Proxmox and OPN/pfSense, Mikrotik CHR

[syd]

Hello!

We know the problem with other than Linux routing platforms in efficiently on Proxmox. Some tests results are shown here:
https://blog.kroy.io/2019/08/23/battle- ... l-routers/

And here the exactly reason of issue is explained (CHR casus):
https://www.youtube.com/watch?v=Jh9HskaS-DA&t=410

So, after seeing this, can anybody well knowing how Proxmox work on this level, tell us is that possible to change its behavior on KVM (by currently available tweaks) to make it doing well or maybe a completely new drivers are needed and it's no hope for now?

I didn't find something useful information on forum (in English), so I hope I do not duplicate questions by making a new thread.
Best Regards.

 

[wolfgang]

Hi,

Netgate the makers of PFSense itself say when it comes to performance you should use tnsr.
tnsr is there new Linux based router platform.
So a company that makes over the last 10 years FreeBSD witch to Linux for performance reason[1].

The network stack in KVM depence on two things Memory speed and CPU frequency.
Ther are the option to speed up the network like vhost, SR-IOV, DPDK, but they are not included in Proxmox VE.

1.)https://www.netgate.com/products/platform.html

 

[syd]

Let me cite explanation of the problem here from a talk given by Mikrotik's team member:

"There is a challenge there and it creates a serious throughput/performance issue when you're using this in ESXi and KVM Proxmox because of the way the buffer are arranged transfer the packets out. Hyper-V on the other hand does not do it that way. Hyper-V has completely different buffer assiggnment for the way it moves packets through the hypervisor and so it actually does not suffer that issue and you can put jumbo frames into Hyper-V without issue.

So in Proxmox packets are assembled into 64k buffers which seems to create MTU issues for the CHR. This affect explicit null the most."


Can somebody tell us if that issue is made by Proxmox's proces which parameters can be changed to right value or just disabled for particular VM to achieve results similar to routers on common Linux distros?

Is it maybe something like "packet buffer fragmentation size" expressed in MTU or kb units somewhere in tweaks settings?

Thank You.