How to make CSR1000v run stable and not use 100% CPU

Moderator: mike

Post Reply
Posts: 8
Joined: Wed Nov 15, 2017 8:44 am

How to make CSR1000v run stable and not use 100% CPU

Post by sven » Fri Feb 14, 2020 11:16 am

Hi all,

I wanted to share some ways to make CSR1000v images stable. I'm using in my viptela SDWAN lab:
- EVE-NG PRO 2.0.6-16
- IOS-XE SDWAN 16.12.1e and 16.12.2r

From the start it was a nightmare. CPUHOG messages, crashes, routing (OSPF/EIGRP) and LAN (VRRP) completely flapping all the time. I tried to increase the CPU to 2 per node (4GB RAM) but to no avail. In the cli of eve-ng, the CPU was always at 100%. All pointed to a change in the code of IOS-XE and how it 'polls' to see if data is received on the wire.

Today I had a major breakthrough and got everything happily spinning like a kitten. As other people may have the same problems, here's what I did:

. Under the hood 1EVE-NG is using QEMU to run the virtual images. The default template when deploying a CSR1000v is to use the following
- Qemu 3.1.0
- and qemu option “-cpu Nehalem”

I changed this to
- Qemu 4.1.0
- and qemu option “-cpu Skylake-Server-IBRS”

The use of 4.1.0 seems to be enough, but since my ESXi is running on “intel xeon e5-2695 v4” and Nehalem is from a decade ago, I used what I think is the corresponding codename “Skylake-Server-IBRS” (so changing this depends on which CPU model you have). This should enable the extra features that this CPU model supports. For example, one of them is AES. If you run Nehalem you will get a warning when booting the CSR1000v saying that AES-HW is not present and that performance may be impacted. I think this also caused the CPU to drop from 100% to 66% (which is the best I have seen anywhere)

Hope it helps,

Uldis (UD)
Posts: 2814
Joined: Wed Mar 15, 2017 4:44 pm
Location: London

Re: How to make CSR1000v run stable and not use 100% CPU

Post by Uldis (UD) » Sat Feb 15, 2020 5:15 pm

upgrade your eve
latest is 2.0.6-34

actually in to even it is hit over 100% it does not harm image working..
if you are running heavy lab please count your eve cores vs lab
it is main reason of HOG errors

Post Reply