Yilong Li
@yilongl_megaeth
Followers
2,698
Following
6
Media
1
Statuses
74
Building @MegaETH_labs | pre: CS PhD @Stanford
Joined November 2023
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Cheney
• 468785 Tweets
Texas
• 229611 Tweets
Michigan
• 189797 Tweets
England
• 148330 Tweets
Mustafa Kemal
• 133881 Tweets
Ireland
• 123315 Tweets
MSNBC
• 74358 Tweets
Lookman
• 47456 Tweets
Nelly
• 38557 Tweets
Grealish
• 32893 Tweets
Arkansas
• 29646 Tweets
Kentucky
• 29036 Tweets
#كاريزما63
• 24590 Tweets
Iowa
• 24026 Tweets
Musiala
• 21052 Tweets
Ohio State
• 16289 Tweets
Penn State
• 16106 Tweets
Pedro Almodóvar
• 15876 Tweets
#BaşkomutanErdoğan
• 14579 Tweets
Sabalenka
• 14004 Tweets
#الهلاليين_يرفضون_لودي
• 13010 Tweets
León de Oro
• 11538 Tweets
Auburn
• 11521 Tweets
Riya
• 10997 Tweets
Sellers
• 10271 Tweets
Wrong. Many of the MegaETH technologies for scaling L2 don't apply to L1. The only reason we can implement such aggressive optimizations is that we can leverage the base layers (Ethereum & EigenDA) for security and censorship resistance.
If you're looking to build a high-perf
2/18) Ironically, MegaETH aims to scale an EVM-based chain to over 100k TPS!
The irony here is that if we implemented this tech on ETH, we would not need "L2 scaling" at all
That is why MegaETH was a glimmer of hope for me, quickly crushed by the realization of its true purpose
3
4
41
24
28
340
State merkleization is not specific to the EVM; it is a common challenge that affects nearly all blockchains.
While the EVM is not the fastest VM available, its execution speed is never the bottleneck in a blockchain's end-to-end performance.
Saying the EVM won't scale is like
The EVM will never scale, no matter how hard you try. Its design limits execution speeds. Just look at state merkleization and storage layout—you'll be horrified.
If you want to truly advance decentralization, you need to develop a better smart contract layer.
3
3
12
5
12
107
I've been saying no to angel investing, but I just had to make an exception to support Pi^2 :)
I've known
@RosuGrigore
for over a decade. We first met in his CS422 Programming Language Design class at UIUC, and I was instantly captivated by the K-framework. This led me to join
Thrilled to announce our $12.5m in seed funding to build universal verified computing for all!
Thank you to our lead investor
@polychain
as well as
@ABCDELabs
,
@bloccelerate
,
@genventurecap
,
@robotventures
, and
@SamsungNext
for supporting us!
We also want to give a shout out
41
137
404
5
20
97
Did you really read our post? We have an entire section dedicated to explaining why we chose to build as an L2 from the very beginning...
3/18) As the documentation was abundantly clear about this:
MegaETH is intended to scale L2s, not ETH itself!
Even if the technology required to scale ETH is identical...
The cognitive dissonance of scaling L2s like this & refusing to do the same for ETH itself is mindblowing!
1
3
32
9
0
71
bing bing da best
Proud to announce MegaETH's $20M Seed Round
MegaETH is the first Real-Time Blockchain, streaming transactions at 100,000 TPS with millisecond latency
Our mission is to turn cutting-edge ideas into everyday reality with fully onchain, interactive & sophisticated apps
✨Story
226
185
2K
9
1
58
TL;DR: reth is already pretty fast; we are going to make it even faster while reducing its resource requirement🤯 Exciting times ahead for eth/acc!
👇To provide some concrete numbers, reth is already capable of running at 1000-2000 mgas/s during historical sync for recent ethereum blocks, which translates to 10000-20000 TPS (note: these are real-world non-trivial txns, not p2p transfers). 🧵1/4
2
10
40
3
5
28
We are so lucky to have you! LFG🔥🔥🔥
yo bing bing left
@Consensys
and joined
@megaeth_labs
as cofounder
wrote some long ass reflection piece on why i love building in the space
remilio 👉👈
30
25
273
6
2
28
@rphmeier
@megaeth_labs
Hey rob, I don't disagree with any of your comments here. But I think you misunderstand the setup of MegaETH and the points we are trying to make in the quoted paragraph.
Block execution and updating state root (merkleization) are two tasks with very different IO patterns. In
3
2
28
“Play long term games with long term people.” says
@naval
—————————————
At
@eigenlayer
we try to live by this virtue.
And why we have
@megaeth_labs
♾️
@eigenlayer
today!
—————————————
A story.
I got introduced to
@yangl1996
via his PhD advisor Prof. Mohammad Alizadeh at
16
21
229
2
1
15
I like the sound of optimistic full nodes so much that I want to keep it even if/when we switch to zk proof at some point:)
#optimisic
I will be happy to repeat the benefits/importance of having optimistic full nodes (that do not re-execute every transaction) as many times as needed. Such nodes will be the kind that most users care about, because they offer a meaningful trade off between confirmation assurance
3
3
41
0
0
12
@txsequencer
Very weird take. Ofc an L2 node can replay all txns in real time at the throughput of the sequencer. You just cannot do it on a low-end laptop.
1
0
13
What should we call an L2 node that stores a full copy of the blockchain state but relies on external proofs (zk or op) for verifying its correctness?
Fake full node
19
Type-3 full node
9
Optimistic full node
39
Lazy full node
37
4
1
12
gmonad
LIVE NOW -- Mega ETH vs Monad | Super Fast Ethereum
Today we're talking EVM, specifically how
@megaeth_labs
and
@monad_xyz
are planning to make Ethereum SUPER fast. We answer:
1 - Which is faster, more decentralized, and more censorship-resistant?
2 - How did Monad and MegaETH
50
55
235
2
0
11
2
1
10
@0xBalloonLover
Anyone can run reth/erigon/geth on a big box & slap block-stm on revm, right? But how far does that really get you?
megaeth isn't zk moon math, but it's not as trivial as you might think
0
0
8
@txsequencer
somewhere between our fake full node & the sequencer (re-execution is a lot more efficient than the 1st time)
1
0
7
@_weidai
Internally we use validating & non-validating full nodes to refer to these two types of nodes respectively. But I can see it might also be a bit problematic bc "validation" is already overloaded :/
0
0
6
@0xtaetaehoho
~1000 TPS for Reth 0.1.0. Obviously outdated but still in the same ballpark I believe.
Source:
1
0
5
@LoganJastremski
@megaeth_labs
We don't want to run consensus during block processing, but that doesn't mean megaeth will be using a permissioned single sequencer. There are multiple ways to decentralize the sequencer:
1
0
4
@Gary_Rong
Hey Gary, I think you might misunderstand the work. More explanation here:
Appreciate the discussion and the clarification!
Snap Sync is definitely a huge improvement over Fast Sync because it does not have to download the internal trie nodes or latch on to a fixed snapshot. I got nerd-sniped into the problem exactly after learning Snap Sync 😆
The
0
0
9
0
0
4
@ShivanshuMadan
@Madhav_goyal_
@keoneHD
@yangl1996
@BanklessHQ
A membership proof proves that a state variable contains a certain value in the state trie. Nope, it's not enough bc a malicious sequencer can equivocate: i.e. send out incorrect state delta & state root to the p2p network but post the correct data to DA later. That's why L2s
0
0
4
@LoganJastremski
@megaeth_labs
1. Sure, there is always tradeoff in design. But it's not true to conclude that megaeth is just a single permissioned sequencer like Nasdaq.
2. Not sure I follow. Keeping up with the tip of the network is literally the definition of syncing state.
1
0
4
@txsequencer
well it's not like we are running a supercomputer. solana-level validator can get you pretty far
0
0
3
@ShivanshuMadan
@Madhav_goyal_
@keoneHD
@yangl1996
@BanklessHQ
They are membership proofs. The proof verification is quite fast & embarrassingly parallel: e.g., check out the multiproofs benchmark here
1
0
4
@smit__rajput
@hotpot_dao
@chandan1_
@paradigm
@megaeth_labs
@yangl1996
As a PhD at Stanford, I spent most time working on low-latency datacenter computing. So, I do know a thing or two about low-latency systems. Claiming "a massive difference b/w the caliber" is hurting my ego a little bit :) But that's OK; the firedancer team is awesome. We just
2
0
3
@LucasEge
This is not a FUD at all ser. Ofc we want zk to succeed but I am just not sure if it's ready for our use case. Besides, the proof system is kinda orthogonal to the core design of megaeth (modulo security properties, etc.)
1
0
3
@0xalank
@yangl1996
@keoneHD
I mean, c'mon, this is like the basics of how L2s work. How do you think zk provers can keep up w/ the sequencer if proof generation incurs orders of magnitude higher overheads vs. native execution.
2
0
3
@LoganJastremski
@megaeth_labs
1. You don't need to run consensus for every block. See Vitalik's post above for other ways to decentralize the sequencer.
2. Data propagation is a big challenge with or without consensus because full nodes need to sync their states with the sequencer.
1
0
2
@IgMosqueira
@IanSNorden
@keoneHD
@yangl1996
@BanklessHQ
@TrustlessState
@RyanSAdams
@hotpot_dao
I think many of your points are very fair! There is def no such thing as the optimal blockchain solution. Anyway, big fan of aptos' work
0
0
2
@ShivanshuMadan
@Madhav_goyal_
@keoneHD
@yangl1996
@BanklessHQ
Watcher nodes are tiny. They use the stateless validation scheme:
1
0
2
@yajinzhou
@megaeth_labs
@monad_xyz
I suppose a different question would be: if slow VMs or storage were the only factors holding us back, why haven't people already solved the blockchain scalability problem?
1
0
2
@ShivanshuMadan
@_weidai
I don't think it's that different from other L2s. The source of truth is always the data in the DA layer. However, most full nodes probably want to work on the data from the p2p network first and then check against those in the DA later.
1
0
1
@schenkty
@varun_mathur
@yangl1996
How is that misleading? We are not advertising 1ms (hard) finality.
1
0
0
@0xalank
@yangl1996
@keoneHD
"Statelessness" means a single verifier/watcher doesn't need to re-execute all the blocks. And they can do so out-of-order.
1
0
1
@LucasEge
@keoneHD
@yangl1996
@BanklessHQ
These are called verifiers/watchers/etc. They use a stateless validation scheme and have very low hw req (at least for OP):
0
0
1
@schenkty
@varun_mathur
@yangl1996
We don't run consensus during block processing, but that doesn't mean megaeth will be using a permissioned single sequencer. There are multiple ways to decentralize the sequencer:
1
0
1
@schenkty
@MrPkar
@yangl1996
The results given out by the sequencer are (potentially slashable) pre-confirmations. This applies to all L2s.
1
0
0
0
0
1