Platform: CloudStack 4.11.2 on CentOS 7 KVM Hypervisor on CentOS 7 I have found some throughput issues on our VirtualRuters and I've tracked it down to CPU IRQ hitting 99% on the VR which was related to NIC interrupts.
I decided to lookup what NIC is being emulated on the VRs; lsmod listed three Intel NICs: 00:03.0 Ethernet controller: Intel Corporation 82540EM Gigabit Ethernet Controller (rev 03) 00:04.0 Ethernet controller: Intel Corporation 82540EM Gigabit Ethernet Controller (rev 03) 00:05.0 Ethernet controller: Intel Corporation 82540EM Gigabit Ethernet Controller (rev 03) All my regular VMs are using Virtio network devices as specified within template settings nicAdapter=virtio When I manually updated the user_vm_details table for a VR with nicAdapter=virtio and restarted the VR everything came up as expected, the VR was started with virtio NICs and the IRQ issue was gone, also the throughput doubled. Now lspci on the VR was showing: 00:03.0 Ethernet controller: Red Hat, Inc Virtio network device 00:04.0 Ethernet controller: Red Hat, Inc Virtio network device 00:05.0 Ethernet controller: Red Hat, Inc Virtio network device The problem I have is getting the nicAdapter setting at systemvm template level like I do for regular VMs so that all VRs are deployed with virtio network adapter. I don't want to set this manually for every network I deploy. So I went to Templates -> Mysystemvm template -> Settings -> Add Setting Name: nicAdapter Value: virtio BUT it just looks to me like systemvms don't honour vm_template_details table where nicAdapter is specified. When a VR gets created, I looked up content of the user_vm_details for the VR and found nicAdapter=virio is missing but I would expect it to be there. If there is anyone running CloudStack with KVM who could help on this that would be great. It might well be a BUG and needs to be reported as such not sure at this stage. If any of the above is not entirely clear please let me know and I will try my best to explain in more detail. Thanks Raf
