5 ms·
I'll take a stab at explaining what happened here: "No swap in production" probably stemmed from someone who worked on the servers, and that person discovered
by fragsworth 7y ago
I'll take a stab at explaining what happened here:
"No swap in production" probably stemmed from someone who worked on the servers, and that person discovered they were swapping too much, which would eventually lead to hard drive decay. And, maybe for budget reasons or whatever, they decided to make it a policy to disable swap, rather than increase RAM or change the underlying software that caused the servers to use too much RAM.
And now, later, on different servers with different software and different specs, the policy still exists, because nobody really paid much attention to why it was created in the first place.
- pjc50 7y agoSwap causes not disk decay but performance decay; the system gets unamanageably slow and would have been better off with a clean crash and failover.
- 8ytecoder 7y agoOnly if there's thrashing. It’s prohibitively more expensive to build a machine that has enough memory for way more than peak usage than to have some swap that can handle some level of overflow pages. Like everything swap has to be monitored and there’s no need To disable it entirely
- nailer 7y agoIt depends. Sudden peaks of latency can get comparatively more expensive on a trading floor than buying an absurd amount of RAM.
- devonkim 7y agoThis reminds me of a previous company that paid consulting fees to have their Postgres configuration tuned. When the hardware was upgraded nobody knew to review the memory settings. After a series of production outages we realized that the new Postgres servers were utilizing 16 GB of RAM when we had 128 GB available.
- tabtab 7y agoIt sounds like a documentation or communication problem. If somebody makes a technical suggestion, such as "don't do X", the reason behind it should be recorded somewhere.