P2P Networking Call #3

Transcript

Kamil Salakhiev: Right. Make sure you request one more time. You have to do that. It's going to try. Thank you. Okay. Hey, Daniel? Hey, Justin? Yeah.

Justin Traglia: Hello.

Daniel Knopik: Hello.

Kamil Salakhiev: We do. What was that? What was that? Excuse me, sir. I don't know what could be.

Anton Nashatyrev: Hello.

Kamil Salakhiev: Awaiting.

yiannisbot: Hey, how are you?

Kamil Salakhiev: Yeah. Hey, Jens.

yiannisbot: Yes.

Kamil Salakhiev: Excuse.

yiannisbot: How you doing?

Kamil Salakhiev: Good, good for you.

yiannisbot: I think.

Kamil Salakhiev: No, I didn't That's that's what we put. You know. about this. Okay. Yeah, let's wait maybe another minute and get started. Yeah, okay, yeah, let's get started. I think today's call not gonna be too long. Yeah, we actually had some presentations planned, but they were canceled last minute, so I think next time.

Agenda will be, kind of more packed. So, yeah, let's start with client updates. Yeah, maybe start with Lighthouse. Sure.

Daniel Knopik: You can hear me, right?

Kamil Salakhiev: Yes.

Daniel Knopik: Yes, perfect. OK. We released a new version last week, and that one ships partial columns, which is enabled by default on Hoodi and — on Sepolia, and from what I can tell, the network looks quite good. So… like, so… our match degree is 6, right? And currently, I'm seeing… Around 2.5 of those six peers on average being partially enabled. So those nodes have rolled out RODs. And I'm actually seeing a drop in node bandwidth on those nodes. Now, last week, it looked kind of same-ish, but apparently around 2 to 2.5 is the point where you can see the first benefits. So no bandwidth dropped, incoming bandwidth dropped. Total bandwidth dropped from 2.5 to…

2, mebibytes per second, so we're seeing an improvement. And actually, the latency, so the, delay from slot beginning, To, data availability has also dropped from around 1.3 seconds on average to around 1.15 on average, but that data is quite jittery, so I'm not sure if it's actually significant yet. But yeah, we're seeing improvements, so that's good. No. Yeah, that's pretty much it from my side for now. We also shipped some vulnerability fixes with that release, so in case you know that you haven't updated yet. Do so.

Kamil Salakhiev: Cool, nice. Yeah, sorry, proceed.

Aarsh Shah: Sorry, Daniel, just some quick questions. First is the improvement in the GA wait time. Have you also shipped the HasBlobs changes, or is it all GetBlobs V3 for now?

Daniel Knopik: Let's just get both for now. So in practice, the latency does not be so bad on the real network to cause…

Aarsh Shah: Okay.

Daniel Knopik: any problems. It's also to be noted, I forgot to mention that, that of those nodes, two are in Australia. And, those actually, benefit more, from that, from, from the improvements than the nodes in Europe. So, it seems to be especially beneficial, for those nodes, which is also nice, to see. Right, and I think on Hoodi, the blob traffic is, of course, far lower than on the blob devnet. So I… so I guess that might also be a reason why GetBlobsWith3 is not as, slow there as on mainnet. Also, I assume, like. I guess also had some… Fixes, right? I think those might be released already, so… so maybe that's… again. a reason why it's not as impactful. But as soon as we have kind of broad adoption on the network, I will also roll out Hasblops on the Sigma Prime nodes in order to see

Yeah, how much there's actually an improvement on Hoodi using it.

Aarsh Shah: Sorry, what are the Sigma prime nodes?

Daniel Knopik: I was saying that I… I'm planning to roll out HasBlobs to the SigmaPrime nodes in order to, see, yeah, measure… measure the benefit.

Aarsh Shah: Okay, and Yeah, the other question I had is, Daniel, do you have a timeline on when you plan to ship the Gloas changes? Hoodi?

Daniel Knopik: I mean, before, I don't think there's actually a timeline for that. Of course, we intend to do so before Gloas ships on Hoodi so that it is immediately online with Gloas on Hoodi itself. There's not really a pressure to release it before, so… yeah.

Aarsh Shah: Sorry, release was maybe the wrong word. I mean, getting the changes ready, and maybe, you know, we can test it out, like, Blob DevNet 0. the… So, like… On my side, I'm almost done with the Gloas changes for partial columns. I'll open it up for review by the team end of this week. So, I was just wondering, you know, if you have a plan in mind where we can deploy the Gloas partial column changes to, say, BlobDevNet0? And just see how it works out, and flush out any bugs we may discover in the flow.

Daniel Knopik: So my PR for Gloss is also ready for review. And I actually deployed that PR along with HasBlobs and GetBlobs34 to BlobDevNet 0. Of course, BlobDevNet 0 is in Fulu.

Aarsh Shah: Yeah, but.

Daniel Knopik: the code path is not actually active, but at least the full loop code path apparently has not regressed, so that's already good to know. And as for testing it with Gloss, I'm not sure if BlobDevNet 0 would be a good venue for that. Maybe we should just piggyback onto Gloas devnet, whatever number.

Aarsh Shah: 6, yeah. Yeah, I think it's gloss. Yeah, it's a definite 6. So that's a good idea like we can our partial column changes as well, on top of the GLOWAS changes they already have on DevNet 6. Instead of using BlobData. Okay, that makes sense to me. Yeah.

Daniel Knopik: Mmhm.

Kamil Salakhiev: Could someone remind me, do we… I remember there was some issue with blobs in Glamsterdam DevNets. Were they resolved? I mean, do we have blobs in Glamsterdam DevNets, or it's not there? Awesome.

Daniel Knopik: I think so.

Kamil Salakhiev: Oh, okay.

Bosul Mun: It's been…

Kamil Salakhiev: Yes. Okay, it was resolved. Great. Okay, yeah, then, thank you.

Aarsh Shah: Kind of makes.

Kamil Salakhiev: So, you're allowed to sell a little dentist there, and you see how it goes.

Aarsh Shah: I'm sorry.

Kamil Salakhiev: Another interesting point is.

Aarsh Shah: That's.

Kamil Salakhiev: To get to know is, like, how many…

Aarsh Shah: Sorry.

Kamil Salakhiev: Notes we actually have that support partial messages. I don't know, actually, the thing from Chaba that he was supposed to present would be very helpful there, because it's basically just teasing some… yeah, it's… allows to analyze the network, I mean, how many nodes run, some protocols, and so on.

Daniel Knopik: Yeah, I mean from the mesh degree, from my observations in the mesh, it's like 2.5 out of six, but I guess that's not necessarily an accurate measurement since it's only, you know, my very limited view of the network based on the sigma prime nodes.

Aarsh Shah: Yeah, so…

Daniel Knopik: Oh, yeah, but I think…

Aarsh Shah: It's.

Daniel Knopik: Yep.

Aarsh Shah: Yeah, that's fine.

Daniel Knopik: Like 30% does not sound that unrealistic.

Aarsh Shah: Okay, so…

Kamil Salakhiev: Yeah. Anyway, yeah, just curious, maybe it's possible to web code some kind of crawler to just get this data. Yeah, it would be very interesting to know that. So… Oh, cool. Yeah, nice, proceeding with quite a bit. I think Arsh reported for TRIZM already. I believe, yeah, we have…

Aarsh Shah: Yep.

Kamil Salakhiev: Also, partial messages shipped, right?

Aarsh Shah: Yeah, yeah, so.

Kamil Salakhiev: Is that okay?

Aarsh Shah: So the Fulu partial column changes are already merged to our release branch. We plan to cut out a release sometime next week, and, you know, then it will go out. For Gloas partial column changes, I'm actively working on it right now, and we plan to, you know, do an internal… that'll be done this week, and the team will do an internal review next week. And, yeah, like Daniel said, there's no hurry to, like, merge or release that, but we can definitely start testing on DevNet 6. Which is my plan as well. And, yeah, most of… I saw both the… the… I saw the PR that you raised for… Get Globs v4 and, ForkchoiceUpdated updated v4 in Prysm, so thanks a lot. So, yeah, I'll definitely do a deep review early next week and do some testing myself as well.

And yeah, thanks for that, contribution. So this, this is it on… For our side, for P2P.

Kamil Salakhiev: Yeah, I think we should do this more, just, don't wait for clients, just implement changes on their side, and yeah, ask for approvals. Uhm… Yes, speaking of GitBlobs before, yeah, Daniel, do you know, like, any plans on Lighthouse side to introduce that at some point as well? Or it is already done, I don't remember. There…

Daniel Knopik: is a draft PR. So that is suitable for testing. There is like one or two code quality things to clean up, but it should be good to go for testing. I've also separated GetBlobsV4 and the updated fcu.

Kamil Salakhiev: Okay.

Daniel Knopik: for… So that can be.

Kamil Salakhiev: Probably.

Daniel Knopik: Tested independently. I'm not sure, what the exact plan is with regards to FCO before rollout, and if that is… Commonly implemented in all the clients yet, or ready, or whatever.

Kamil Salakhiev: Yeah, if I'm not mistaken, like, the differences, yeah, we have… Fcuv4, it has some changes for Sparse Global Pools, and also changes for… There's no sound. Gas, something like that. Yeah, you're in. yeah, once we have them, yeah, we can, But yeah, actually already Prysm supports… I don't know, in Prysm we have both forkchos updated and GetBlobsV4, right? So we can basically use… I mean, this Bosworth PR contains that, so we can test sparse blob pools properly already. Okay, cool.

Bosul Mun: Yes, but there are some bugs that I will investigate it. For Deb? In Glamsterdam Diagnostics, there's sparse pool merged. And also we have purchase updated v4 and gap loss v4 there. So once any ECL client have it on DevNet, then we can see the impact of.

Aarsh Shah: That's false.

Bosul Mun: But, first of all, and…

Aarsh Shah: So…

Bosul Mun: I'm also curious… I'm also curious, how many EL clients…

Aarsh Shah: Immediately.

Bosul Mun: have a sparsible pool on the DevNet, or…

Aarsh Shah: I don't know.

Bosul Mun: I think anyone from EL…

Aarsh Shah: Busy.

Bosul Mun: here, except…

Aarsh Shah: Yeah, absolutely.

Kamil Salakhiev: Yeah,

Bosul Mun: And I'll, I'll go to Disc.

Kamil Salakhiev: Yeah, I remember,

Aarsh Shah: Israel.

Kamil Salakhiev: They had draft PR for that.

Aarsh Shah: Yes.

Kamil Salakhiev: Right.

Aarsh Shah: A month ago, I took.

Kamil Salakhiev: Checked today, they have… I mean, the PR is not draft anymore, it's just open for review.

Aarsh Shah: Double.

Kamil Salakhiev: So I assume they should ship this in the nearest future? Maybe if we can ping them, they will ship it faster?

Aarsh Shah: Only at the top, so I'm not…

Kamil Salakhiev: Yeah, so at least, yeah, for them, I, for them, yeah, I can confident that, yeah.

Aarsh Shah: Yeah, stupid.

Kamil Salakhiev: In the near future. Not sure about Nethermind?

Aarsh Shah: Okay.

Kamil Salakhiev: Yeah, if we can check with.

Aarsh Shah: Absolutely.

Kamil Salakhiev: Great.

Aarsh Shah: Oh, on which.

Kamil Salakhiev: Network.

Aarsh Shah: I'll be testing sparse prop pool. Do we have a demo for it already?

Kamil Salakhiev: I see. Yeah, both.

Bosul Mun: Not dedicated DevNet, but it's… her guest is merged on DevNet 6, but nobody has… Okay. None of the CEOs have has implemented FCU and GAP Plus before, so. I don't think it's very effective right now. And for me, I… locally tested using Kurtosis with 5-coded prism, but when I revised it and developed, there were some problems, so I'm still investigating.

Aarsh Shah: Okay, do we know how we plan to test the sparse block pool work? Like, let's say once Lighthouse and Prysm both… integrated, and it looks good, do we plan to test Spark's block pool on a dedicated DevNet? or piggyback on some existing testing. And the reason I have this, I want to understand the timeline of When Prysm should merge, The sparse block pool work to its release branch. once I've reviewed it.

Bosul Mun: I think, as… It's faster, better, because… Yes, the blog.net zero, we can use the… So. Launching DevNet is… Not. very, it doesn't have to be very scheduled. Once we have clients ready, we can launch it, so… I would say faster is better.

Aarsh Shah: Okay. Okay. And, Daniel, if I understood correctly, you already finished the get blocks before changes, right?

Daniel Knopik: Yeah, a draft car. Darn.

Aarsh Shah: Okay, so.

Daniel Knopik: That's it.

Bosul Mun: You can also use the lighthouse and launch the devnet with lighthouse get Nethermind. Erigon. Lunch dedicated, yeah. Have that for those 4 clients, and maybe join Prysm later. The reason I worked on Prysm is because I'm most familiar with Go. Since Daniel already tested his branch with GAP, I think it's… good to test.

Kamil Salakhiev: Yep. Okay, yeah, I also checked today before the call. Yeah, I can, in theory, try to run shadow tests, I mean, using ETH shadow with, I mean, now we have like Prysm and Guest, both supporting sparse block pool. So we can maybe simulate 500 nodes approximately. This would take maybe entire night to simulate maybe 10 minutes of simulated time. But if it's useful, yeah, we can give it a try. just to see how it goes. I mean, we have some infrastructure that can Run it. Anyway, I mean, this infrastructure is doing nothing right now, so maybe I'll just leave it for today night, and yeah. We'll see how it works. Okay, yeah, I think it's clear.

Okay, good. I don't know, like, Anton, I see you also on the call. Maybe… do you have anything from Teku's side that you want to share, or it's all good? You're right.

Anton Nashatyrev: To be honest, not much details about Teku, but… I see that they are working hardly on PeerDAS, so they made several fix releases. So yeah, probably PeerDAS is already in production, hope so. So yeah. Yeah, but nothing more from the table side.

Kamil Salakhiev: Yeah, great. Yeah. That was very. Anticipated update. I mean, for quick. Nice. Okay, cool. Yeah, we don't really have any other items in agenda. Yeah, unless anyone wants to bring up anything. And… Okay, nice. Yeah, I think maybe then we can wrap up for today. So thanks for joining. Yeah, next time we'll have some stuff to present. So yeah. There was just some last minute calculations. Anyway. Okay, thanks for joining. See you. Bye.

Anton Nashatyrev: Bye bye.

Bosul Mun: Bye!

Aarsh Shah: Hi, thank you.

Transcript adapted from ethereum/pm, CC BY-SA 3.0.