pve kernel 4.10.17-1 wrong ixgbe driver

dfenwick

Active Member
Jul 16, 2017
26
1
43
59
pve-kernel-4.10.15-1 from the iso install has ixgbe driver 5.0.4 (correct version)

I updated my host and the ixgbe driver in pve-kernel-4.10.17-1 has been downgraded to the 4.4.0-k, which is the very old and outdated ixgbe driver.

Any chance this can get rolled back into the 4.10.17-1 kernel in the non-subscription distribution?
 
Please read the Intel network driver sourceforce page:

https://sourceforge.net/projects/e1000/files/ixgbe stable/5.2.1/

Note that while we attempt to keep the driver version number (5.2.1) in sync with its counterpart in the Linux kernel that has similar functionality this is far from authoritative. If you are using a newer kernel.org kernel or distro it is likely that its ixgbe driver is at least as up to date as the out of tree (OOT) driver found here.

This is why we stopped to include this driver and use the upstream kernel driver.

Did you run into any problems with the upstream driver?
 
It's missing all of the tunable parameters they added for the 5.x drivers. In particular, it's missing all of the IRQ and RSS steering parameters that are pretty much required in order to get anywhere near 10Gbe in the interfaces. The 5.x driver was quite a long way ahead of the 4.x driver with regard to tuning. If you do a modinfo on each driver, you can see the parameter differences. I'm not exactly sure where the Ubuntu/Debian guys *got* this driver, because the current kernel sources have the driver at version 5.1.0 as of May 2017.

On a side note (I know you work a lot on this, Dietmar - I've read through a lot of the git stuff, so I figured you might appreciate this) I managed to crash my storage array today 4 times before I realized what was happening. It's a Dell PS-M4110. It really does not like it when you do low level TCP tuning with sysctl. I actually *crashed* both the primary and backup controller cards on each of my SAN shelves just by tuning TCP on my Proxmox hosts. I didn't even know I could do that. I think it didn't like tcp_tw_reuse, tcp_tw_recycle, and dsack or some combination of the three.

Incidentally, you guys wouldn't happen to have some reasonable documentation for actually *getting* anywhere near 10Gbe out of open-iscsi would you? I'm getting nothing even remotely close. I can get line speed with iperf3, but iscsi is barely scratching at 1.5Gbps. I've tried just about everything and can't seem to make it go faster. It could very well be the Equallogic array itself. I have no idea what else to tune at this point.
 

About

The Proxmox community has been around for many years and offers help and support for Proxmox VE, Proxmox Backup Server, and Proxmox Mail Gateway.
We think our community is one of the best thanks to people like you!

Get your subscription!

The Proxmox team works very hard to make sure you are running the best software and getting stable updates and security enhancements, as well as quick enterprise support. Tens of thousands of happy customers have a Proxmox subscription. Get yours easily in our online shop.

Buy now!