Thread by David Fauchier

twitter.com
1 Recommender
1 Mention
Recommend
Post
Save
Complete
Collect
Thread
There are many debates being had about this but I want to focus on one thing — FTX did not have a nice matching engine! In fact it had a very bad and laggy matching engine with latencies orders of magnitudes worse than their competitors.

It was almost certainly implemented in Python (you could tell because of the way that certain floating point numbers were formatted in their web socket messages) which is uh not the language you would pick if “fast” or “reliable” were things you cared about.
It’s true that they had better uptime than many competitors but that’s primarily because they throttled activity *very* aggressively and had huge tail latencies. The exchange rarely fell over, it just degraded massively under stress.
What they did have, which was a major selling point and that traders loved, was a functioning portfolio margin system. On FTX you could trade spot, margin and futures all in one account, with collateral posted in multiple currencies and seamless margin financing. It “just worked”
Compare it to Binance, where by default you have separate wallets for trading spot, spot margin and futures. You have coin-margined and stablecoin-margined futures, and the stablecoin-margined futures are quoted in both USDT and BUSD and they don’t share margin by default.
Binance lists three different BTC perpetuals (BTCUSD_PERP, BTCUSDT and BTCBUSD with different margin systems and different fee schedules). It’s a mess. There is also a separate fee schedule for spot trading.
By comparison FTX just had BTC-PERP. If you’re a retail trader who wanted to get some crypto leverage it was *way* easier to get started on FTX than on Binance or OKX or Huobi or BitMEX.
So it’s no surprise that they were able to scale quickly and attract both retail and professional traders.

But their matching engine? That absolutely sucked.
Mentions
See All