Glamsterdam Repricings #8

Transcript

Stefan Starflinger: Hey, Zoom is working!

Spencer Taylor-Brown: Hey, guys.

Stefan Starflinger: Hey, how are you doing?

Spencer Taylor-Brown: Doing well here. We've had, like, the first dose of sun in Edinburgh. Of the year, so… Yeah, it's been good. I've been enjoying it.

Stefan Starflinger: dice.

Spencer Taylor-Brown: A little bit, you know, it was fun. It feels a bit like a fever dream, though.

Stefan Starflinger: Yeah.

Spencer Taylor-Brown: A lot of youth.

Stefan Starflinger: I mean, it's 30 degrees, so… I could do with a little bit of a cool… cool down. We don't really have air conditioning in Europe. That much, so I don't have any, so I'm a little bit fucked.

Maria Silva: Hello, hello. Sorry, I joined, and then I couldn't put my earphones to work, but yeah. How is everyone? I understood you're talking about the good weather we have in Europe at the moment.

Stefan Starflinger: Yes, the much too good weather.

Maria Silva: Sorry.

Stefan Starflinger: It's, 30, 30 plus degrees.

Louis Tsai: 30? 36th degree? It is 35 here. It's so.

Stefan Starflinger: Yeah, that's even worse. Okay, I'm not complaining anymore.

Maria Silva: Doesn't seem like we have a lot of attendance today, maybe we can give it a few more minutes to see if more people join, but I think also this would be maybe a shorter call.

Stefan Starflinger: Yep.

Maria Silva: Hi. Drugan, how are you today?

Dragan Rakita: Hello. I heard you complain about better. I have air conditioning, so I'm good.

Stefan Starflinger: Thank you for this information.

Dragan Rakita: Yeah. It's very vital, to be honest, at least for me. Vital.

Stefan Starflinger: Yeah, at some point, your brain just slows down. And, without air conditioning. Oh, great.

Dragan Rakita: Yeah.

Stefan Starflinger: My, my memory.

Maria Silva: Okay, cool. I think we are 3 minutes in, no more people are joining, so I guess we can just kickstart. So the agenda, I have, bal-devnet-7 updates, then maybe we could have, like, a discussion on the open issues for 8037 and what we do there. Then I'll give a quick, benchmarking update for, 7904 and the ETH Transfers EIP, so, 2780. Yeah, and then I wanted the Tienne to kind of discuss, moving the discussions to ACDT, because I feel like 8037 is mostly stable, and the next CIP is a… Once they are included in the DevNet, I think they'll be… Simple enough… simple enough that we can just have the discussions on ACDT, But yeah, let's, let's start. So for bal-devnet-7,

Stefan, I don't know if you want to give a quick update, or Spencer.

Stefan Starflinger: Yeah, I haven't actually had that much time to look into things, I've mostly been doing benchmarking. But everything has been working pretty well, so I wasn't pinged, really about any issues to look into, because I mostly have a monitor for any consensus-related stuff, but I think we've been having 100% participation with EVM fuzzing for quite some time now, which is really good news, so I'm very happy about that. I think there's only some small update from Spencer regarding the new release of the test, and I have only one question there. Which would be, do we need to, or can we just roll it out on the current bal-devnet-7? Like, I wouldn't mind if the bal-devnet-7 breaks or something.

But I'm not too sure, like, what the chances are. Like, if it's 100% that while bal-devnet-7 will break, then it's not really worth it, then we could just move it to Glamsterdam devnet. But if it's, like, maybe 20%, I think we should just risk it.

Spencer Taylor-Brown: Yeah, so it's just, extended coverage, the next release. and a… Spec change, for… Let me, let me link the PR. And I… I don't think this… This would break? Or maybe it would… it's hard.

Stefan Starflinger: What is the chance we fit it with EVM fuzzing?

Spencer Taylor-Brown: To be honest, I need to check if… what… if… what clients are doing. Of what… if they're actually aligning with this check, if they're aligning with eels, or if they're doing… a different check already. but yeah, I think… Truly. I can check whether… I can check whether this change would break the DevNet, based on what clients are doing now. if it would break the DevNet, then I won't include this in the next release, and I'll… we'll include this in the Glamsterdam devnet 5 release instead. If it doesn't or I won't break it, I will, I will input it. But yeah, like, the TLDR is the… The next release will mostly just be… well, it's almost entirely test… test coverage.

Yeah, it's nice to not have a breaking… Devnet with 1837. So… Or a stable one, maybe that's the phrase.

Maria Silva: But so… but so you are saying that it could make a sense to just move one of those changes to the next, Glamsterdam devnet? Is that, what, Stefan, you are… you are proposing?

Stefan Starflinger: I'm proposing to just try to include it in bal-devnet-7. I looked into it a bit, but I'm not too certain if it will break it or not. But I think, like, we can risk it. I don't think we need to keep bal-devnet-7 around for too long. I think it would be interesting. I mean, a client can make the changes, and then try to sync to bal-devnet-7, and then they should see if it works anyways. I don't think it's such a big risk, and if it's fine, then I think we just… Keep it there, otherwise, we put it to the Glamsterdam devnet. I think it's fine to leave it in the release, I don't think we have to exclude it from the release.

Dragan Rakita: This depends, basically, the Geth has more legs. check while, red, and regulating here more strict. That means… the block built by RET is going to be accepted by Geth, But the block built by Geth is going to be rejected by Red. So it's probably going to… Yeah, going to break networking, depending on amount of the clients that accept that block. It could potentially be splitted. or invalidates some. If there is a lot of clients that has stricter, check. then the get block is going to be just trolled. But if there is more clients that are more relaxed, then… Yeah, a lot of other clients are going to basically be stuck.

I would move it just to not risk it, move it to, like, Glamsterdam devnet 5.

Maria Silva: Also, I would also add that we are relying now on bal-devnet-7 for benchmarking, so I would also like to keep that as stable as possible for now. I mean, at this point, the compute repricing 7904 is kind of, like, relying on us getting measurements for Geth on bal-devnet-7, and they are still merging their optimizations there. So yeah, I think I'll… it's… I would maybe lean more towards Having things not breaking, rather than… Chancing it. Does… does that sound good, Spencer? Is it okay to move it to… Glamsterdam Devon at 5.

Spencer Taylor-Brown: Yeah, no, perfect, sounds good. We'll do that.

Maria Silva: Okay, anything else on bal-devnet-7?

Stefan Starflinger: I don't think so, other than the benchmarking effort, I would, say. Right.

Maria Silva: I can… I can give an update there, after. But first, I just wanted to… I just wanted to discuss, then, Glamsterdam DevNet 5, and there were a couple issues on 8037. Yeah, I was expecting to have a wider… attendance in this call, but I guess most of the people that are here already… have already commented on these two changes, so I'm not sure if there's anything else to add here. But I… I would say, like, the two big open questions in my mind are… In this EIP, so I'll… in this, PR to the EIP, this idea of the clo… the call data flow accounting. Yeah, I feel that the current, accounting we are doing in the specs and the EIP is not the most… it's not the most correct one, and this change is…

more correct. And also, it clarifies how it should work after 7778. Which I think now is kind of, like, hand-wavy on how it… supposed to work. So I think this is an important change, and then the second one is the… is mostly, clarification, so it shouldn't change the, specs, it's mostly just, adding more, more ink through there. I think then there's this, PR from Gary that, essentially, clarifies, or extends the refunds we provide for 7702 authorizations. I think this is more in line with the spirit of the EIP. Where you are trying to give as much refunds as possible, or you're trying to align as much state GAS with state creation. But it is changing how… because it's giving additional refunds for authorizations that we didn't have before. So also.

Curious on what you think about this one. But yeah, do we have any comments on these two… these two changes?

Dragan Rakita: I think the code on the first PR needs to be changed. I think we had discussion in the Discord about how that needs to go. But yeah, floor… removing floor gas from the block-related gas variable makes sense. And on second period, it does make sense. we just need to figure… I need still to look at the PR, but it makes sense to, like, refund the state if the authorization Already happened. There's some edge cases, I think, but either way, it shouldn't be that hard to support.

Maria Silva: Right. Any more comments on these?

Spencer Taylor-Brown: I think for the… 7778… alignment. Yeah, I didn't really understand it at first, but I think, you guys clarified in the Discord, so I'm happy with that change for the… For Gary's change, the 7702, Personally, I wouldn't want to add it, because I think it will add more edge cases, but then it is more in line with the EIP, so…

Maria Silva: Right.

Spencer Taylor-Brown: Yeah, I think, I'm happy with both, to be honest, and maybe… maybe it's worth paying to ACDT. and… I think it would be good to include that when aligning on, I guess what EIPs are formally included in Glamsterdam devnet 5? And to be honest, I think we should… just have everything in the next DevNet.

Maria Silva: Right, so I think that's… that's a good point. So, it seems like the call data floor is a bit more… obvious that we should do it, because it's kind of like, fixing something that is not fully correct. While Gary's PR is most… is more like your PR extending functionality in a way, so yeah, maybe it's a good thing to kind of, like, discuss with a wider audience, I agree. So yeah, maybe it's just I can… likely just merge my PR for now after I do the update, and then we'll… we'll take Gary's PR for… for discussion. Does that sound good, or do we have any, issues with this. Okay, anything else to discuss on open issues for 8037?

Okay. So maybe I can give a quick update on benchmarking. So let me see if I can find it. So I have a sort of, like, short report with the key takeaways of where the numbers… are right now, so I'll just post it here for everyone. Let me get it, but the TLDR is, maybe I can even share my screen, but the TLDR is essentially We ran our analysis on bal-devnet-7. And we did it for 7904, so the compute repricing, and 8038, the state access repricing. And there's, two important things. So, first is, for 7904, we are still waiting for Geth to merge their optimizations into this branch, and so because of that, they're… they are still performing build

worse than the… than the… the anchor for the CIP, which is the 100 million gas per second anchor. But besides Geth, all other clients, and this includes Nethermind, which was one of the clients that had the biggest, change in codebase between 80… between bal-devnet-3 and bal-devnet-7. So for… for Nethermind, we have confirmed that they are performing better than the anchor on all opcodes, and the same for all the other clients. So the only thing missing now is for Geth to merge their optimizations, which we should expect, these are next week, and then we can start, measuring them, and then Close off, these… these analyses, for sure. And so it's looking like, again, still likely that 7904 won't… won't be included, or…

I think at the end, it will mostly be turned into an informational EIP, where we say we did the analysis, all operations were performing better than we needed them to perform to raise the gas limit, and so this is, like, a no-op EIP. There's no… there's no change needed, but we still, like, record the analysis and the takeaways. Then for 8038, we are still, seeing some improvements, but, they are not, yet, good enough, so especially for Erigon, they are still performing, poorly, and they are aware and working on optimizations. Then, for Reth, we have an issue on the how we are kind of, like, benchmarking them, and Rafael is coming, from, out of office today, and he will start checking that and see, because right now the performance we are seeing in our benchmarks is worse than what we would expect based on their tests.

For Besu, they are still working on optimizations, and they should start to land between these and next So yeah, so it's still… A lot of, things still happening, there's still some optimizations to be done, but I think we are starting to get there. And then we started also to run the analysis for 2780, so this is the… it transfers EIP, and the… the context here was, the… The initial, essentially the initial EIP, had this flavor where, let's break down the cost of an ETH transfer into its components, and then we price them… we price the components instead of having, like, this, generalized price. But when we do that, with the current gas costs for things like estate access, estate rights, we actually, in ETH transfer.

goes about… goes over the 21K, which, it's not feasible. And so what we are doing now is, like, we will… we started to benchmark the ETH transfers with, so this is, like, a stateful benchmark, where we take into consideration mainnet size for the account tree, and we are starting to see some results, but results are not reliable enough, so I think there's still some work to be done here. But yeah, so it's… it's a work in progress, still not finalized, and we still also expect some optimizations that we are observing from the State Access EIP to also help with 2780. So… Then, what does this mean? So, essentially. I think we are still in the same place as we were last time in terms of final numbers, so we can probably include 2780 and 8038 in Glamsterdam that

Glamsterdam devnet 5, but they won't be with final numbers, and I think between the next two weeks, I think we'll start having more, stable numbers, and maybe we'll even start having, like, a proposed set of values for both of these EIPs. But yeah, so this is the… the current status. I don't know if there are any comments or questions on these.

Dragan Rakita: That sounds fine. I think we are on the, like… it's not related to the gas repricing and everything around that. It's more about what's left. I think we… have additional 2 DevNets, if everything goes fine, and we can ship Glamsterdam one after that. I think the EIP-2780, reduced intrinsistic transaction gas, I would like to tweak it a little bit. But other EIPs and changes, like, look… Perfectly okay. So, there is not a lot to… to work on. That would be my comment.

Maria Silva: Great. Yeah, I think since we have time, what would you tweak in 2780? Is it, the simplification on the cases?

Dragan Rakita: Yeah, I would simplify it even more. Basically, I would remove two things. Remove the case where we take into account if the target is precompiled. Remove the case when the transfer is to itself. Just, yeah, consider it as… don't even track dead cases. Additionally, I need to check this, but the warming around the transaction target seems unnecessary. I would consider both the… target and delegation target as the cold, warm, cold loaded, and just marketed that. Yeah, I think that would be it. I thought maybe it's… to have… I thought about the idea to have, like. Let's take the higher maximum gas, then… Direct refund.

If, like, account is present, account is… Delegation is not present, or something like that, but… Yeah, this, this, this feels, like, unnecessary. But either way, some kind of simplify… basic removing those kind of edge cases could basically simplify the IP.

Maria Silva: Right. Yeah, I think it's the… I agree that there's some simplification we can do there. I think it would be clear what simplifications we should do when we have more stable results on benchmarking, because then it could say, like, depending on some cases, it could make a really difference in terms of prices, if we keep the edge case versus just price on the worst case. And if that's, like, a… like, for the precompiles, for… for instance, I think it's, I mean, I'm okay with removing it, but it could be, like, a case where actually a lot of people will use it, and it would be a meaningful decrease in gas costs, then it… there could be an argument that the complexities is warranted.

So yeah, I agree in general that it would be nice to kind of simplify it, but also, let's do it once we have a clearer view on the final numbers as well.

Dragan Rakita: I agree.

Maria Silva: Cool, anything else here? Maybe a question from me, do we have a sense for the timeline for Glamsterdam devnet 5?

Stefan Starflinger: I think we will just… I mean… probably, we will set a time on the ACDT, on Monday, and the scope. And I would assume, The week… So, in 2 weeks? Something like that.

Maria Silva: Okay.

Stefan Starflinger: Maybe earlier, if I would ask Barnabas.

Maria Silva: Okay, perfect, thank you. Yeah, just, just, just curious, Okay, then to wrap up, I was actually… and I'm… I think I will talk with Barnabas about this, but I was thinking about, stopping the breakouts, because I feel like more and more we don't have as many things to… to kind of discuss, or are mostly doing updates. So I think it would be a good time to kind of, like, stop doing these breakouts and moving discussions entirely to ACDT. Would you guys agree with this, more generally? Yep.

Stefan Starflinger: Sounds good.

Maria Silva: Okay, so let's do it. I'll ping Barnabas and see. Yeah, are there any more topics that, anyone wants to bring up, or should we wrap it up? Okay, then we can wrap up. Thank you all for coming. Speak soon.

Dragan Rakita: Right.

Spencer Taylor-Brown: See you guys. Bye-bye.

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