[Gluster-devel] Client memory footprint

[Pavan T C]

On Wednesday 24 August 2011 02:38 PM, Gordan Bobic wrote:
> On Wed, 24 Aug 2011 11:48:04 +0530, Pavan T C <tcp at gluster.com> wrote:
>> On Wednesday 24 August 2011 09:02 AM, Emmanuel Dreyfus wrote:
>>> Emmanuel Dreyfus<manu at netbsd.org> wrote:
>>>
>>>> Another issue: glusterfs client eats a lot of memory. It grows up as
>>>> big
>>>> as 222 Mo in a few minutes. Here is top(1) output:
>>>> PID USERNAME PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND
>>>> 19913 root 83 0 222M 122M RUN 6:52 42.97% 42.97% glusterfsd
>>>>
>>>> I do not think there is a leak, since that remain fairly stable.
>>>> That is
>>>> still big. Is it a number considered reasonable? Is there a way to tune
>>>> memory usage for smaller systems (cache size?)
>>>
>>> No answer to that one. Are the numbers I have something common, or is
>>> there a problem to fix?
>>
>> This is normal.
>
> Perhaps a more appropriate question would be: "How can this be improved?"
>

The glusterfs server process uses iomem pools. Initially, it starts with 
one pool. Under memory pressure, it allocates more, but does not give 
back the pool when the pressure decreases. It can be debated whether 
this is a bug, but one can come up with heuristics to free the pool.

Pavan

> Gordan