Linux 7.4 Rewrites the Rules of Kernel Efficiency With a 100x Workqueue Boost

Linux version 7.4 is working on making an important change to its performance that will be useful in high-end servers. The change involves improving how Linux flushes its workqueues, which are used by the Linux kernel to run tasks asynchronously.

In the past, if Linux flushed its workqueue, it had to check all the workqueues for all the CPUs in the system even though most of those CPUs had nothing to do. The new change makes Linux look at workqueues only when they have been used after the last flush.

Longtime Linux developer Tejun Heo implemented the change, which has already landed in the Linux 7.4 development code. The improvement targets unnecessary overhead, particularly on systems with many CPU cores.

Benchmarks Reveal a Dramatic Performance Jump

The early numbers are impressive. On a 192 CPU, two node system, flushing an idle per CPU workqueue increased from 40,000 operations per second to 4.2 million. An idle unbound workqueue improved from 450,000 to 4.3 million operations per second.

This demonstrates the amount of overhead involved in bookkeeping rather than actual processing. This improvement is especially helpful for the Linux file system, XFS, which may flush its workqueue with each log force operation.

Still, the improvement is dependent on the workload. Experiments conducted on extremely active queues showed that per CPU workqueues were not improved in any way; unbound workqueues, however, became 15 to 20 percent slower.

Our View

This is one way to illustrate why kernel programming is vital even if there is nothing new that pops up on the user interface. The reason why it is necessary to get rid of any unnecessary overhead is that it will lead to efficiency when working with lots of cores on the server in storage operations.

Want to know more about the improvements that are going to be made to Linux performance? Share this post with your fellow Linux fans.