[Linux-cluster] Question about memory usage

[Robert Peterson]

Danny Wall wrote:
> Are there any resources that can help determine how much memory I need
> to run a SAMBA cluster with several terabytes GFS storage? 
>
> I have three RHCS clusters on Red Hat 4 U4, with two nodes in each
> cluster. Both servers in a cluster have the same SAN storage mounted,
> but only one node accesses the storage at a time (mostly). The storage
> is shared out over several SAMBA shares, with several users accessing
> the data at a time, via an automated proxy user, so technically, only a
> few user connections are made directly to the clusters, but a few
> hundred GB of data is written and accessed daily. Almost all data is
> write once, read many, and the files range from a few KB to several
> hundred MB.
>
> I recently noticed that when running 'free -m', the servers all run
> very low on memory. If I remove one node from the cluster by stopping
> rgmanager, gfs, clvmd, fenced, cman, and ccsd, the memory get released
> until I join it to the cluster again. I could stop them one at a time to
> make sure it is GFS, but I assume much of the RAM is getting used for
> caching and for GFS needs. I do not mind upgrading the RAM, but I would
> like to know if there is a good way to size the servers properly for
> this type of usage.
>
> The servers are Dual Proc 3.6Ghz, with 2GB RAM each. They have U320 15K
> SCSI drives, and Emulex Fibre Channel to the SAN. Everything else
> appears to run fine, but one server ran out of memory, and I see others
> that range between 16MB and 250MB free RAM. 
>
> Thanks,
> Danny
>   
Hi Danny,

I'm not sure what you mean by a Samba Cluster, but I can tell you this:
There's a group working on clustered samba, but it's not quite ready yet.

There's no hard fast rule regarding memory usage.  It seems to me that
the 2G of RAM on your nodes should be enough to run a respectable cluster.
If you're having a memory problem with GFS, you may have hit this 
bugzilla bug:

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=214239

The fix won't be generally available until RHEL4U5.  I think it's in the CVS
repository now if you built your environment from source.  It's been tested
at several sites now, and seems to help a great deal.

Regards,

Bob Peterson
Red Hat Cluster Suite