Jones, Evan, Write Latency and Block Size, here. Don’t know him; like the data; I’ll read the blog for a bit. Let’s play I’m not going to pay a lot for this muffler. I’m thinking, just for a target, why would I willingly allocate more than 10 microseconds per message on all logging (local, prime broker, regulatory, venue, etc, etc) of cash equity HFT messages with off-the-shelf hardware and no substantial mods to the Linux kernel? When you look under the covers, or you know more than me today, maybe it has to be 20 or 30 microseconds but for this game you have to explain (account for the time) for anything above 10 microseconds.
Honeyman, Linux NFS Client Write Performance, 2001, here. This is one of those deals, when honey speaks you listen even if it was ten years ago, but it can get messy, witness, here. Don’t ask why. Never explain, never complain; Fun People; cwtvomk (Come wipe the vomit off my keyboard), that’s just honey warming up, respect.
Smith, Greg, Tuning Linux for low PostgreSQL latency, here.
Linux Dev Center, Low latency in the Linux Kernel, here.
Torvalds, email archive, here.
Westnet, The Linux Page Cache and pdflush: Theory of Operation and Tuning for Write-Heavy Loads, here.