Dynamic processor mapping for virtual machine network traffic queues

摘要

An algorithm for dynamically adjusting the number of processors servicing Virtual Machine Queues (VMQ) and the mapping of the VMQ to the processors based on network load and processor usage in the system. The algorithm determines the total load on a processor and depending on whether the total load exceeds or falls below a threshold respectively, the algorithm moves at least one of the VMQs to a different processor based on certain criteria such as whether the destination processor is the home processor to the VMQ or whether it shares a common NUMA node with the VMQ. By doing so, better I/O throughput and lower power consumption can be achieved.

主权项

1. A method of dynamically adjusting a network traffic queue associated with at least one virtual machine queue to processor mapping comprising the steps of:
determining whether a first load comprising a first virtual machine queue associated with a first virtual machine is present on a first processor; determining whether the first load exceeds an upper threshold for processor load on the first processor; determining a first resulting load on a second processor based on the first virtual machine queue and a second load on the second processor; moving the first virtual machine queue to the second processor when the first load exceeds the upper threshold and the first resulting load does not exceed the upper threshold for the second processor, wherein the second processor is a next physically closest processor to an electronic path upon which the first virtual machine queue resides; and wherein the proceeding steps are performed by at least one processor.