you look at this application,.

K

But having a transport protocol inside if you go to next slide, I made that application red, which means that this application is doing somewhat unexpected. That can easily happen when the transport protocol is within and then one solution is to remove that application. But maybe that's that application user and something can be done, but it takes the time for the application developer to change his application.

K

So what we can do is on next slide is to is to change the rules in disgust and policy controller and and apply a nap all I policing solution for that, given application in this case and and to propagate, do such things to be to the database is seen enough and you don't have to act on the application immediately on that case, and you don't have to disable the application in media thing that case either, which can be an interesting consequence of of solving some trust issues here.

K

But but these are both of these examples, this one than the previous one I in a way wide example is not meant to be like final solutions or recommendations in any way, but to delist rate. What can happen if we, if we think about this trust issues, please go to next slide?

K

Okay, so this is a summary about Russell trust. We believe trust has to be handled within the device. User user should have control in the end, but that control should be almost invisible during using the application's, sometimes even even always when, whenever the user can accept, meaningful evil befalls there, and we propose a trust and policy control function which can do all this. On behalf of the end user and actually databases of OS vendors and network randalls might be included there.

K

Okay next slide, please. So in summary, we put requirement and transport protocol framework to achieve a healthy ecosystem, forced evolution. We propose solutions, or rather we propose ideas to meet these requirements. He thinks spot is really important there. We also think the trust and enforcement issues had handled and represented for some ideas for this, and then there are several open questions, especially in the area of trust and enforcement. What is the task of I-80 of their hear? What is next, what is missing? Is it I ATF thing at all?

K

Do you the potential gains with this? Justify this kind of complexity or or can we have something similar, simpler and good enough and I think that was my last slide.

J

Brian Trammell um I've been up here for a while I'd like to go back to that. Actually, you know that's a really tiny knit. Thank you for putting together a consistent set of requirements and asks for how this would work. My first question would be: there seems to be a little bit of a.

J

Preoccupation- maybe in this in this design, with not trusting the transport protocol and I wonder if you could say anything about that and why that seems to be a first order. Design principle of this system.

K

Because we believe that fast evolution means much more freedom and and if there is that much more freedom, absolute development, some application developers will have the is not so friendly transport protocols by accident or by lena.

J

And my friendly, you mean further.

K

J

Fairness: congestion control, yeah, okay, okay,.

K

And I I think I think we actually shoot hello I mean if you, if you take our transport protocols, idea transfer protocol I think this should be trusted by default because they are. We know that they are behaving well.

J

I mean you know that you're at sea says that they should yeah I mean, would I can I can implement tcp on top of spud yeah? You can I can do it in like a couple of hours, but I wouldn't want you to trust it, because.

C

I wanted to go fast, so.

J

I probably set the initial window to something like give a thousand but I, don't think I don't want to I, don't I, don't think you want to cross any TCP. I can write in a couple of hours, but that's.

K

The point that I make that's why I said by default, if you by any means, realized that this flavor of DCP has this problem, you might want to change your database and apply some kind of policy which you guys and.

J

People in line behind me feel free to throw me out of line is this is gonna? Take a while, um let's go backward, um can you show me the show me, the big, the big color for architecture slide? Please this one? Yes um yeah!

J

This is secure, so well the thing that the thing that's here that isn't like, so you have a set of requirements, you're, basically pointing to a bunch of other work going on in the IETF, and thank you for it for at least I read the draft and it is a coherent vision of the future. I won't say anything more about it.

J

This seems hard and dystopian.

K

A

J

Yeah um and I don't want to say anything uh nice about other standards organization, so I won't um I'm okis on the hardness, in that the trust and policy controller is now on the attack target in the attack service. This is the basically the place. The different one on the Internet I will own that in order to make it work at all, you need to solve the key distribution problem and, if you're going to solve the key distribution problem, you can apply that solution to better things.

J

Can you go back to like slide 6 I think it was slide 66.

L

J

No nope nope nope 72.6 the thing that scares the hell outta me here or is the statement. It shall be possible for the end hosts to opt out of middle box cooperation.

K

Ornithol theme yeah: this is it because.

J

It al giant difference between those two things and I think that that kind of exposes the vision behind the vision here that that whole bit makes me a little bit angry, actually that the idea that ok well we're going to we're going to build something in the middle of the stack that's going to handle all of the trust and distribution which fortunately, I won't work. So that's good.

J

At least we won't be able to build that part um and and then we're going to make it transparent to the user, so they don't see what's going on and maybe, if maybe, if the application developer is nice to them, and maybe if they know enough, maybe we'll let them click the box that turns off the value-add and I. Don't think that's the Internet I mean I, don't know, I hope it isn't.

K

Not I mean that's enough that they know the purpose with this one. The purpose is to allow this and I. Think opting in is a is a better I think actually from what you say, I should have definitely written. It should be possible for the entrance to opt in. Yes, she had.

J

A very minimum.

C

K

If the middle box cannot provide any demonstrated value, then he will not they're.

J

Very minimum um I'm, actually gonna I'm gonna go get back and wine after.

A

That there is, when I understand that.

C

This state, your name, may come on.

A

Microsoft, hey, there are many requirements expressed here and what you listen here is also what was presented as the motivation for spurred and other things like that and, and basically I have a hard time, believing that it is either in good face or good for the internet or whatever of it. What there is a real problem.

A

The real problem is that if we want evolution possible, we want to be able to not trust the end-to-end transport I mean there is no reason that network shelters, the end-to-end transport, and it is true that the current situation in which we believe, if that the end-to-end Transport, has to be at in a certain way. Otherwise the whole internet will collapse, that's actually very fragile.

A

So so what we should study is what is the proper way of in the network, ensuring that the network is stable, even if the end-to-end transport is not as I trusted to do all that kind of stuff, and there are solutions in the in the range of active queue, management, traffic or filing, etc. That makes sense, and you can do active you.

A

It's happening mmm that you you, you make sure that each of your users are the same as this by and large cannot step on yours or users food.

A

If you do that, then it doesn't matter what particular transporter to use our users, because when they cannot help the other guys and I think that if we take that in those terms as in I, provide I manage my network to make sure that even if someone is beheads, it only affects himself and not the other guys.

A

Then we have a much much more decoupled interface. We don't have to go spy into exactly what the other guy is doing. We can have a generic contract and a visionary contract like that is probably much easier to have than trying to go. Look at exactly what is implemented in each box.

A

K

I think the thing you are saying over as I understand, it is a searing day that we want to have this kind of, like always, fair share for everyone and I.

A

Think I don't know that. That's that's an easy to understand. Example of one possible policy. You might have a policy that says I mean if you pay ten times more about.

K

Whatever it's not about, okay, share, independently of the content and the importance of the content and the state of the application, absolutely.

A

K

A

If I have my, if I pay for my network, who are you to say that this stream of music is more important than my gameplay I mean I,.

K

Think that's why it's very important to be able to switch this whole losing hope. I I do believe that there are cases, then, when the resources are released, cars and and then you can have much better over a quiet of experience out from your network. If you allow that kind of differentiation, and if you don't accept that it's really funny.

A

I truly do not aight witness. I I particular I. Do not believe that there is a trade-off between latency and earth or between will I began bandwidth because those kind of thinkin, if they are real, they can be dealt with end-to-end with proper application for our collection or whatever. Then you can do that.

K

I think there are entering solutions. I think solutions at the batala can be more effective. Also, antenna solutions can I mean like just take an example. If you stake fair queuing and you take take lad, but then fact Ewing most likely destroys the lad, but behavior look.

A

I mean I: don't want you d to debate the details on how to implement queue management? mmm What I'm saying is that please.

M

Use the mic to you: okay,.

A

What I'm saying is that expressing the problem in the specific term of how to prevent congestion of the network? It's much more likely to get us a good architecture than having an abstract discussion about the abstract, possibility of abstracts middle boxes.

K

Yeah it's I mean then then, in a way point taken. We can. We can have examples of like concrete solutions where this can be useful and- and we are, we are happy to accept comments there why it can be done without communication.

L

Hong Kong hi, presently a large pizza pepperoni to provide a seasonal sinking of the temple service, but I, as here are some concerns on some technical issues. Example: the middle boss. We all know that the in the pastures are multiple cascades, midibox, you're, you're you're, talking about the middle, both collaboration, McCutchen's lomatic, formidable in cascade, how to fund the sender or receiver talking or covering for them beetle. Not just doesn't Nestor hop in a box. Oh yeah.

K

That's far from being trivial, I I think there is a. There is a simple case when there are two access providers, there are basically middleboxes of deals to access providers and when it comes to Tranzit in between is the task of of the access providers network to find something right for for this kind of traffic. But I do understand that there is this complexity of several middleboxes and I. Don't have the answer for that.

I

Pars occurred so when you, when you have a network that has resource constraints right, you.

C

I

Either spend money on doing something like this or you can spend money on putting in more capacity. Why would we do this instead of putting in more capacity.

K

Because you're, probably.

I

Hungry because.

K

Hopefully it costs less or there is a point till you can so more capacity like.

I

Even if you could put in a little bit more capacity, so this stuff will never get you more, but this is this is this? Is all about policing and regulating and making you know, scheduling that just put in more capacity right, it's building an architecture around administrating scarcity is not a good idea.

K

Yeah I mean if, if the cost of additional capacity is reasonable, that I do and do you agree with you, the question is: why are we having these in many countries, the stupidly small, like mostly cuts in mobile network? This is because we create artificial under load in the mobile to be able to operate without, but your service, or is it just the pricing trick? That's that's a good question. What.

J

So if we're going to start talking about pricing and costs in and end capacity, Brian Trammell, um if you can go back to the big big colorful bunch of boxes slide, so no way actually go back to one slight before because it's actually, this shows the interactions, so that trust and policy controller is going to need a key for basically everything that has an as a capability associated with it. In.

K

This particular idea for.

J

K

J

The amount of money that it's going to take me to build and operate that pki, how many 10 gig links can I get and how many more, how many more radios can I put on the power I'm guessing it's more than 1pk eyes are not cheap. All right, I mean not even the engineering effort of trying to build this I mean this is a. This is a boxing line, diagram and I'm, not gonna, actually go into into power.

J

You'd built the Fox line diagram, but but this Russian policy controller is going to need to basically make I mean even just the even just that the I've got all of these roles and I'm trying to express the policies to it. Even just that part is an enormous amount of management. Overhead, mmm yes, and.

K

Then the building.

J

The pki, in order to make that work and getting all of the keys to all the devices, is an enormous amount of overhead hi I'm, trying to see where the win is.

K

But I I mean to I I. Think I get false a few points. I might have the photo part of the point. So for that, but still.

J

They have two points. One is once you build this there's so many other really evil, evil, evil, evil things you can do with it that I don't want anybody to think that this is a good idea um and two I. Don't think that it is in economic I mean unhappy, but that I don't think it is an economically viable idea, but so.

K

But then that I have a come come to a question: how much did it take to develop at back glass for for the roses? How.

J

Much will it take to develop adblock plus for the browsers and.

K

They did it help the users, you know I decision or did it did it hurt that uses decision I'm.

J

Not sure what the I don't understand, the question: okay,.

K

Adblock Plus's I know.

J

What advertisers, but.

K

That that wasn't developed by the operators- it now don't have to be necessarily developed by this. The operators dist austin police.

J

Know somebody has to develop.

K

J

K

Going to develop right so very good question someone who would like to get the advantages of, for example, midibox cooperation, for example, selecting transport protocols, but at the same time, would like to have control, and that should I think it could be a community development and and and what I am vision. Is a community development, not an evil corporation development. For this one, and especially the day, even if this is developed by the cooperation, I think it should be Sybilla to replace the database to a database controlled completely by communities.

K

B

K

So it's clear that we.

B

Closed the lines after Matt.

J

I will take this offline because I can't.

A

J

F

Matt Mathis, um so one of the things that is influencing design of the internet today is that there is a the trajectories for the cost of memory and the cost of bandwidth or actual actually have substantially different slopes, and the problem is that memory has to beat Moore's Law squared in order to keep up with the links.

F

The reason is is because, as the links get faster, you both need to make the memory faster and you make the memory bigger, and you see this in things like even simple switch high speed switches, the queuing time is going down because you can't afford the memory to make the queuing time anywhere near the delay bandwidth product, the state burning, this kind of protocol in the middle of the network blows up compared to the cost of the bandwidth you're you're, essentially on the wrong side of the the migration.

F

What we're seeing today in the urn, that is, that the edges you can do fancy things like red and all that kind of stuff. The core is doing drop tail on a 20, millisecond, cue or something or smaller, and those cues are getting smaller. As the switches get faster because you can't afford the memory I I.

K

Completely agree with your segment, my vision, dar, is that there are boxes policy controllers, for example, in in cellular architecture, which do you look.

F

At relatively close to the edge mmm they're, relatively close to the edge yeah.

K

And they should in a way apply existing quiet or service techniques in both directions based on this. But it is definitely not something to be terminated in high speed switches and it's not something to be terminated, didn't most switches, all right dose or whatever.

B

Okay, thank you very much.

B

So before we close the session humidity early, that's just a friendly reminder from people to think about if they would be willing to stand as a transport error director nominee in the coming selection process- and you have comment la razón, yeah.

I

Lotsa girls actually have a um so I'm a bunch of people that are academics in the transfer layer came to me over the years and also now and said I would consider doing it, but I need support. You know my university. Isn't supporting this or um I.

I

Don't have a project that supports it at the moment, and so, if, um if you are at a company and I'm looking into this for a netapp right, but if you are at another company, please look if you can fund the university professor, even part-time right, if a bunch of us get together and pool money, um if you can make it impossible for them to run. That would be good for everybody. So I'm happy to be the contact point for anybody with money and I will or.

B

You will be the contact point.

I

B

You are stepping down and right it's an.

I

Offer from from our side, let's see.

B

If people are afraid to talk with large, because there could edit or whatever you're free to talk with Lars, but we expends what it is, cannot so over to be like the clear on point, we can come to and say like by the way we would be able to offer money. Yeah.

I

And it doesn't need to be the full thing right. I don't know about a full-time, it depends what I costs right, but it can be a fraction of that. If multiple do it.

E

Right so just to bring back up something we talked about previously in Dallas, we spent a hour.

E

On Friday effort on the last session in dallas on friday, talking about you know what it was possible for us to do to make things easier for people to be transport area directors we ended up with. I guess five sets of questions that came out of that. The group we would I tried to talk.

E

I've tried to start a conversation about this on the tsv area, mailing list kind of sending, in a summary of that discussion from the minutes and I have gotten some input, and actually one of the things was we I went in I went into this session on Dallas kind of and I think other people did as well kind of thinking.

E

Well, you know if you had companies that were supporting an area director that that would be a point of conflict and that right, you know what I'm giving as a result of us having this conversation is that that does not seem to be as big a concern for people as I thought. It would be so I I thank people for the interaction that we've had so far because it's been clarifying for me and for for Martin and I think for the rest of the area.

E

Having said that, that's really a conversation that we need to continue to have really seriously within the transfer area.

E

Oui oui, the I ASG, said that we were going to try to do some process. Experiments that we'd be announcing about this timeframe, to try and do try to arrange things so that the time requirement for transport area Drescher's would be less and and actually area. Directors in general would be less and we really haven't come up with any experiments, so I'm.

E

Thinking that the more work we can do within the transport area, to arrange things for transport area directors for us to increase the number of people who are willing to be considered as nominees um I, think the better that I think the better. It may turn out that there comes out a great set of plans from the isg. That's how our problems- and that will be awesome, but that hasn't happened yet, and you know meanwhile, Oh Harold I think said: he's sending out the call for nominations next month.

E

So please keep those conversations that conversation going on on the tsv area mailing list, because it really matters to the transport area.

F

Memphis, google, so I tried to have this conversation inside a google a couple years ago and it went wrong in two ways, one of which is easily fixed. The word support is actually a euphemism if it doesn't contain the word financial and that actually got misunderstood.

F

It required a cleric some iteration and that caused the second thing, which is a perception of conflict. We've been bringing Google in particular than bringing a lot of business before this area. Directorate and people said we can't risk they're being problems from that and the fix to that is probably doing something at I'll say. I sock level that may not be the right answer, but it's some broader level where the money is blended for multiple sources and multiple reasons.

F

So it's not specific, not at all. It's there's some distance between any policy and and the funding sources.

B

So you made the comment last time and since then we both haven't been active, so we are talking in the back already, but we don't have anything. We can tell you. So it's just like.

N

Yes wonder what Matt said: aaron falk, I am, the network came up. There was nothing on the room for a long time, but then there was one comment: I kind of missed it. So rolling back to the discussion. The last presenter that we had there's a question about where this discussion is going to continue.

B

So the UM the idea is, if this is like general transport protocol evolution discussion that should take place on transport area. If it's a discussion about fines and offered services that should go for taps, for instance, but if it's like general, how do we will transfer protocols? What are the requirements? What are the mistake we have seen in the past or things we haven't fixed? That should happen on transport area. So far,.

N

Okay, so aaron falk. I have a comment on my own about the area director thing I just want to point out to those people in particular. If there's any academics in the room or listening that this whole stack evo thing is actually going to bring a lot of interesting work into the ITF. It.

M

N

Has and I think it'd be kind of an interesting time to be an area director and be in a position to help guide that work through and help make it successful. I think there are lots of examples in the ITF where the area director has played an important role in making a work, be coherent and working groups be focused on the right stuff and run overlapping and current etcetera etcetera.

N

So I would encourage folks that if you are, if you have a research interest or an architectural interest, that this is actually kind of the time to be a transport ad, that's.

E

Dirt excellent excellent point: um I I, know that more many people in this room know this better than I do, but, but just the just the idea that you know transport protocols were kind of the land that time forgot as far as us being able to get anything deployed for a while and that you know that just isn't true anymore yo. You know we're we're working beat through that now we're that was kind of a stall yo, a stalling point for a bunch of years, so I agree very much with uh with Aaron this I.

E

You know this is this is a great time to be in transport.

B

All right, so, if anybody's here in the room who hasn't signed the blue sheet, yet you can still come up front and sign it, and otherwise I think we are spot-on. It's nine minutes our past seven and time for the bits and bytes. Thank you for being here and see you next time and.

E

We should mention that we will be at this and bites if anybody wants to chat with us about being an area director.

B

D

Ablution and they.

M

B