Hello,
I’ve recently installed Peplink FusionHub on a KVM Linux hypervisor and am looking to optimize its performance, particularly for 10Gbps throughput.
My KVM host specifications are: Intel Xeon Gold 6230 @ 2.10GHz / 64 GB DDR4 / NICs: 2x10G Intel X710
For the FusionHub VM, I’ve configured it with: RAM: 16GB (with 1GB hugepages), vCPUs: 8 and network: virtio-net-pci with vhost=on.
In my lab, this configuration delivers good performance, reaching about 4Gbps throughput (2Gbps inbound + 2Gbps outbound) with overall FusionHub CPU utilization around 25%.
But during the tests, I’ve observed that vCPU utilization is not always optimally distributed. Depending on the traffic pattern (sessions/protocol type etc…) I observe that only two vCPUs heavily loaded (near 100%) while the remaining six vCPUs are underutilized.
This suggests that the workload isn’t being effectively distributed across all available vCPUs.
So my question is:
Does FusionHub support vhost multi-queue for virtio-net interfaces?
If I configure for example vhost queues=4 or more will FusionHub properly leverage these queues to distribute the network load more evenly across its 8 vCPUs?
Are there any further KVM-level optimizations you would recommend to improve performance and ensure better CPU core load distribution within the FusionHub VM in multi-core enviornment?