Hi Brian,
I said it a large number of times already. I removed the bonding and then 2
months later I reinserted it with the hash suggested by Glasgow. The new
bonding configuration made the incoming traffic, that didn't need it, better
but didn't change anything on the outgoing. I looked for a while at the
behaviour of the machines network cards with ifstat and observed that the
outgoing traffic always uses 4 links but the incoming traffic always use only
one.
We suspect there is a mismatch in how the switches and the machines interprets
lacp. I was about to make some more tests with a cisco switch (rather than
DELL) and the help of NNW people when the DRI bid was announced and I put that
aside. I hope I can still carry out the tests with some of the machines that
will remain on 1Gbps switches.
cheers
alessandra
Alessandra Forti
NorthGrid Technical Coordinator
University of Manchester
On Mon, 19 Mar 2012, [log in to unmask] wrote:
>
> Last of my historical analysis of T1S-UK_T and vice verse ( plus RAL to
> T”DS).
>
> What would be interesting to know is for the following channels there was a
> period where rate improved dramatically.
>
>
>
> http://bourricot.cern.ch/dq2/ftsmon/channel_view/multi/SLACXRD&UKI-NORTHGRID-MAN-HEP&TRIUMF-LCG2&TRIUMF-LCG2&UKI-NO
> RTHGRID-MAN-HEP&SWT2&UKI-NORTHGRID-MAN-HEP/RAL-LCG2&CERN-PROD&UKI-NORTHGRID-MAN-HEP&UKI-NORTHGRID-LANCS-HEP&NIKHEF-
> ELPROD&RAL-LCG2&INFN-T1/2011-02-09/2011-06-23/168/
>
>
>
> Now it was almost a year ago, but if we can work out what changed in late
> march ( and then again two months later;
> we could actually make progress.
>
>
>
> Until after the network upgrades at T2s have been completed, best method to
> track is to follow the page Duncan
> pointed out last week ( thank you)
>
> Brian
>
>
>
>
>
>
>
>
|