Re:Swap, swap, swap, and bad places to work
by blunte on Friday August 09, 2019 @07:00AM
(#20653135)
It seems to me that if you hit the point where you really need swap, then you're already in trouble. Maybe that swap gives you a little buffer before things get really bad, but chances are that will just keep you unaware of your impending problem until they go critical (unless you have lots of good monitoring/alerting).
[ Parent
| on HN ]
Re:Swap, swap, swap, and bad places to work
by azernik on Friday August 09, 2019 @07:23AM
(#20653262)
You would think so, but as Rachel points out the Linux kernel displays some pathological behavior with noswap that even a tiny amount of swap lets it get around. Something weird is going on in the memory management code.
Re:Swap, swap, swap, and bad places to work
by VectorLock on Friday August 09, 2019 @10:10AM
(#20654449)
I'm not sure what shes observing other than "feels weird" but my observation is the exact opposite, but my observations usually have some measurements to them.
Re:Swap, swap, swap, and bad places to work
by aardvarklegend on Friday August 09, 2019 @10:45AM
(#20654818)
It requires a perfect storm of just shy of 100% used memory and a lot of mmaped io. In that case the mmaped can get shunted to a handful of pages (or even one page) and so you lose all ability to have any block io larger then one page size. And every page of memory involves a fully blocking io request. Itβs most certainly real for certain workloads (databases are common).