[gfarm-discuss:50601] Re: help me. gfarm2 broken pipe error

[Osamu Tatebe <tatebe@xxxxxxxxxxxxxxxx>]

Hi Suan,

# Please do not send this kind of emails to gfarm-announce ML, which
# is assumed to announce new software releases mainly.

How many nodes does your cluster have?  We experienced similar
issues during the data analysis challenge.  When accessing the
Gfarm file system from more than hundreds of nodes, the MDS is
down.

For now, we notice two reasons.  One is a thread unsafe code in
multithreaded execution region, which is quickly fixed on August
9 in the revision r4009.  The other is lack of resources of the
MDS.  I suggest to use a multicore machine with more than 16 GB
memory to support hundreds of clients.

At first, please try to the most recent Gfarm code in the
subversion respository at SourceForge.net;

% svn co http://gfarm.svn.sourceforge.net/svnroot/gfarm/gfarm_v2/trunk gfarm_v2

After installing the new version of Gfarm, you need to restart
gfmd and all gfsds (several times for safe).  You do not need to
execute config-gfarm or config-gfsd script again.

Thanks,
Osamu

On Tue, 23 Sep 2008 11:19:24 +0900
"chosuan" <chosuan@xxxxxxxxxxx> wrote:

> I'm using gfarm 2.1.0 in rockscluster 4.3.
> 
> All functions are working well, but whenever I use qsub engine at
> rockscluster,
> 
> after a few minutes or hours, gfarm2 metaserver's gfmd daemon goes down and
> DFS gets cracked.
> 
> The error message for this is "broken pipe error".
> 
>  
> 
> Suan cho