Hi all,
Our pool node has been taking a bit of a pounding during the FTS transfer
tests from RAL. The machine has 16GB of RAM all of which was being used to
handle the transfers. You will be able to see the high memory and CPU
usage using the Ganglia page for our pool node:
http://mon.epcc.ed.ac.uk/ganglia/?r=day&c=ScotGrid-Edinburgh&h=dcache.epcc.ed.ac.uk
8 parallel transfers were being used during these tests.
Have any of the other Tier-2s (IC, Lancaster) experienced this sort of
behaviour during sustained transfers? What about RAL? Andrew Sansum
definitely mentioned that RAL and SARA may have been seeing a problem like
this that was solved by reducing the number of parallel transfers.
I think it would be good to find out if this was just an issue with
Edinburghs setup or a more general dCache issue due to java processes
consuming large amounts of pool node resources. This may be an issue we
see more of once all Tier-2s get an SRM (assuming they use dCache that
is!).
Any information would be useful.
Thanks,
Greig
--
=======================================================================
Dr Greig A Cowan http://www.ph.ed.ac.uk/~gcowan1
School of Physics, University of Edinburgh, James Clerk Maxwell Building
DCACHE PAGES: http://www.gridpp.ac.uk/deployment/admin/dcache/index.html
=======================================================================
|