Cached RAM clean

H_TeXMeX_H · 10-03-2011, 04:43 AM

I've never had that happen because of cached memory, but then I don't run any servers or apache, so it's possible.

zer0patches · 10-04-2011, 08:24 AM

Quote:

Originally Posted by syg00

I not normally a fan of resurrecting dead threads, but you're destroying potential evidence doing that. Try echo-ing "1" and see if it has the same effect.
What is you swappiness value ?.

It's currently set to 60, the default. I have read I can use a more conservative value since I have lots of ram. Typically 4-7gb free for whatever that's usually used for caching. Unless there was a runaway process eating memory it should never run out.

free -m always shows 4-7gb free so I don't think that is the case.

The machine does typically swap about around 100k, which I would think is negligible to performance.

echo 1 has the same effect.

It just seems to me that mem requests are not freeing memory as intended. I don't know how to figure out if this is an application issue or a system issue.

syg00 · 10-04-2011, 06:14 PM

Quote:

Originally Posted by zer0patches

The machine does typically swap about around 100k, which I would think is negligible to performance.

Hmmmm - depends what's swapping. I'm of a like mind that you shouldn't be swapping at all with that amount of memory. I'd be looking at 30 as a max setting - mainframes we generally recommend "0" (yes, zero). This is only a recommendation value, not a per-centage or anything. You really want to bias against (any) swapout activity.

Quote:

echo 1 has the same effect.

Good - this was just to ensure pagecache was really the issue, not SLAB/SLUB.

zer0patches · 10-06-2011, 12:44 PM

Quote:

Originally Posted by syg00

Hmmmm - depends what's swapping. I'm of a like mind that you shouldn't be swapping at all with that amount of memory. I'd be looking at 30 as a max setting - mainframes we generally recommend "0" (yes, zero). This is only a recommendation value, not a per-centage or anything. You really want to bias against (any) swapout activity.Good - this was just to ensure pagecache was really the issue, not SLAB/SLUB.

I'm probably going to adjust swappiness to 0 and see what happens. See if I can get swap usage down to 0, we have more ram than we need and should never have an issue running out of actual free mem.

Might give me an idea if swapping is the problem, I suppose if it's swapping lots of really tiny files it could come down to hard drive performance being the issue with seek and r/w times. It is a standard sata 1g 7200 rpm i'm sure.