Обсуждение: Recommended Swap space
Hi,
What is the recommended swap space for postgres or 9.0.11 or 9.2.3 on linux(3.0.58-0.6)
RAM on the hox is 32GB.
Thank you
Nik
My experience is that you're best off either with a swap space that matches or exceeds physical memory, or none at all. The linux kernel swap daemon (kswapd) gets confused and behaves badly with small swap spaces, especially the more memory you have. Good news is that hard drives are cheap and 32G isn't a lot of memory so I'd set it up for 32G + a tiny bit. Once a machine gets to larger memory sizes (128G and more) I just turn off swap.
The really messed up bit is that the problems with the kswapd won't show up for weeks, months, or sometimes even longer. The symptoms of a kswapd problem is that swap is mostly full, but there's LOTS of free memory / kernel cache, and the kswapd is busily swapping in / out. Page faults skyrocket and the problem can last for a few minutes or a few hours, then clear up and not occur again for a long time.
tl;dr: Either physical memory + a little or none.
On Fri, Apr 12, 2013 at 3:05 PM, Nik Tek <niktek2005@gmail.com> wrote:
Hi,What is the recommended swap space for postgres or 9.0.11 or 9.2.3 on linux(3.0.58-0.6)RAM on the hox is 32GB.Thank youNik
To understand recursion, one must first understand recursion.
On Fri, Apr 12, 2013 at 6:15 PM, Scott Marlowe <scott.marlowe@gmail.com> wrote: > The really messed up bit is that the problems with the kswapd won't show up > for weeks, months, or sometimes even longer. The symptoms of a kswapd > problem is that swap is mostly full, but there's LOTS of free memory / > kernel cache, and the kswapd is busily swapping in / out. Page faults > skyrocket and the problem can last for a few minutes or a few hours, then > clear up and not occur again for a long time. Isn't that a NUMA issue?
No, I've had it happen with NUMA turned off. The NUMA issues are with zone_reclaim_mode. If you have it set to 1 (default on a lot of big machines) it can cause serious problems with performance as well.
On Fri, Apr 12, 2013 at 3:21 PM, Claudio Freire <klaussfreire@gmail.com> wrote:
On Fri, Apr 12, 2013 at 6:15 PM, Scott Marlowe <scott.marlowe@gmail.com> wrote:Isn't that a NUMA issue?
> The really messed up bit is that the problems with the kswapd won't show up
> for weeks, months, or sometimes even longer. The symptoms of a kswapd
> problem is that swap is mostly full, but there's LOTS of free memory /
> kernel cache, and the kswapd is busily swapping in / out. Page faults
> skyrocket and the problem can last for a few minutes or a few hours, then
> clear up and not occur again for a long time.
To understand recursion, one must first understand recursion.