Community discussions

MikroTik App
 
lmatys
newbie
Topic Author
Posts: 28
Joined: Fri Aug 23, 2013 1:39 pm

KVM/QEMU CHR performance

Sat Apr 22, 2023 11:18 am

Hello,

We have prepared physical server for the chr:
- dell r630
- 2x CPU E5-2690 v4 @ 2.60GHz, 28xcpu cores (set to max performance, hyper threading disabled)
- 32 gb ram
- ethernet controller Intel 82599ES 10-Gigabit SFI/SFP+ (bonded 2x10ge)

Virtualization env:
- centos 7
- kvm/quemu
- vm guests virtio drivers
- set 24 cpu cores and pinned to guest vm
- 4xcpu left for the hyper visor

CHR:
- installed 6.48.6 long-term
- 16 gb ram
- 24 vcpu’s
- three interfaces, templete configuration below:

<interface type='bridge'>
<mac address='52:54:00:a2:c7:c4'/>
<source bridge='nm-bridge'/>
<target dev='vnet0'/>
<model type='virtio'/>
<driver name='vhost'/>
</interface>

In general, chr works ok but we can observe strange cpu usage (few of them) on the kvm itself (not at the chr mikrotik system).
All pinned cores to the chr are being used, but three of them have much more load than the others (on the kvm perspective).
If have check it in details, and it is coneccted with the numer of the network interfaces exposed to the chr (i believe that higher cpu load of these three cores is caused by the ‚virtioX-input.0 processess).

Please take a look below:

[admin@bgp-chr-2] > /system resource irq print
Flags: ro - read-only
# IRQ USERS CPU ACTIVE-CPU COUNT
0 1 i8042 auto 0 9
1 ro 4 serial auto 5
2 6 floppy auto 1 2
3 9 acpi auto 2 0
4 10 usb4 auto 3 0
usb3
virtio2
5 11 usb2 auto 4 0
usb1
6 12 i8042 auto 5 3
7 14 ata_piix auto 6 0
8 15 ata_piix auto 7 0
9 40 virtio1-config auto 8 0
10 41 virtio1-requests auto 9 19 400
11 42 virtio0-config auto 10 0
12 43 virtio0-input.0 auto 11 1 224 102 354 <— HERE
13 44 virtio0-output.0 auto 12 118 097
14 45 virtio3-config auto 13 0
15 46 virtio3-input.0 auto 14 863 033 066 <- HERE
16 47 virtio3-output.0 auto 15 18 392
17 48 virtio4-config auto 16 0
18 49 virtio4-input.0 auto 17 680 380 338 <- HERE
19 50 virtio4-output.0 auto 18 31 231
[admin@bgp-chr-2] >

During a medium load, most of the vcpu’s are about 70-80 % idle, but these three vcpu are about 1-8% idle.
Is that a normal behavior? Could it cause the bottleneck in the network and troughput/performance?

Best.
 
elbob2002
Member Candidate
Member Candidate
Posts: 252
Joined: Tue May 15, 2018 8:15 pm
Location: Ireland

Re: KVM/QEMU CHR performance

Sat Apr 22, 2023 9:45 pm

You're running BGP I presume going by the VM name.

BGP performance is much improved in v7. There are threads aplenty elsewhere about BGP performance so that's likely to be your culprit.
 
lmatys
newbie
Topic Author
Posts: 28
Joined: Fri Aug 23, 2013 1:39 pm

Re: KVM/QEMU CHR performance

Sat Apr 22, 2023 10:12 pm

You're running BGP I presume going by the VM name.

BGP performance is much improved in v7. There are threads aplenty elsewhere about BGP performance so that's likely to be your culprit.
True, it is a bgp router.
But in this case, it looks like a packet forwarding load insted of handling bgp process.

Who is online

Users browsing this forum: No registered users and 4 guests