OLVM: Explanation of Virtual Interface Multi Queue (Doc ID 2947548.1)

Last updated on JUNE 11, 2023

Applies to:

Goal

What is Multi-Queue:

By default, each virtual interface has only 1 queue, so network-related applications running in a guest with multiple vCPUs will not benefit from those same vCPUs.
This became a bottleneck when a single CPU can generate traffic faster than a single virtio-net queue can consume it, this is more prevalent in intra-host networking where the physical adapter speed is not a factor.
This issue was solved with the introduction of multi-queue. With multiple queues, each vCPU can handle traffic with its own queue.

What are the defaults in OLVM:

• If Multi-Queue is enabled, each interface will have a maximum of 4 queues
• The number of queues will never be bigger than the number of vCPUs (VNICs in a VM with only 2 vCPUs will only have 2 queues)
• Since the maximum is 4, a VNIC in a VM with 64 vCPUs will only have 4 queues. 

What are the drawbacks:

• More queues, more CPU Load in the Guest VM
• Increasing the number of queues above 4 may have adverse effects on a VM already starving for CPU
• Increasing the number of queues above 4 in a VM with plenty of CPUs available may improve intra-host networking performance or even reduce the number of dropped packets 

Solution

My Oracle Support provides customers with access to over a million knowledge articles and a vibrant support community of peers and Oracle experts.