[mira_talk] Re: insufficient temp space on cluster nodes...
- From: Bastien Chevreux <bach@xxxxxxxxxxxx>
- To: mira_talk@xxxxxxxxxxxxx
- Date: Tue, 9 Oct 2012 08:27:11 +0200
On Oct 9, 2012, at 3:34 , Langhorst, Brad wrote:
> Mira jobs are failing on our cluster due to insufficient local tmp space.
> It's not easy to upgrade these.
>
> However, these nodes are connected to fast storage on a 10G ethernet
> connection - probably this is faster storage than local.
> Is it possible to use mira with tmp on NFS, or tmp on nfs + tmp on local?
In principle yes. You can either start MIRA directly in a directory which is on
NFS, then everything will be there. Or you can redirect specifically the
"*_assembly/*_tmp" directory with
-DI:trt=/path/to/whereever
However, MIRA will print out a warning if it detects it is running on NFS. The
reason is that users reported terrible performance of earlier versions of MIRA
on NFS attached storage (10x slower or worse). The protocol comes with a
certain overhead and writing to files over NFS behaves differently
(blocking-wise) than local files. I have since located and eradicated some of
the causes for the slowdown, but cannot guarantee that MIRA runs at full speed
if the tmp directory is on NFS.
I'd be interested on feedback though :-)
Best,
Bastien
--
You have received this mail because you are subscribed to the mira_talk mailing
list. For information on how to subscribe or unsubscribe, please visit
http://www.chevreux.org/mira_mailinglists.html
Other related posts:
