Processing power doubles endlessly, it seems, and become less and less expensive. Applications, at the same time, are developed with less of resources efficiency in mind and thus demand more and more resources. And so the spiral grows. While in most cases it is not an issue and performance is not impacted in some instances you have to squeeze performance out of your environment. Possible scenarios include:
- VMs or applications where traffic packets are relatively large.
- VMs/apps with multiple concurrent connections, where network traffic moves between guest VMs on the same host, guest VMs across hosts or to the hosts themselves, or from guest VMs to an external system.
- To provide relief to vNIC RX packet drop rate increases (per KB 4540 - Handling packet drops from vNICs in AHV) where CPU contention of guest/host has already been ruled out.
For those particular instances, there is a solution.
This is not a “turbo” button. Do not use this for all your VMs.
Rule out other sources of contention first.
Upgrade to the lastest AOS and VirtIO and monitor the performance of your environment afterwards.