So, in the Storage Group meeting today, it was agreed that we'd try to
do some more regular tests of rates to Manchester,Glasgow,QMUL from
BNL (for North America, and problematic for most sites), and SARA (for
Europe, problematic inbound for most sites, and usefully right next to
where the UK GEANT2 link enters Europe, so there's no real network in
the way).
Whilst we've not got anything automated up yet, Wahid and I just did
some tests pulling a 2GB (random content) file from ATLASSCRATCHDISK
at SARA to Glasgow.
One transfer up (GLA -> SARA) : 2048MB in 30.3 seconds = 67.6MB/s
(sonar test said: 3.01MB/s)
Four transfers back (SARA -> GLA): 7MB/s +/- 1MB/s
(sonar test said: 0.63MB/s)
Now, we have altered our network configuration at Glasgow since the
Sonar test, and we're still asymmetric in the results, but our values
are 10x the values for the Sonar test!
I hope the Sonar test itself will also show such an improvement on Friday...
Sam
On 28 February 2011 11:28, Alessandra Forti <[log in to unmask]> wrote:
> Thanks, it'll be useful to talk to the NNW sys admins to understand the
> outgoing problems we have.
>
> Manchester reinstalled the DPM node at the end of January might explain the
> "pretty stable" instead of the "very stable". As said in another thread I
> have increased the max tcp snd/rcv windows to 12MB on Friday so I'm waiting
> to see if it had any effect on the sonar tests this week. However while with
> lcg-cp tests I got better results on the incoming traffic and saw tcp
> windows grow to MB size the outgoing traffic is still pretty bad and there
> is something blocking the sender window from growing.
>
> cheers
> alessandra
>
> On 27/02/11 21:56, Graeme Stewart wrote:
>
> Hi
>
> As promised here is the matrix of UK T2s <-> T1s, as measured with the ATLAS
> 'large file' (2GB) sonar tests. The averages data from the 3 sonar 'pings'
> made so far.
> The sheet is coloured in the following way:
> < 3MB/s Red
> 6.4 MB/s Yellow (this is the average for T2->T1)
>> 10MB/s Green
> The colours change linearly from red -> yellow -> green, which might not
> work on older versions of Excel. I have produced a PDF as well.
> The second sheet grabs a few plots from these pages which show evolution
> through time:
> http://bourricot.cern.ch/dq2/ftsmon/channel_view/multi/UKI-SCOTGRID-GLASGOW_DATADISK&UKI-SCOTGRID-GLASGOW_DATADISK&UKI-SCOTGRID-GLASGOW_DATADISK&UKI-SCOTGRID-GLASGOW_DATADISK&UKI-SCOTGRID-GLASGOW_DATADISK&UKI-SCOTGRID-GLASGOW_DATADISK&UKI-SCOTGRID-GLASGOW_DATADISK&UKI-SCOTGRID-GLASGOW_DATADISK&UKI-SCOTGRID-GLASGOW_DATADISK/IN2P3-CC_DATADISK&PIC_DATADISK&INFN-T1
> _DATADISK&BNL-OSG2_DATADISK&SARA-MATRIX_DATADISK&FZK-LCG2_DATADISK&NDGF-T1_DATADISK&TAIWAN-LCG2_DATADISK&TRIUMF-LCG2_DATADISK/2011-01-03/2011-02-27/134/
> http://bourricot.cern.ch/dq2/ftsmon/channel_view/multi/UKI-NORTHGRID-MAN-HEP_DATADISK&UKI-NORTHGRID-MAN-HEP_DATADISK&UKI-NORTHGRID-MAN-HEP_DATADISK&UKI-NORTHGRID-MAN-HEP_DATADISK&UKI-NORTHGRID-MAN-HEP_DATADISK&UKI-NORTHGRID-MAN-HEP_DATADISK&UKI-NORTHGRID-MAN-HEP_DATADISK&UKI-NORTHGRID-MAN-HEP_DATADISK&UKI-NORTHGRID-MAN-HEP_DATADISK/IN2P3-CC_DATADISK&PIC_DA
> TADISK&INFN-T1_DATADISK&BNL-OSG2_DATADISK&SARA-MATRIX_DATADISK&FZK-LCG2_DATADISK&NDGF-T1_DATADISK&TAIWAN-LCG2_DATADISK&TRIUMF-LCG2_DATADISK/2011-01-03/2011-02-27/134/
> http://bourricot.cern.ch/dq2/ftsmon/channel_view/multi/UKI-LT2-QMUL_DATADISK&UKI-LT2-QMUL_DATADISK&UKI-LT2-QMUL_DATADISK&UKI-LT2-QMUL_DATADISK&UKI-LT2-QMUL_DATADISK&UKI-LT2-QMUL_DATADISK&UKI-LT2-QMUL_DATADISK&UKI-LT2-QMUL_DATADISK&UKI-LT2-QMUL_DATADISK/IN2P3-CC_DATADISK&PIC_DATADISK&INFN-T1_DATADISK&BNL-OSG2_DATADISK&SARA-MATRIX_DATADISK&FZK-LCG2_DATADISK&NDGF-T1_DATADISK&TAIWAN-LCG2_DATADISK&TRIUMF-LCG2_DAT
> ADISK/2011-01-03/2011-02-27/134/
> for Glasgow, Manchester, QMUL in the T2->T1 direction.
> Glasgow is very stable, Manchester pretty stable, QMUL less so. Very rarely
> do T1s swap places though.
> Cheers
> Graeme
>
>
|