AllCoreDevs - Testing #090
Transcript
- jochem-brouwer
gm 😃
- emma (reth)
gm
- Yassine Ferhane
gm
- nixo
gm
- Justin Traglia
Gm
- Justin Florentine (Besu)
gee em
- Toni Wahrstätter
Reacted to gm 😃 with "🙏"
- Toni Wahrstätter
Reacted to gm with "🙏"
- spencer
gm!
- Toni Wahrstätter
Reacted to gm with "🙏"
- Toni Wahrstätter
Reacted to gm with "🙏"
- Toni Wahrstätter
Reacted to Gm with "🙏"
- Toni Wahrstätter
Reacted to gee em with "🙏"
- Toni Wahrstätter
Reacted to gm! with "🙏"
- Pooja Ranjan
Replying to "gee em" Audio version please 🙂
- Justin Traglia
- spencer
Whats a stable devnet?
- Parithosh Jayanthi
Its a new thing we’re trying out
- spencer
Replying to "Its a new thing we..." Sry I was joking! But also a nice question to answer 😆
- spencer
Reacted to "Its a new thing we..." with 😅
- nixo
Reacted to "Its a new thing we’re trying out" with 😂
- Mario Vega
Reacted to "Its a new thing we’re trying out" with 😂
- Chris Berry
Reacted to "Its a new thing we’r..." with 😂
- Parithosh Jayanthi
Reacted to "Sry I was joking! But also a nice question to answer 😆" with 😂
- Stefan Starflinger
Reacted to "Its a new thing we’r..." with 😂
- potuz
Reacted to "Its a new thing we’re trying out" with 😂
- Toni Wahrstätter
Reacted to Its a new thing we’r... with "😂"
- Justin Traglia
To prevent caching, nice.
- FLCL (Nethermind)
Reacted to "Its a new thing we’re trying out" with 😂
- Justin Traglia
Reacted to "Its a new thing we’r..." with 😂
- Justin Traglia
Reacted to "Sry I was joking! Bu..." with 😅
- Nico Flaig
Reacted to "Its a new thing we’r..." with 😂
- Barnabas
We should make a decision by EOD.
- Ben Adams
Is StateActor using FlatDb for Nethermind now?
- Louis
Replying to "Is StateActor using ..." I think so. Cc @jochem-brouwer
- Marius Van Der Wijden (M)
Sad that we have to move away from the 100mgas anchor, but great that we can move on and ship a substantial improvement
- potuz
If clients manage 75 yes
- Nico Flaig
can we reduce attestation deadline to 2 seconds?
- potuz
I would love this ^ cc @carlos
- Edgar | ethrex
we will look into fixing cold
- Łukasz Rozmej
as we are discussing here "worst case bottleneck scenarios" and not normal blocks - do we have to have 100% of clients always below the limit? I think if we have 75% of the network than we are fine IMO. And typical blocks are usually faster.
- Maria Silva
These are worst case blocks
- Łukasz Rozmej
to be honest I don't understand how Reth has such good numbers for state access kind of looks like something is off with benchmarking
- Justin Florentine (Besu)
Replying to "as we are discussing..." i think you're suggesting we can predict market share
- Ameziane Hamlat
My understanding is that our current repricing doesn’t take into account ePBS. They are based on the 3 seconds execution window. So 75 Mgas/s anchor point targets 225 Mgas gas limit without ePBS
- draganrakita
Replying to "to be honest I don't..." Probably magic
- Csaba
What’s the adversary model behind these worst cases?
- Chris Berry
Reacted to "Probably magic" with 😂
- Yassine Ferhane
Reacted to "Probably magic" with 😂
- —
Łukasz Rozmej: Replying to "to be honest I don't understand how Reth has such good numbers for state access kind of looks like something is off with benchmarking" yeah looks like it
- Ben Adams
Worse case on massive state
- draganrakita
Replying to "to be honest I don't..." What to say, it works
- Csaba
Replying to "What’s the adversary..." I mean, is it a continuous stream of worst case blocks being generated by builders? Or is it a single one?
- Toni Wahrstätter
75 mgas with 300M block gas limit would still be a quite conservative and safe choice, leaving enough buffer and a path torwards 450m.
- Ameziane Hamlat
Reacted to "75 mgas with 300M block gas limit would still be a quite conservative and safe choice, leaving enough buffer and a path torwards 450 mgas. " with 👍
- Nico Flaig
isn't it 9 seconds?
- Ameziane Hamlat
Replying to "75 mgas with 300M bl..." 75 Mgas/s 🙂
- potuz
I very much like these numbers if we do manage to get to 75 on clients
- Barnabas
Guys, can we just agree on 200M gas for glamsterdam, and do the fork, then move on from there? We will have months to evaluate what is exactly safe, already going from 60 to 200 is a huge lift.
- jochem-brouwer
Thanks for this discussion. We have to realize that if the mainnet blocks as of now do not hit these limits, we know that if this is exploitable (and profitable) in the future, then this will be attacked. We have the vision on this now so this is why we target worst case, not avg case
- potuz
Replying to "isn't it 9 seconds?" What is 9 seconds?
- draganrakita
Reacted to "Guys, can we just agree on 200M gas for glamsterdam, and do the fork, then move on from there? We will have months to evaluate what is exactly safe, already going from 60 to 200 is a huge lift." with ➕
- milen | Erigon
Reacted to "Guys, can we just agree on 200M gas for glamsterdam, and do the fork, then move on from there? We will have months to evaluate what is exactly safe, already going from 60 to 200 is a huge lift." with ➕
- spencer
Reacted to "Guys, can we just ..." with ➕
- Chris Berry
Reacted to "Guys, can we just ag..." with ➕
- Nico Flaig
Replying to "isn't it 9 seconds?" payload has to be available at 6 seconds in prev. slot, and attester need it executed at 3 seconds in next slot?
- jochem-brouwer
Heeft gereageerd op "Guys, can we just ..." met ➕
- Justin Traglia
Reacted to "Guys, can we just ag..." with ➕
- Parithosh Jayanthi
Reacted to "Guys, can we just agree on 200M gas for glamsterdam, and do the fork, then move on from there? We will have months to evaluate what is exactly safe, already going from 60 to 200 is a huge lift." with ➕
- Nico Flaig
Reacted to "Guys, can we just ag..." with ➕
- Maria Silva
Replying to "isn't it 9 seconds?" The proposer needs to finish by the start of the next slot
- Ben Adams
Reacted to "Guys, can we just agree on 200M gas for glamsterdam, and do the fork, then move on from there? We will have months to evaluate what is exactly safe, already going from 60 to 200 is a huge lift." with ➕
- Nico Flaig
Replying to "isn't it 9 seconds?" hmm yeah kinda true was assuming proposer might be fine but attesters vote for parent
- Toni Wahrstätter
Yeah, glam with 200, check how things play out, and then go for 300 before hegota (assuming things are fine) sounds good
- draganrakita
Reacted to "Yeah, glam with 200, check how things play out, and then go for 300 before hegota (assuming things are fine) sounds good" with ➕
- jochem-brouwer
Heeft gereageerd op "Yeah, glam with 20..." met ➕
- spencer
Reacted to "Yeah, glam with 20..." with ➕
- nixo
Reacted to "Yeah, glam with 200, check how things play out, and then go for 300 before hegota (assuming things are fine) sounds good" with ➕
- Stefan Starflinger
Reacted to "Yeah, glam with 200,..." with ➕
- Simon (@siladu) | Besu
Replying to "What’s the adversary..." I believe it’s a worst case single block, with minimal/no caching…but maybe someone else can confirm
- jochem-brouwer
Does this have the new numbers wrt access lists?
- potuz
I think the suggestion is to fix the relative costs right now as a decision, and then later decide if we target 75 or 50
- Marius Van Der Wijden (M)
Replying to "What’s the adversary..." Yes, single full (1559 max) block
- Toni Wahrstätter
Replying to "What’s the adversary..." Yeah, we would have 6s max for execution. The worst-case block must never take longer than 6s to execute
- potuz
At any rate we can right now anchor the 6 seconds on PTC and we can move to attestation deadlines
- Edgar | ethrex
good for us too
- draganrakita
Ask for someone else from geth team 😄 just to confirm
- Justin Traglia
Reacted to "Ask for someone else..." with 👍
- Ben Adams
I'd like account_write lower; but it is what it is 🤷♂️
- Maria Silva
Reacted to "I'd like account_wri..." with 🥲
- FLCL (Nethermind)
does it affect access lists?
- Justin Traglia
Replying to "good for us too" Great thanks 🙏
- Maria Silva
Replying to "does it affect acces..." Yes, it will. As they are a transformation from cold_storage_access
- Ben Adams
Replying to "does it affect access lists?" we made access lists cost more I think?
- FLCL (Nethermind)
Reacted to "Yes, it will. As they are a transformation from cold_storage_access" with 👍
- FLCL (Nethermind)
Reacted to "we made access lists cost more I think?" with 👍
- Ben Adams
Replying to "does it affect access lists?" Since they are mostly pointless
- Justin Traglia
Erigon & Reth, do you support these changes?
- draganrakita
Reacted to "Erigon & Reth, do you support these changes?" with 👍
- milen | Erigon
Reacted to "Erigon & Reth, do you support these changes?" with 👍
- draganrakita
Replying to "Erigon & Reth, do yo..." Yes we do
- jochem-brouwer
Antwoord verzenden naar "does it affect acc..." I think idea is to price access lists same as if you would do normal storage load (so there is no incentive anymore to use access lists in tx)
- milen | Erigon
Replying to "Erigon & Reth, do yo..." Yes for Erigon. Similar feedback to Besu
- Ben Adams
Replying to "does it affect access lists?" Is double block data cost (in both tx access list and block access list)
- draganrakita
In general if something is not correct, mitigation is lowering block gas, this does not require hardfork but validators to change how they run client.
- Justin Traglia
FYI, Csaba, I moved your discussion item to the EL breakout section. We’ll start the breakouts after this concludes.
- jochem-brouwer
Good point Ben :)
- Csaba
Reacted to "FYI, Csaba, I moved your discussion item to the EL breakout section. We’ll start the breakouts after this concludes." with 👍
- Mikhail Kalinin
Where to see numbers on this?
- Justin Traglia
Replying to "Where to see numbers..." I would also like to see numbers.
- pk910
Replying to "Where to see numbers..." We should not test weak stuff on hoodi. Yea, it's a testnet, but it's a stable testnet for the comunity, that we promised to not break as there's a lot of work involved to rebuild
- Nico Flaig
@Enrico Del Fante (tbenr) can't you prepare the block in prev. slot already?
- Carlos
https://ethresear.ch/t/is-the-slot-0-reorg-cost-fixable-epbs-attestation-deadline-study/25338
- Enrico Del Fante (tbenr)
Replying to "@Enrico Del Fante (t..." We anticipate some of the work but not all the work on attestations
- Justin Traglia
Reacted to "This is the analysis..." with 🙏
- Carlos
https://ethresear.ch/t/is-the-slot-0-reorg-cost-fixable-epbs-attestation-deadline-study/25338" with 👍
- Enrico Del Fante (tbenr)
Replying to "@Enrico Del Fante (t..." We try to use all possible attestations we receive
- Nico Flaig
lodestar has similar problem, we have worst cases of like ~800ms to build in the beacon block in some cases
- Carlos
https://ethresear.ch/t/is-the-slot-0-reorg-cost-fixable-epbs-attestation-deadline-study/25338" with 🙏
- Carlos
https://ethresear.ch/t/is-the-slot-0-reorg-cost-fixable-epbs-attestation-deadline-study/25338" with 🙏
- Enrico Del Fante (tbenr)
Replying to "@Enrico Del Fante (t..." Ye that’s why I implemented a time limit on attestation preparation
- Mikhail Kalinin
As long as there are no equivocations
- potuz
yes
- Toni Wahrstätter
Will come up again with discussions around shorter slots and we'll need more numbers / analysis anyway
- potuz
Reacted to "Will come up again with discussions around shorter slots and we'll need more numbers / analysis anyway" with 👍
- Justin Traglia
https://ethereumfoundation.zoom.us/j/89441658268?pwd=yjqf4TheEb1R7ZSuesCdOAy5tdyVk7.1
- Parithosh Jayanthi
Reacted to "Will come up again with discussions around shorter slots and we'll need more numbers / analysis anyway" with 👍
- Enrico Del Fante (tbenr)
Replying to "@Enrico Del Fante (t..." But it can still take some time, we can easily reach 500ms
- FLCL (Nethermind)
discv5-only is cool, let's go. What can be the plan/timeline for mainnet?
- Stefan Starflinger
Reacted to "discv5-only is cool,..." with 🛳️
- draganrakita
Reacted to "discv5-only is cool, let's go. What can be the plan for mainnet?" with 🛳️
- spencer
Reacted to "discv5-only is coo..." with 🛳️
- Edgar | ethrex
Se ha reaccionado a "discv5-only is coo..." con 🛳️
- Mario Vega
Reacted to "discv5-only is cool, let's go. What can be the plan/timeline for mainnet?" with 🛳️
- spencer
Reacted to "discv5-only is coo..." with 🥏
- spencer
Reacted to "discv5-only is coo..." with ✌️
- spencer
Reacted to "discv5-only is coo..." with 5️⃣
- Csaba
Reacted to "discv5-only is cool, let's go. What can be the plan/timeline for mainnet?" with 🛳️
- Marius Van Der Wijden (M)
Isn't this just snap/2?
- jochem-brouwer
- jochem-brouwer
- Marc
Reacted to "discv5-only is cool, let's go. What can be the plan/timeline for mainnet?" with 🛳️
- FLCL (Nethermind)
tx pool warmup may be more efficient and evict such warm items
- —
Łukasz Rozmej: Replying to "to be honest I don't understand how Reth has such good numbers for state access kind of looks like something is off with benchmarking" are you delaying writes beyond benchmark?
- jochem-brouwer
Antwoord verzenden naar "tx pool warmup may..." yup, good point!
- Karim T. (matkt)
I see also that as a last step after snap in order to maybe move all this data in L0 as you said
- jochem-brouwer
exactly, it is not critical
- Mario Vega
Reacted to "exactly, it is not critical" with 👍🏼
- jochem-brouwer
its a nice to have to hot-boot syncing node
- spencer
https://github.com/ethereum/pm/issues/2174#issuecomment-5167784043
- jochem-brouwer
Maybe bosul in cl breakout?
- emma (reth)
Replying to "to be honest I don't..." 🪄
- jochem-brouwer
Lets agenda this for ACDC after check with Bosul
- Mario Vega
Reacted to "Lets agenda this for ACDC after check with Bosul" with 👍🏼
- FLCL (Nethermind)
this would speedup sparse blob testing
Call summary
Targets
- •Glamsterdam devnet 8: short-lived named testnet, dependent on repricing finalization this week - 00:06:16
Decisions
- •Repricing numbers frozen: SLOAD stays at 2,100; account writes set to 9K; anchor fixed at 75 Mgas/s; all EL clients approved - 00:34:15
- •Attestation deadline stays at 3s for Glamsterdam fork; 2s lowering deferred post-fork via coordinated client release - 00:44:00
- •discv5-only (no discv4 fallback) adopted for devnet 8 and targeted for mainnet - 00:45:24
Highlights
- Repricing Updates:
- ·New anchor lowered from 100 to 75 Mgas/s due to costlier account test cases; 21K ETH transfer is hard ceiling - 00:07:21
- ·EIP-8038: account writes increase 8K→9K; SLOAD stays at 2,100 (no increase from current); ETransfers EIP unchanged - 00:11:09
- ·At 75 Mgas/s anchor: Erigon still red/yellow (optimizations pending); ethrex has one cold-access gap; Besu/Geth at edge but expected to improve - 00:13:31
- ·At 75 Mgas/s, theoretical max block limit ~450M gas; 300M considered safely reachable if all clients reach green - 00:18:49
- El Breakout Discv5:
- ·discv5-only mode confirmed for devnet 8; all clients passing on at least one branch; fallback must be disabled - 00:45:24
- ·Two clients lacked a flag to disable discv4 fallback (Nimbus EL confirmed); teams asked to add the flag before devnet 8 - 00:46:36
- El Breakout Eip 7702:
- ·EIP-7702 engine API mandatory endpoints raised; Geth and Nethermind passing Hive simulator; concerns about FCU placement unresolved - 00:59:57
- ·Action: check with potuz on spec changes, then agenda EIP-7702 engine API discussion for ACDC - 01:01:56
- Fork Status And Schedule:
- ·Glamsterdam devnet 7 stable for 20 days; 90%+ participation, minor disk/node issues - 00:03:47
- ·Glamsterdam devnet 8 targets new CSpec release; should be short-lived named testnet with updated repricing - 00:06:16
- Epbs Attestation Deadline:
- ·ePBS makes worst-case slow blocks an automatic reorg attack; worst-case sizing must be respected, not average - 00:23:22
- ·Attestation deadline lowering to 2s: main blocker was one operator combining high min-bid with timing games on Prysm - 00:37:00
- ·Decision: keep attestation deadline at 3s for fork; gather hard numbers post-fork, revisit lowering to 2s without a hard fork - 00:44:00
- El Breakout Block Access Lists:
- ·Block Access Lists (EIP-8189) enable hot-booting SnapSync: pre-loading last N blocks' access items into cache cuts cold-read penalty - 00:49:50
- ·Snap2 healing phase can replace trie healing with BAL execution; ~79% of first-touch items touched in prior 256 blocks - 00:53:25
- ·Consensus: BAL hot-boot is a nice-to-have post-release optimization; does not affect worst-case benchmark numbers - 00:58:02
Action Items
- •Maria Silva / SEAL - Update EIP-8038 repricing numbers; publish new execution-specs release - 00:34:22
- •EthPandaOps - EthPandaOps: gather hard block-propagation and state-transition timing numbers at fork for 2s attestation deadline analysis - 00:43:02
- •jochem-brouwer / Mario Vega - Confirm EIP-7702 engine API spec changes with potuz; bring to ACDC agenda - 01:01:56
Key decisions
Attestation deadline stays at 3s for Glamsterdam fork; 2s lowering deferred post-fork via coordinated client release
PFIDeferred 2s lowering post-fork via coordinated client releaseRepricing numbers frozen: SLOAD stays at 2,100; account writes set to 9K; anchor fixed at 75 Mgas/s; all EL clients approved
IncludedRepricing numbers finalizeddiscv5-only (no discv4 fallback) adopted for devnet 8 and targeted for mainnet
devnet 8discv5-only mode confirmed for devnet 8