Intel 82576 Jumbo frames issue (Supermicro X8DTH-6F), with fix

[bana]

On 8.0.1 beta4, my gigabit ethernet interfaces can not allocate receive buffers when asked to use Jumbo frames. This can be fixed by do in the following:

login as root and type enter the following commands:

mount -uw /
echo kern.ipc.nmbclusters=\"131072\" >> /boot/loader.conf
echo hw.igb.num_queues=\"4\" >> /boot/loader.conf
sync ; reboot

 

[JimGat]

Bana,

What is the symptoms?
Where did you see it running out of receive buffers?



We are running a FreeNAS 8.01 x64 on a supermicro chassis with Core i7 6GB Ram and two on board Intel and one Intel Pro Quad Port Card. 6 Gig ports using the igb driver.
We are currently only using one of the Quad Port Gig Ports for NFS to 3 VMware ESXi 4.1

Before upping the kern.ipc.nmbclusters="128000" we were having the non-responsive management GUI and the SSH console described in another post.

After adding the kernel parameter everything works very well and will out perform our other Linux based NFS stores.

The problem we are having now is the storage network card (Intel Quad Port) will stop communicating. Seems that it does that on all 4 ports. From the FreeNas console you can ping the IP's of the card. Nothing Shows down, but not ping-able from the ESX Boxes. Disconnecting the switch has no effect (HP Managed switch). There are no packet errors that have occured. We are using Jumbo Frames MTU=9000. NFS is still active on the Management IP side, I was connected Via SSH on the management IP. Dowing Just the IGB2 (primary IP on the intel) seems to allowed the entire card to work. I am going to post in another thread If I can find one more aplicable. .

Thanks for any help in advanced and thanks to the Team for such a great NAS Product!

 

[bana]

I needed to up both the clusters and the queues, mind you mine is dual port. If either number was left unchanged, i loose all connectivity. You might want to experiment with 8 queues.