Following mostly the advice of
http://fasterdata.es.net/fasterdata/host-tuning/linux/
mainly slight change is to make maximum tcp window size between 4-12M (my recommendation at the moment has been to say 8M)
default window size people have between 128k and 1M
minimum window size, majority have either 4k or 8k, some go as much as 128k or even 1M !! (erroneously in my un-substantiated opinion)
For a local transfer the tcp settings are less important I t might be we have too many transfers concurrent to get a high rate.)
Liverpool currently have the following number of transfers per fts channel.
Ral-Liverpool= 10 for Atlas, 5 for t2k
Liverpool-ral= 15 for Atlas, 5 for t2k
Star-Liverpool= 10 for Atlas, 5 for t2k
Personally I am not sure I would want to change this that much since you need high concurrent transfers to deal with small files.
Brian
-----Original Message-----
From: John Bland [mailto:[log in to unmask]]
Sent: 05 April 2011 14:24
To: Davies, Brian (STFC,RAL,ESC)
Cc: [log in to unmask]
Subject: Re: UK FTS channels that don't make the SONAR test
Hi,
What are the generally accepted TCP Window sizes these days? We may have
missed a tweak somewhere.
Or we might have to take it up with the uni network guys.
John
On 05/04/2011 13:40, [log in to unmask] wrote:
> So the following channels are those which would not make the 10MB/s for
> atlas transfers are in the following channels:
>
> http://bourricot.cern.ch/dq2/ftsmon/channel_view/multi/UKI-SOUTHGRID-BHAM-HEP&UKI-NORTHGRID-LIV-HEP&UKI-NORTHGRID-SHEF-HEP&RAL-LCG2&UKI-SCOTGRID-DURHAM&UKI-LT2-RHUL&UKI-LT2-UCL-HEP&RAL-LCG2&RAL-LCG2&RAL-LCG2/RAL-LCG2&RAL-LCG2&RAL-LCG2&UKI-NORTHGRID-LIV-HEP&RAL-LCG2&RAL-LCG2&RAL-LCG2&UKI-LT2-RHUL&UKI-LT2-UCL-HEP&UKI-SCOTGRID-GLASGOW/2011-03-01/2011-04-05/19/
> <http://bourricot.cern.ch/dq2/ftsmon/channel_view/multi/UKI-SOUTHGRID-BHAM-HEP&UKI-NORTHGRID-LIV-HEP&UKI-NORTHGRID-SHEF-HEP&RAL-LCG2&UKI-SCOTGRID-DURHAM&UKI-LT2-RHUL&UKI-LT2-UCL-HEP&RAL-LCG2&RAL-LCG2&RAL-LCG2/RAL-LCG2&RAL-LCG2&RAL-LCG2&UKI-NORTHGRID-LIV-HEP&RAL-LCG2&RAL-LCG2&RAL-LCG2&UKI-LT2-RHUL&UKI-LT2-UCL-HEP&UKI-SCOTGRID-GLASGOW/2011-03-01/2011-04-05/19/>
>
> so these are
>
> ral-Liverpool
>
> Liverpool-ral
>
> ral-RHUL
>
> RHUL-ral
>
> ucl-ral
>
> ral-ucl
>
> Birmingham-ral
>
> Durham-ral
>
> ral-Glasgow
>
> For Liverpool, RHUL and ucl its bi-directional
>
> But for Birmingham Durham and Glasgow its single directional.
>
> Glasgow is an interesting issue.
>
> On the 25^th we increased from 5 to 20 concurrent transfers due to clear a
> backlog of files. For two weeks before the channel average was 11.0 MB/s but
> two weeks after it is 7.1 MB/s ( so now failing) this can be seen in these
> two plots:
>
> Before
>
> http://bourricot.cern.ch/dq2/ftsmon/multi_spacetoken_view/RAL-LCG2/UKI-SCOTGRID-GLASGOW/2011-03-11/2011-03-25/12/0/1/
>
> After
>
> http://bourricot.cern.ch/dq2/ftsmon/multi_spacetoken_view/RAL-LCG2/UKI-SCOTGRID-GLASGOW/2011-03-25/2011-04-08/12/0/1/
>
> I see this as possible evidence that number of concurrent files is a factor
>
> An url to compare bi-directional can be seen here:
>
> Durham is here:
>
> http://bourricot.cern.ch/dq2/ftsmon/channel_view/multi/RAL-LCG2&UKI-SCOTGRID-DURHAM/UKI-SCOTGRID-DURHAM&RAL-LCG2/2011-03-11/2011-04-25/24/
> <http://bourricot.cern.ch/dq2/ftsmon/channel_view/multi/RAL-LCG2&UKI-SCOTGRID-DURHAM/UKI-SCOTGRID-DURHAM&RAL-LCG2/2011-03-11/2011-04-25/24/>
>
> Liverpool is here:
>
> http://bourricot.cern.ch/dq2/ftsmon/channel_view/multi/RAL-LCG2&UKI-NORTHGRID-LIV-HEP/UKI-NORTHGRID-LIV-HEP&RAL-LCG2/2011-03-11/2011-04-25/24/
> <http://bourricot.cern.ch/dq2/ftsmon/channel_view/multi/RAL-LCG2&UKI-NORTHGRID-LIV-HEP/UKI-NORTHGRID-LIV-HEP&RAL-LCG2/2011-03-11/2011-04-25/24/>
>
> Brian
>
--
John Bland [log in to unmask]
System Administrator office: 220
High Energy Physics Division tel (int): 42911
Oliver Lodge Laboratory tel (ext): +44 (0)151 794 2911
University of Liverpool http://www.liv.ac.uk/physics/hep/
"I canna change the laws of physics, Captain!"
|