Alibaba Proposes A Group Balancer For The Linux Kernel Scheduler

With more organizations such as Alibaba configuring their servers to share CPU cores/resources among applications these days rather than exclusively assigning CPU cores to individual applications/tasks, the Chinese company is proposing a new "group balancer" for the Linux kernel scheduler.

The proposed Linux Group Balancer is focused on reducing resource conflicts when sharing resources among tasks. This balancer is focused on balancing groups of tasks across groups of CPU cores.

A "request for comments" on this Group Balancer (GB) was sent out today and was summed up as, "Thus what we need, is a way to ease confliction in share mode, make groups as exclusive as possible, to gain both performance and resource efficiency. The main idea of group balancer is to fulfill this requirement by balancing groups of tasks among groups of CPUs, consider this as a dynamic semi-exclusive mode. Just like balance the task among CPUs, now with GB a user can put CPU X,Y,Z into three partitions, and balance group A,B,C into these partition, to make them as exclusive as possible. The design is very likely to the numa balancing, task trigger work to settle it's group into a proper partition (minimum predicted load), then try migrate itself into it. To gradually settle groups into the most exclusively partition."

Alibaba's benchmarks on a 128 core CPU server found the Redis performance to benefit from 2~10% faster in the GB mode compared to the standard sharing mode.

We'll see what comes of this patch series.