Opinion: auto-scaling is not really good with burst that need to be handled in matters of milliseconds.
-
Show this thread
-
TIPS: 1. use bare-metal servers 2. ensure your infra/services can handle 10x spikes in milliseconds, not minutes 3. UI should rely entirely on websockets 4. push users towards websockets 5. use standard DBs only for historical data, serve real-time data via in-mem processes
9 replies 4 retweets 45 likesShow this thread -
Replying to @paoloardoino
Or properly implement caching? and structure / config your DB? Use GRPC instead of plaintext websockets? That said Bitfinex is functionally more complete than others and way more responsive ^^
1 reply 0 retweets 0 likes -
Replying to @JaspervGelder
gRPC is cool, but I kind of prefer working directly on ZeroMQ. It's really powerful and removes central brokers. Kind of helps keeping latency down. We're experimenting really hard with reliable multicast too.
1 reply 2 retweets 1 like -
Replying to @paoloardoino
Risk of losing messages and ordering issues are not really ideal if you want to keep your exchange auditable. For the user / algo trader however multicast would be awesome it is ok if you drop messages but rather Fast/incomplete > out of order and complete ;p
1 reply 0 retweets 0 likes
Reliable multicast is what is being used by Nasdaq for order input. Of course is not a Swiss knife that can be used for everything ;) Also you can have separate listeners that record to disk what you're broadcasting.
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.