libp2p Team Meetings, 15 Apr 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: 2019-04-15 libp2p sync meeting

Description

Notes here: https://github.com/libp2p/team-mgmt/pull/31

A

Okay, hello, everyone: this is the lid p2p. We can see April, 15 or 16, depending on where you are. um Let's just get the Crypt bad, open and I'll post the link, but I have it.

A

Alright, if you don't have the crook right here, it is in the chat for the resume.

A

Okay, so I'll be moderator. Does anyone take notes.

B

I would volunteer but I'm having some net issues. So maybe not this time. Okay,.

A

All right look any other volunteers: I can try. I can also try take notes if no one is willing to do it. I.

C

Can take notes if necessary, okay,.

A

All right we'll take it John I'll, take it all right. I'll put you in a recording link. I will add later so the only so the only thing we have on the agenda are the we should changes to weekly updates yeah. So the only thing we have on the agenda are weekly updates. Is there anything?

A

Well if there's something else, you want to add to the agenda: do it in the as a background task and we'll run through the weekly updates, and then we can come back to it or else, and the meeting okay, so I'm, just gonna go in the order that they're in here at accounts, you're you're, oh you look unfinished I can't tell you wanna, come back to you.

B

Know I can start okay, good I've just been experimenting with keeping a work log, so I've put everything it just.

B

I guess I start with what I'm blocked on I'm still waiting for functional gauges to be supported. You know, consensus that'll, make a lot of stuff a lot easier to do, particularly around the routing table and some other things.

B

I don't want to start having metrics for and I'm stuck on trying to view last week's meeting, I guess we'll sort that out I would have a band there, I'm, currently working on a consensus, tracing and I'm using it I'm just experimenting with it, essentially how to get in various places, I'm hoping in particular debugging stream, pooling stuff and the other one that would be interesting. Is that provide there's an issue about provides taking a very long time.

B

I think tracing would be very insightful there, so I'm working towards that and the things that I've done recently I've spoken to several people from file queen who had a lot of DHT questions. Those were very productive conversations reviewed a few metrics related PRS and we've merged one into the DHT, with some basic stuff to get started.

B

What else I've started spinning up a few ipfs demons in Australia I'm, just experimenting with them tracking metrics on them see how they go just running them on BBS and that kind of thing. So that's mostly what I've been doing.

A

How many sounds good sounds good yeah, let's keep going through the updates and then okay, so the next one is Jacob he's not here. As for its mellow night for him, so I will I'll. Just let people read read this themselves. If you're watching the video you can find it in the lid p2p / PM repo issue number was 11 I. Think, oh sorry, sorry, no and not it's not an issue. It's uh it'll be a PR emerged. Pr, the Tories link, J's notes.

A

Okay, the next person is me so yeah, so I wrote up all the performance reviews now. I have to write the summaries, which is more work than I realized. The security audit is also go. Live. Pdp is also starting this.

A

We are sorry it's technically starting next week, but this is the week where I gather a bunch of background materials and send them over to the auditors so that they have some context so I'm working on pulling that together and those are the two I think those are the two main things don't work out for the project. I.

A

Would this some of us are going to Berlin for a conference in May and I want to see if we can do immediately to meet up, but I'm also not sure I have a lot of bandwidth between security audit and all the other loose ends. I have so I'm trying to I?

A

Have this idea generally that we should try to do when whenever a critical mass of people are gonna go to be in one city at the same time, which you just tried to do with p2p meetups, so I'm trying to see if we can make that happen in time for Berlin, okay,.

C

Dad Tara Nemo yeah.

A

C

Of forgot that what's happening again, let's talk afterward, okay, sure.

A

I can talk to you, yeah I do think I do think it's sold out, but let's come back to it afterwards. Okay, okay, so vasco is next also at middle a night for him. So I I just kinda. Let people read the read the notes: okay, John you're next.

C

Alright, so last week, I was looking at totally big P and the these api's and data structures behind the swarm and the host, and this week and I was giving some thoughts to what does what does what do we actually need to handle messages? In addition to streams in particular, sort of the ultimate goal of messages is to be as stateless as possible, but there's various places. You need state like where you doing the encryption, and you know if we have you know different different names for protocols across sessions.

C

We need some notion of sessions, so the question is: what does this look like, and can we can I come up with something that handles both streams and message as well, so I've been trying to propose something for that this week, I need to talk to Stephen and I think it was Youssef who was interested in this architecture, so I'm going to talk to him talk to set up meetings.

C

I also want to understand what our goals are with with relay and kind of how it got architect of the way it is so I want to talk to be so about that as well.

D

One question: these message is supposed to be a reliable.

C

Unreliable at least initially but I think that's a great question of how do I I'm thinking when I'm thinking messages, I'm thinking something based on UDP initially, but then there's the question of what do we want to be able to layer on top of that to get, for example, reliable messages? So that's a good question. Yeah.

D

We should probably talk because in quick message proposal.

D

C

Let's do that my other question actually related to that, since this came up, I'm really curious about thoughts on using encryption in these protocols versus something on top. Obviously, from you no point, if not reimplemented things and not having crypto bugs clearly, it's nice to use existing features but I'm thinking in terms of if we're relaying. What does that look like I, don't know if you have any thoughts on that, because you really want more end-to-end encryption, for example, right.

E

So for me, I got I would actually like to expose relay before encryption, just as an option where users can choose with what they want to. But that's that's gonna require some Rebekah's. What.

C

Do you mean by relay before encryption where well, it's.

E

Hard to do you.

C

Know I think the fact there's nobody doing.

E

C

That's what I think we should do J, but the problem is that just means we can't rely on encryption as part of an existing protocol. Well,.

E

So, ideally, the shell be layered with all this rain. So like you negotiate something abusing X, then you go to the next thing. You give a transfer automatic, rips, whatever, but suppose certain services on transports that dope on Mac encrypt before the encryption layer, the problem this is like. If you do anything at this layer, it gets really dangerous because now, like any control, message drill, it open any messages, control messages and people will start messing with them. Just because they're trying to be nice like right.

C

E

They're, not least what it doesn't matter: they're still gonna mess with them and do things you don't want to do, which is really why everyone's move just say: encrypt everything we don't care, it doesn't cost that much the nation. Actually encryption is negotiation, the very beginning with RSA. For us, the actual encryption of the screen is like pretty much free, so doing a link, layer, encryption and then like an independent christian, is probably okay.

C

I mean that's. That seems to me the most practical way forward, but we should set up a meeting to talk about this. Yes,.

E

And if you'd also like to discuss packet really versus circuitry, like some limitations of circuit, really yeah,.

C

I'm not surprised, let's cut this offer right now, but let's schedule a time this week, I think I. Think Yusef was interested in some of this too.

C

Okay, that's it for me.

A

A

John I took notes for you. Thanks now turn up pass a baton back to you again: I'm Martin, you're, muted, I.

D

I'll start with what I'm doing next next on my my plate is dealing with multi stream 2.0.

D

Right I think I'll start with it. This week, in the last, in the last two weeks, I worked a lot on on quick tracing, I added support or for the big trace tool to to my quick implementation and immediately found a few bugs with that and fix them, and.

D

Merged all the changes for the next ipfs release, so quick support will be. It will be a lot faster and effective to working with, with the key update design team in the in the quick working group. This one there's one area of the spec: that's not not really clear.

E

At the moment and.

D

We are, we are precisely defining when, when you're dropping out, all generations of keys I could demo the quick trace. If you want to I, have I have to trace this here might be I'd be so people can see what quickly.

D

E

A

I was saying yeah.

D

You see anything.

A

D

Okay, so this is this- was a quick trace, it's written by by Google it's and it's basically a log file format and a and an Open GL based renderer. What do you see here in blue is packets being being sent out and you get all kinds of man metadata here you see which which sprint frames are in a packet.

D

So you can see here we are sending out a bunch of packets and then in green. We see a hundred milliseconds later because I, the RTT on this connection is 100 milliseconds we get an AK for those packets and it's all, it's all nice.

D

So when you look at this was one of the first traces I got out of out of quick trace and what you immediately see that things are going going wrong here. What we are supposed to do is send out all the packets spaced out over the over the RTT kind of wanted to have a smooth line here and we are sending out all the packets, basically in one bunch, so waiting 100 milliseconds for the for the ACK and then sending another bunch of packets. So then was the eye fixed.

D

Now the pacing rate takes care that we send out. That gets continuously.

E

Quick question: should we just be increasing the window such they're, always sending or receiving packets, or is that not or do we just slow down or some rate so.

D

You're increasing the congestion window every time you get an AK or what you want to put, but let's train on the network. Yes, you want to space out the whole congestion that you have at moment overload the RTT.

E

God, okay, so that way you have to just yeah otherwise you're, basically, because.

D

Otherwise is that if they are like small buffers in between and you send a whole congestion window at the same time, you might lower you just imagine, I first get egged lost at the tail and then yeah, just a new window will draw so I have a second trace. I think my screen is much hurt anymore.

D

A

I think we lost your screen.

D

Okay, so this is after I fixed the back. If you can see like things, are going going a lot smoother here. What I'm having here is the connection with a 20 millisecond RTT and a fixed bandwidth, and so it so you can see like we are sending our data. We are getting we're getting an axe, very creasing, the the congestion window, and then something happens here. Let me zoom in so in red are packets that are lost, and this is exactly what how congestion control should work.

D

So what you do is you send more and more data into the network until packet loss happens and then, when packet loss happens, you're you, you stop increasing your congestion window, you reduce it a bit and keep sending it that right, and by doing this, you find out what the bandwidth available is- and this is this is what we are doing here. So we see it like there's packet loss occurring, we are not increasing the congestion window anymore and quic is finding finding the bandwidth of that connection and then continuously sending with a penguin.

D

So this this quick trace I find it really cool. It's super helpful to debug. What's going on, if you can see, you can see it plotted like this and then they'll probably be a few other bucks valid point, but the one I already fixed the one with the pacing rate, it already give the speed of like effective to connect stuff. That's a pretty big deal. This.

A

Is really cool so if you zoom it in the top right, there shouldn't be any red anymore unless the characteristics of the connection change that right, y-you-you.

D

Continues the condition control continuously probes, the network, so you'll have some loss from time to time, but but it will be will be very, very little loss. Okay,.

A

Yeah, okay, now that's yeah! There's.

D

Basically, no loss going on anymore.

A

Yeah yeah, okay, I, understand Thanks.

E

You mentioned a motor stream 2.0 spec is this anyone that I can see. Excuse me I, believe you talked to a bull, the stream tube in it right, yeah.

D

But I said it's mixed.

E

Okay, sorry I think that you finished limited okay, but.

D

We should probably talk sometime. Yes,.

A

Yeah I think you guys and roasted all talk, because I think you have different conversations going on like pilot bilaterally search, so yeah.

A

Okay desert, so those are all the updates.

A

Yeah I don't know it doesn't, does anyone have anything else they wanted to discuss or do what guys want to call it a day.

A

All right thumbs up, I, don't hear anybody saying they have more topics, so I'm gonna call it okay! Well thanks! Everyone for coming and you'll see this video it'll probably take me a half an hour or something but it'll be up and yeah. Yes, John I'm, hoping you can figure out johner and acrylics. Maybe you guys can figure out with Cole what what happened to the old last week's video I think you probably didn't get in, but this is so file. Guess.

A