Hi Brian
I have heard of this being an issue, but can't recall any details. Any
chance you could repeat test but use castor@cern? Not sure how much
this will tell us since the systems are so different, but if we do see
similar behaviour I think we just note it since clearly this will be
important when migrating disk1 files from dcache.
Shaun
>-----Original Message-----
>From: GRIDPP2: Deployment and support of SRM and local storage
>management [mailto:[log in to unmask]] On Behalf Of
>brian davies
>Sent: 06 July 2007 11:38
>To: [log in to unmask]
>Subject: srmcp pushmode assymmetry for transfers in atlas
>RAL-Lancaster tests.
>
>With regard to the comments about dcache<->castor transfers being
>slow, I have looked at transfers in both directions between lancaster
>and RAL and notice that transfers are consistently slower when using
>the CASTOR server to control the transfer. Is this a known issue of
>interoperability?
>
>srmcp -pushmode=true srm://<RALCASTOR>/<filename>
>srm://<LANCSDCACHE>/<filename>
>it takes on average 8.5 minutes (wall clock) per file for 10
>files in a row
>however without the pushmode to control the transfers it only takes
>46s per file.
>With the reverse diretion of file transfer ie files from Lancasster to
>dcache it is
>48s per file transfer with pushmode=true ( ie lancaster controlled)
>it is when controlled form RAL
>
|