[Samba] Clearcase, Samba, and mnode values

[Kathy]

Thanks Volker, regarding Valgrind.  And thanks for the reply Jeff.
You guys have made me feel better about Samba, because it supports
what I was noticing in that although Samba memory usage was
increasing, I was not seeing it consuming the amount of memory that I
saw being cached.  Maybe I should be asking Jeff this offline, since
it isn't a Samba question, but I would love to know how (if possible)
to tell how much is reclaimable RAM.  I have seen certain applications
bring a system to its knees (Cadence comes to mind) once all "free"
memory is cached and we have to reboot because the application no
longer works.  To me, that is a problem with how Cadence is managing
memory.  Yet on our other Clearcase server, I see it running with
everything cached and is completely happy.  Perhaps the only way to
know if the app is happy in this state is to let it run and see what
happens?  Unfortunately, the persons I have been talking with at IBM
are still pointing at Samba and the fact that I was 0.01 out of spec
on Samba version.  I need to find another resource to talk to there.

Thanks --

Kathy

On Thu, Sep 8, 2011 at 11:55 AM, Jeff Layton <jlayton at redhat.com> wrote:
> On Thu, 8 Sep 2011 10:14:47 -0700
> Kathy <banshee135 at gmail.com> wrote:
>
>> That's possible and yesterday I was looking at possibly using Valgrind
>> to see if I could dig further into that idea.  I've never used it
>> before, though, so not sure if there is an easier method to detect
>> kernel memory leaks.
>>
>> And about static things in swap, I agree.  I have noticed on our old
>> Clearcase/Samba server, that it consumes all the memory down to about
>> 150M plus 72k of swap and just sits there like that.  Seems to be fine
>> and can run for 2 months or longer like that.  That server, though,
>> has only 4 gigs of memory and so I was assuming that it did that
>> because it didn't have a lot of memory.  However, this new Clearcase
>> server, which has 32 GB of memory appears to perhaps want to do the
>> same thing.  So I began to wonder if that is just normal behavior --
>> i.e., it caches all its memory.  But I think it's a problem because
>> people started to report Clearcase running really really slow when it
>> got down to almost nothing left and it just seems odd that it would
>> consume all 32 GB of memory in less than 12 hours.
>>
>
> That's normal. Linux will use up as much free RAM as it can to cache
> file data, based on the principle that free RAM is wasted RAM. What
> really matters is not free RAM, so much as *reclaimable* RAM.
>
> If the memory is clean (meaning that it doesn't have data that needs to
> be written back out), then the kernel can just free it on a
> least-recently-used basis when the need arises. If not, then the kernel
> will require more active participation to free up memory, which is
> comparatively slow.
>
> I think you'll probably need to step back and determine what the
> application is doing when it becomes slow. It may very well be that
> there is a problem with memory allocation at that time that's causing
> the slowdown. But, you can't really assume that or you might end up down
> a rabbit hole that has nothing to do with the real problem. Determining
> that will probably require help from IBM as only they have real insight
> into clearcase -- it's a closed source program, after all.
>
> Either way, it's highly doubtful that this has anything to do with
> samba.
>
> --
> Jeff Layton <jlayton at redhat.com>
>