[Gluster-users] mnt-mail.log - memory ???

[Michiel van Es]

Craig,

I suffer teh same issue, the sync; echo 3 > /proc/sys/vm/drop_caches 
speeds up my glusterfs.
How often should I run it?
I have it scheduled once a day:

0 8 * * * sync; echo 3 > /proc/sys/vm/drop_caches

Michiel
On 5/13/10 12:14 AM, Craig Carl wrote:
> Bryan -
>     Not a memory leak. Usually what happens is the kernel is using that space as disk cache. To check for that run "top", kernel disk cache is reported as "cached". I expect that is what is using all of the memory. In theory all of that memory is available for user processes immediately upon request but the actual behavior seems to vary widely based on disti.
>     Two possible solutions.
>
> 1. Run "sync; echo 3>  /proc/sys/vm/drop_caches" as root. This drops the entire cache on the host. Kinda like using a sledgehammer to kill a fly. Some users put this in cron.
>
> 2. Adjust the swappiness. This is the right solution for the issue. There is a really good explanation of the process here(1). In the 2.6 kernel we can adjust the aggressiveness with which the kernel flushes disk cache out of memory. In most distributions the default value is 60, the closer you move the value to 100 the more frequently the kernel will page inactive items out of the cache. I'd suggest you experiment a little bit to find a good value for your environment.
>
>      Once you get the swappiness tuned it would great if you could let us all know what you decided on and how it is affecting memory utilization.
>
> (1) http://www.linuxhowtos.org/System/Linux%20Memory%20Management.htm
>
> Thanks,
>
> Craig
>