Dr D J Colling wrote:
Hi Lassi,
The iowait we observed happened when we where doing dCache->dCache
transfers using fts.
You can see the iowait difference between srmcp and fts on this page
http://www.gridpp.ac.uk/wiki/IC-HEP#Transfer_Tests_2006-02-22
We compared how the data is writen on the disk when using fts or srmcp
http://www.hep.ph.ic.ac.uk/~georgiou/dcache_fts/fts-0.png
http://www.hep.ph.ic.ac.uk/~georgiou/dcache_fts/srm-0.png
you can see the difference yourself. Basically using fts causes the disk
to jump when writing while srm doing fine.
The transfer rate we could achieve using srmcp (with phedex) was close
to 400Mb/s while with fts we never reached more than 140Mb/s for
dcache->dcache transfers.
More tests haven been achieved by configuring the fts channel to use
srmcp rather than gsiftp third party copy. It improved the situation
but not dramatically. There will be an upgrade to 1.5 at ral next monday
and I hope we can make another set of tests.
Cheers, Olivier.
cc: gridpp storage list
> Hi Lassi,
>
> We see a considerable degredation in performance moving to FTS ... from
> memory we went from something 50MB/s (firewall limited) to something
> like 20MB/s ... and the CPU load on the dcache pool nodes goes through
> the roof. Which makes me quite nervous about this switch. Has anybody
> else seen changes in performance when making this switch?
>
> These have been general transfer tests not done with PhEDEx but I have
> no reason to believe that would make a difference.
>
> I have cc'ed Olivier van der Aa the LT2 technical manager who has been
> performing these tests (with other people around the UK) and Kostas
> Geourgiou who is our sysadmin who has been investigating them.
>
> All the best,
> david
>
> On Thu, 1 Jun 2006, Lassi A. Tuura wrote:
>
>>
>> *** Discussion title: PhEDEx Discussions
>>
>>> CNAF is a Castor site afaik and that's why I suspended the
>>> transfers. If you and
>>> Daniele are sure, that it is not a problem confirms to do transfers
>>> in pushmode
>>> from CNAF, I will of course re-enable them.
>>
>> Bonus points for planning to switch to FTS soon, and even more
>> brownie points for actually switching to FTS :-)
>>
>> Lassi
>>
>> -------------------------------------------------------------
>> Visit this CMS message (to reply or unsubscribe) at:
>> https://hypernews.cern.ch/HyperNews/CMS/get/phedex/207/4/1/1/1/1.html
>>
--
- O. van der Aa - Imperial College London -
- LT2 Technical Coordinator -
- tel: +442075947810, +442071005426 -
- SIP: [log in to unmask] -
- fax: +442078238830 -
- http://surl.se/agtu -
|