top | item 46868983

(no title)

stuxnet79 | 26 days ago

This scheduler takes advantage of sched_ext which is a relatively recent kernel capability. I really thought that EEVDF was the last word on scheduling in Linux, at least for the short-medium term.

Apart from trolling I am really curious what other useful functionality sched_ext enables. One of the primary reasons CFS was replaced with EEVDF was because it has a better set of defaults which don't require tuning / patching which is exactly what sched_ext simplifies.

discuss

order

xevrem|26 days ago

the main reason why sched_ext is taking off is that EEVDF and CFS before it are -general- cpu schedulers, so while they work well for most tasks, they don't work as well as they could in certain situations like heavy server loads, low-latency loads (like multimedia production and gaming), or security sensitive loads. These schedulers are built and or specifically tunable to these types of loads, such as scx_lavd, scx_bpfland, and scx_cosmos. They each approach things differently, such as using core compaction, sharing or unsharing per core task queues, being way more cache context aware, keeping hot cores hot, etc. And the perf boost in these specific contexts is very real. Heck, Meta started using scx_lavd for better task load balencing over EEVDF on their servers.