[elrepo] ixgbe

[Schramm, David]

There could always be a better way, but in this case, the NFS server is an appliance and very good at what it does. This isn't ultra-high IOPS, this is more large IOs (enough to pull about 1GB/sec or more per node) which is why we are trying to squeeze out as much as possible from the hardware. So NFS is the only protocol in play right now.


From: elrepo-bounces at lists.elrepo.org [mailto:elrepo-bounces at lists.elrepo.org] On Behalf Of Grigory Shamov
Sent: Thursday, March 02, 2017 8:13 AM
To: EL Repo General Mailing List
Subject: Re: [elrepo] ixgbe


Could it be that switching from NFS to Lustre is a better way than doing it through drivers and kernel?

Grigory Shamov
________________________________
From: elrepo-bounces at lists.elrepo.org<mailto:elrepo-bounces at lists.elrepo.org> <elrepo-bounces at lists.elrepo.org<mailto:elrepo-bounces at lists.elrepo.org>> on behalf of Schramm, David <david.schramm at tdstelecom.com<mailto:david.schramm at tdstelecom.com>>
Sent: Thursday, March 2, 2017 7:37:25 AM
To: elrepo at lists.elrepo.org<mailto:elrepo at lists.elrepo.org>
Subject: [elrepo] ixgbe

Hi all,
We've embarked on a project to push the limits of NFS for an analytics cluster (RHEL 6) and seeing how many of the more advanced features in the Intel cards (82599 in this case) can't be tuned with the 4.0.1-k version, looking to see about a module for the 5.x version ( 5.0.4 is the latest I believe). The vendors involved with this effort have all suggested tweaking the cards for this application due to the high bandwidth needed over few streams. What do you think?

thanks
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.elrepo.org/pipermail/elrepo/attachments/20170302/76cbc7f7/attachment-0001.html>