On Tue, Nov 25, 2008 at 02:11:10PM +0100, Graeme Stewart wrote:
> This means we have saturated our i/o capacity with a cluster which is
> 15% full of analysis jobs.
>
> I think we need more network cards and bigger switches. I am astonished.
Just out of curiosity what is the value of your rfio/dcap readahead? In
the past I've seen some CMS jobs here spending most of their time
downloading data because they read root files in a semi random way[1].
Since their reads are in ~4K blocks with a readahead of 1MB or so they
ended up downloading each file a few hundred times or so.
Cheers,
Kostas
[1] http://cmsdoc.cern.ch/cms/cpt/Computing/Technical/documents/20080422/iosummary/iosummary.html
|