scammo (@scammo_) header
scammo (@scammo_) profile picture

played the game @S2_FNF | prev: @sphere_labs @heliuslabs

🇰🇷🇺🇸
Joined January 2022
Photo by scammo (@scammo_)Photo by scammo (@scammo_)Photo by scammo (@scammo_)Photo by scammo (@scammo_)Photo by scammo (@scammo_)Photo by scammo (@scammo_)Photo by scammo (@scammo_)Photo by scammo (@scammo_)Photo by scammo (@scammo_)Photo by scammo (@scammo_)
monday too sunday all i eat is sundae
3
6
44
9,129
software engineering is a low status job in korea and that is so confusing/weird to me
515
512
288
19,242
3,282,861
During my exit interview at Citadel, the HR manager asked me why I was leaving. Being young, hot-headed, and eager to do more exciting things with my career, I told her point blank: "I want to build cool shit." I'm lucky to have spent my career doing just that. I built RFQ trading bots, systematic options trading engines, and volatility curve auto-fitters. I built the core smart contracts for compressed NFTs on Solana, powering hundreds of millions of cryptographically controlled digital assets. I built a fully on-chain spot orderbook that has processed over 60B in notional volume. And now I'm building a highly efficient and verifiable financial computer that can atomically execute arbitrary conditional escrows. Looking back, I have some wisdom to share with my younger self and other young coders with similar ambitions: "Cool shit" != business problems Writing sick code will rarely be the primary value add, and logic systems need to impact messy and unpredictable human systems. It turns out that the real "cool shit" (building things that work/make money/gain adoption) is often really boring. Systems thinking > algorithmic thinking Both are important, but having a good grasp of the big picture leads to higher velocity in the long term. And it's way harder. Cleverness reduces complexity If you want to understand a complex system deeply, the easiest way to make progress is through clever (mental) abstractions. When you want to change that complex system, the best approach is to make small extensions to the existing (software) abstractions. Lastly, navigating people systems is harder than engineering ever will be. Building a company has been the most challenging and rewarding thing I have done in my career, and most of the challenges there do not involve writing software. If you are a young and ambitious software engineer who's interested in building cool shit, please consider applying to @ellipsis_labs. I'm confident in saying that our focus is on solving real business problems, but despite that, our tech stack is abnormally interesting if you like thinking about and tinkering with computers. Please shoot me a DM or apply using the link below!
15
20
5
304
29,403
Remember when I shared this almost exactly a month ago? nitter.gq/alessandrod/status/187… It's fixed. For the first time ever, we're actually building blocks in parallel. I've highlighted a block our dev validator produced in _under 110 milliseconds_. Do you know what this means? Do you understand how much faster replay is going to get? Do you see how much capacity we have to increase block sizes now? I'll share a full perf breakdown once we optimize things. I'm confident we can push this below 100ms. (having this much fun should be illegal?!)
Replying to @alessandrod
Next let's look at the threads you probably most care about: the ones that process the user txs that end up in the blocks. At the very start of the first block, all the workers execute transactions in _parallel_. Parallel good. Unfortunately the work quickly becomes _sequential_: for most of the time, solCoWorker04 is the only thread executing transactions, while the other threads _sleep_ and do nothing. Why? This is most likely a bug in the scheduler and the reason why I've been doing this profiling to begin with. I think I know where the bug is, more on this in the next couple of weeks. But so you know what this means? It means that not only we're not close to limits, but we have capacity to process 4x the current load, and we've been working on big improvements to further increase capacity.
59
72
22
283
143,210
Random thoughts now that things have settled a bit (going to regret saying this in 3, 2, 1...) What happened the past few days is amazing, and I'm still buzzing. We did great, despite what some of the usual larpers are saying (more on that below). Agave ingested traffic at peak up to 20x what we usually process, and did it without breaking a sweat. I knew we could handle it from internal benches, but now I got proof with real demand. QUIC, dedup, sigverify, feeding the txs to the scheduler... all worked perfectly and with plenty of capacity to spare. The cluster as a whole did great. The one thing that didn't work perfectly was the scheduler, and that sucks, but schedulers are hard, bugs happen and this was a known bug. Remember the scheduler issue I posted about last month? That's what happened. Same bug, exacerbated by higher load. Bad timing. We've been working on a fix, and also @apfitzge has started working on a new scheduler implementation that seems to do very well with the kind of load mnb receives today. If you'd asked me last week, I would have told you with almost absolute confidence that the probability of seeing this level of demand - sustained for such a long time - in 2025 was close to zero. I would have told you that "decentralized Nasdaq" is a fantastic north star, but not something that affects my daily job too much, because I like to focus on concrete improvements to solve problems we face in the next 6-12 months, not years. I'm glad to admit that I was very wrong. Demand skyrocketed out of nowhere and it will again. We got hit with demand that other chains can only dream of. We did great and are going to do much better next time. I am beyond excited. And yet apparently there are people still claiming that solana dOesN't sCaLe? To those of you with bag bias: whatever nobody cares it's just exhausting to have to keep scrolling past you please stop talking about things you don't really understand. But to those of you that actually work on other chains (this includes some CEOs apparently smh): what is wrong with you? How can you see what happened and conclude that we didn't scale? If we swapped places and you handled an event of this magnitude, I'd be on twitter congratulating you, and on slack yelling at everyone to step up our game. Jfc don't be sore losers, be better.
40
70
27
352
73,317
scammo retweeted
All glory to Christ 🙏
3
1
14
1,672
Job is never finished
Replying to @alessandrod
jobs never finished
14
51
30
794
139,314
scammo retweeted
The best in the world.
34
21
7
322
59,879
scammo retweeted
45/47 🇺🇸
15
7
1
65
29,178
1/ There is a tug of war happening with Solana node catchup times. Agave performance is a big focus right now, but this mean aggressively removing the "slack in the line" startup relies on in order to catch up quickly 🧵
5
21
6
107
22,287
new year, new scheduler? github.com/apfitzge/agave/tr… showing some initial promise in bench-tps. is radically simpler than the current central scheduler. possibly results in less parallel blocks but higher fee collection
9
5
3
86
53,388
Replying to @ruggedwithjack
IMO good traders sell (risk off) as r/r in market subsides Not because of a $ goal you had in your head. Fucks up your vision Playing crypto as a game instead of like its real money
2
4
84
5,118
if you aim to hit 6-7 figure ports for a certainty, NEVER roundtrip. ALWAYS sell aggressively when green. if you aim to hit 6-7 fig pnls, be ready for 10s of the most SOUL-SHATTERING roundtrips for weeks on end. you cant have both at once.
26
64
8
1,103
88,370
Roundtripped 1M+ It's okay. Sometimes you lose, sometimes you win. but take profits when you're giga up.
57
11
2
241
21,307
Most days I wake up and immediately get an aneurysm from someone being dead wrong on the internet. The other day it was from someone claiming that "software scaling has reached its limits". I shared an agave perf profile screenshot, here's how to read it 🧵
19
37
10
164
91,574
I worry about the future generation of engineers over-relying on AI. The ways to become cracked: • Getting your ass kicked by the compiler • Urgently fixing prod bugs you introduced • Writing, reading, and deleting a lot of code The feedback loop is important.
19
10
4
123
17,187
scammo retweeted
These guys are the goats @S2_FNF
8
7
29
9,537
scammo retweeted
S2vember - $8M+
13
11
58
16,273
Hello am korea man, not too good with politics, quick tldr on what's happening and why ur bags are nuking
97
191
55
2,111
686,031
the alfa is out now learn how to use durable nonces for better priority fee discovery, limited execution windows, and one-of-many strategies tempxyz svbstack com /p/durable-nonces
15
11
6
91
28,164