weizhouapache commented on PR #7333: URL: https://github.com/apache/cloudstack/pull/7333#issuecomment-4117737708
> > We had to implement this change using the agent.hooks.libvirt_vm_xml_transformer.script hook on KVM hosts to enable multiqueue for Virtual Routers. > > After applying to virtio interfaces, we observed a significant improvement in throughput, confirming that VRs are currently limited by single-queue virtio by default. > > It may be worth considering extending support for these NIC settings to System VMs (Virtual Routers, Internal LB), since they can become network bottlenecks in high-throughput environments. > > A possible approach could be to leverage System Offering External Details, where key/value pairs are already supported. For example: • nic.packed.virtqueues.enabled • nic.multiqueue.number > > This would provide a clean, native and persistent way to tune virtio NIC performance for System VMs, avoiding the need for custom libvirt hooks. > > @matheusfontes Thanks for sharing. This is very interesting, and it makes sense to consider extending VM settings support to Virtual Routers. > > Regarding the approach, extension-specific details are typically handled through the CloudStack extension framework. We could consider extending support for additional settings in service offering details. At the moment, service offering details are mainly used for constraint service offering such as min/max CPU and memory. This might provide a more flexible and native way to configure such parameters, including NIC-related options, @matheusfontes can you try with some key/value settings of the SYSTEM template ? <img width="2417" height="1265" alt="image" src="https://github.com/user-attachments/assets/a9e890c1-e6c2-49db-9c67-0835ee4d8cd1"; /> -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
