Metauser

NetXen Incorporated NX3031 Multifunction 1/10-Gigabit Server Adapter

November 18, 2009 Posted by mitch | work | 17 Comments

So, I installed my first 10GigE card the other week... for a high throughput NAS head for a digital printing company I contract for. For the last couple months I've been planning and preparing to swap out their entire network core with new Juniper equipment... I've installed a couple stacks of the Juniper EX4200 switches, and we even purchased a 10GigE module for a 300 meter fiber link on both ends of the building. As part of this... we thought we could improve their through put to the NAS server (CentOS 5.4 x86_64 with Samba and SAN disk, and RedHat Cluster Server) by trunking ports. As soon as I had two 1Gbit copper links trunked to these... the speed seemed to be cut in half with samba... SSH seemed to be very fast... some googling and I'm seeing reports that in a bonded nic senario, samba drops to half speed... but NFS and other protocols seem to be able to take advantage of it... Now... I'm sure an upgrade to samba would fix this... but I don't want to introduce new variables going into their busy season... So, we purchased a 10GbE card for the server so we wouldn't need to trunk it.

Moved over to this server with the NetXen 10GbE card... pluged it into the Juniper Switch... the link works fine... we move the digital presses over to it and it works fine for a while... after a couple hours running, I start getting errors in the log:

Nov 17 20:34:54 gondor kernel: NETDEV WATCHDOG: eth4: transmit timed out
Nov 17 20:34:54 gondor kernel: netxen_nic eth4: transmit timeout, resetting.
Nov 17 20:34:58 gondor kernel: NETDEV WATCHDOG: eth4: transmit timed out
Nov 17 20:34:58 gondor kernel: netxen_nic eth4: transmit timeout, resetting.

Did some googling, and I've found an article to turn off the tcp segmentation offload

so I run:
ethtool -K eth4 tso off

and you can check it with:
ethtool -k eth4

I will update this post if it makes any difference

---- UPDATE ---- Nov 30, 2009 3:55 PM MST ----
After looking around... HP wants to use a nx_nic driver instead of the netxen_nic driver that comes with CentOS/RedHat Enterprise by default. So, I've compiled the SRPM with my build box, and Installed it on the new server. Version 4.0.516. I then upgraded the firmware on the NIC to the same. I will do some more testing to see if this is the fix I require....