Community discussions

MikroTik App
 
artkomnet
just joined
Topic Author
Posts: 2
Joined: Thu Feb 13, 2020 4:43 pm

VM performance - x86 vs CHR

Tue Sep 14, 2021 1:25 pm

Hello.
We have x10sdv-tp8f + xeon D-1518 + 4Gb ram + 128GB ssd mvme
I did some simple tests to choose the most efficient solution for the main router (nat, firewall, queues, pppoe server).
I tested loopback performance: btest (10 seconds average) send udp / tcp and queue test (6x remove & import script for 500 queues).
ROS 6.48.4, Winbox 3.30

CCR1036 (for comparison)
btest - 1016 Gbps udp / 18 Gbps tcp
queue test ~ 9 seconds

Windows Server 2016 Hyper-V
btest - 187 Gbps udp / 41 Gbps tcp
queue test ~ 7 seconds

Proxmox x86
btest - 820 Gbps udp / 38 Gbps tcp
queue test ~ 1 second

Proxmox CHR
btest - 825 Gbps udp / 34 Gbps tcp
queue test ~ 2 seconds

I would prefer CHR because of the licensing model, but I noticed that when working with winbox, it can freeze / hang when importing scripts and navigating and loading the interface is much less responsive.
Nothing like this happens on an x86 VM. Everything loads in a flash.

Are there any specific settings in proxmox that may affect the performance of the virtualized mikrotik?
Maybe Extra CPU Flags in proxmox?
 
artkomnet
just joined
Topic Author
Posts: 2
Joined: Thu Feb 13, 2020 4:43 pm

Re: VM performance - x86 vs CHR

Thu Sep 16, 2021 3:05 pm

We already have an active p10 license for CHR

I'm currently testing "Traffic Generator" from MUM pdf - https://mum.mikrotik.com/presentations/ ... 963921.pdf
It's strange, but applying the performance tricks given in this forum does not increase performance but reduces it.

Turning off HT does nothing. Adding a multiqueue for ports only increases the core load. I get the best performance with the default proxmox settings.
 
elbob2002
Member Candidate
Member Candidate
Posts: 252
Joined: Tue May 15, 2018 8:15 pm
Location: Ireland

Re: VM performance - x86 vs CHR

Sat Sep 18, 2021 6:05 pm

Out of curiosity what type of NIC have you?

I've seen issues with some Intel NICs on Linux where the NIC would what would seem like half it's advertised speed. Affected me on XCP-NG.

The solution was to turn off TCP Segmentation Offloading for the interface like this:
ethtool -K <interface> tso off gso off
It seemed to be specific to Linux 4.15 kernel though.
 
User avatar
Amm0
Forum Guru
Forum Guru
Posts: 3169
Joined: Sun May 01, 2016 7:12 pm
Location: California

Re: VM performance - x86 vs CHR

Wed Sep 22, 2021 9:03 am

First, I'd look at what CHR virtual machine network settings, specifically the emulated network card on Hyper-V – that should not be that slow if this was all done on same hardware.

But since X86 already works on same hardware, using PCI passthrough for the network card may possible and/or good idea if you're looking to maximize performance. PCI passthrough requires multiple ethernet ports since once passthrough, you couldn't get to the Proxmox virtual server if all ethernet ported are passthough'ed. Now without passthrough, all the traffic would hit the Linux/Proxmox via a software bridge – likely quick stop, but won't happen on X86 version, maybe the difference in raw speed. A quick forum search turns up... so a little unclear what to expect with Proxmox:
viewtopic.php?f=15&t=175046&p=857202&hi ... io#p857202
...now whether CHR includes same network drivers as X86, hard to know without test it.

We normally use VMWare with CHR – never an issue. I believe the hypervisor is free as long as you don't want the management features – but you can read the exact rules. With ESXi, we use the VMXNET3 network and in most cases for CHR enable Promiscuous Mode on virtual switch too. We don't use passthrough since the CHR in our case is the router for other VM guests on same box.

Who is online

Users browsing this forum: No registered users and 8 guests