|
|
Log in / Subscribe / Register

PC Gamer on the scx_horoscope scheduler

PC Gamer has run an amusing review of the scx_horoscope scheduler for Linux, which uses astrology to optimize scheduling decisions.

The scheduler is full of bizarre features, like its ability to perform real planetary calculations based on accurate geocentric planetary positions, lunar phase scheduling (the full moon gives a 1.4x boost to tasking, apparently) and "zodiac-based task classification".

That latter feature is easily one of my favourite bits. Specific planetary bodies "rule" over specific system tasks, so the Sun is in charge of critical system processes, the Moon (tied to emotions, of course) rules over interactive tasks, and Jupiter is assigned to memory-heavy applications, among others.



to post comments

April Fools Fodder?

Posted Jan 28, 2026 16:37 UTC (Wed) by cruff (subscriber, #7201) [Link]

Seems like something you'd see on April Fools Day. Does it modify its behavior if the computer is placed under a pyramid?

Effort

Posted Jan 28, 2026 17:07 UTC (Wed) by welinder (guest, #4699) [Link] (1 responses)

Maybe a bit too much spare time.

But, hey, if that scratches your itch go right ahead.
.

Effort

Posted Jan 29, 2026 1:43 UTC (Thu) by prittner (subscriber, #110534) [Link]

I believe it's "itch.io" not "itch go"

Amusing

Posted Jan 28, 2026 18:15 UTC (Wed) by jd (guest, #26381) [Link] (6 responses)

One big problem is that filesystems and schedulers tend to get developed in large numbers but the amount of theory behind them seems to be extremely limited.

Comedy schedulers like this actually are useful for understanding how much of a scheduler's efficiency comes simply from time not being wholly random and how much comes from the nature of that non-randomness. These are "placebo" schedulers in any such analysis, if you like.

I would love to see someone compile a complete set of schedulers that have either made it into the kernel or have been offered. (HP's heirarchical scheduler system plus their sample schedulers would be examples of proposed ideas that were never adopted, I think the Red Hat M:N scheduler briefly was). If these ideas were all placed alongside each other today in the modern kernel, it would be interesting to see how they'd rank - say, on overheads, throughput of high-priority tasks, average throughput, frequency of erratic behaviours, etc.

I doubt anyone has the time to actually do this, but it would allow us to see clearly how ideas across time compare (regardless of the limitations of the setting they were in at the time of implementation).

Amusing

Posted Jan 28, 2026 19:42 UTC (Wed) by eru (subscriber, #2753) [Link] (3 responses)

Wouldn't it be bizarre if the horoscope scheduler performed better than the mainstream ones on some workloads? Proof for astrology? :-)

Amusing

Posted Jan 29, 2026 6:13 UTC (Thu) by k8to (guest, #15413) [Link]

Oh, we already know astrology exists.

Amusing

Posted Jan 29, 2026 8:47 UTC (Thu) by leromarinvit (subscriber, #56850) [Link]

At least it would explain why some things only work if the stars align just right! :-)

Amusing

Posted Jan 29, 2026 9:22 UTC (Thu) by mjg59 (subscriber, #23239) [Link]

Knowing whether there are cases where pretty random scheduler decisions improve performance over decades of carefully tuned algorithms would certainly give insight into whether that work has been beneficial...

Amusing

Posted Jan 29, 2026 10:21 UTC (Thu) by paulj (subscriber, #341) [Link]

> Comedy schedulers like this actually are useful for understanding

That was surely the motivation behind this work. To help highlight how much (if any) of work in 'real' schedulers has (in net effect) been on the same plane as astrology.

Amusing

Posted Jan 29, 2026 18:53 UTC (Thu) by riking (subscriber, #95706) [Link]

"into the kernel" -- this is a sched_ext scheduler, so it will work on any system with sched_ext compiled in!

Pure gold...

Posted Jan 29, 2026 14:51 UTC (Thu) by csamuel (✭ supporter ✭, #2624) [Link] (2 responses)

Love this work. Also https://github.com/zampierilucas/scx_horoscope#acknowledg...

> "In space, no one can hear you schedule."

Pure gold...

Posted Jan 29, 2026 22:19 UTC (Thu) by CChittleborough (subscriber, #60775) [Link]

Pure gold...

Posted Jan 30, 2026 8:21 UTC (Fri) by vasvir (subscriber, #92389) [Link]

Also from the github page:

...using astrology to schedule CPU tasks is:

Scientifically dubious
Cosmically hilarious

Why only astrology?

Posted Jan 29, 2026 15:15 UTC (Thu) by GhePeU (subscriber, #56133) [Link] (8 responses)

Brilliant. But why only astrology? I want something like this but based on the local weather.

And a qdisc that matches nearby road traffic conditions.

Why only astrology?

Posted Jan 30, 2026 10:28 UTC (Fri) by intelfx (subscriber, #130118) [Link] (7 responses)

> And a qdisc that matches nearby road traffic conditions.

"Nothing can match the bandwidth of a 18-wheeler loaded with HDDs"

"But you can match the latency, tho'"

Why only astrology?

Posted Jan 30, 2026 14:27 UTC (Fri) by Wol (subscriber, #4433) [Link] (1 responses)

> "But you can match the latency, tho'"

Can you?

They've done a couple of real-life trials of IPoAC https://en.wikipedia.org/wiki/IP_over_Avian_Carriers, one in South Africa, one in Australia, and on both occasions the pigeon won ...

We'd probably get the same result again, even now ...

Cheers,
Wol

Why only astrology?

Posted Jan 31, 2026 13:07 UTC (Sat) by intelfx (subscriber, #130118) [Link]

> Can you?

"Matching the latency" of an 18-wheeler means making it worse, not better :-) Which, is exactly, if playing along GP's joke, what a qdisc that delays Internet traffic according to nearby road traffic conditions would be doing.

18-wheeler

Posted Jan 30, 2026 20:50 UTC (Fri) by cbushey (guest, #142134) [Link] (4 responses)

Wouldn't micro sd cards surpass the density of hard drives now?

18-wheeler

Posted Jan 30, 2026 21:10 UTC (Fri) by Wol (subscriber, #4433) [Link] (3 responses)

> Wouldn't micro sd cards surpass the density of hard drives now?

Probably, but price/TB they're lousy in comparison. What's the capacity of an HDD now? I've got two 8TBs, and they're SMALL!

I think you can get 1TB SD cards. And then of course you have all those assorted SSDs - you could probably send them by IPoAC, but losing packets would be a little pricey :-)

Cheers,
Wol

18-wheeler

Posted Feb 2, 2026 9:39 UTC (Mon) by taladar (subscriber, #68407) [Link]

microSD is up to 2TB actually and apparently for enterprise NVMe drives they are up to 128TB per drive available and 256TB per drive announced to be available in 2026.

18-wheeler

Posted Feb 4, 2026 4:45 UTC (Wed) by gdt (subscriber, #6284) [Link] (1 responses)

A 18-wheeler will move a lot of data, but often we don't want to just move data, we want to move it and then process it.

That is, the bandwidth to the storage matters. A Micro-SD card with the UHS-III specification interface has a maximum bandwidth of just under 5Gbps. A quick look at NVME SSDs suggests a benchmarked read bandwidth of 110Gbps.

The current generation of fast long-distance networks are running coherent-optical 400Gbps ethernet within one channel of 30-channel DWDM transmission. Previous generations were at 100Gbps and 10Gbps (as channels within 64-channel DWDM transmission).

Issues of TCP tuning to one side, for sequential transfers it's hard to see the argument for loading up the trailer if the bottleneck is the small length of 'network' from the CPU to the storage.

This gives a hint as to where scientific instruments have been heading: not writing sensor output to disk, but streaming that across a fast network to the supercomputing cluster which does the initial analysis and reduction of sensor output to observation data, then sequentially writing those observations to disk. Or putting that another way, the internet is part of the instrument.

18-wheeler

Posted Feb 4, 2026 8:05 UTC (Wed) by Wol (subscriber, #4433) [Link]

> That is, the bandwidth to the storage matters. A Micro-SD card with the UHS-III specification interface has a maximum bandwidth of just under 5Gbps. A quick look at NVME SSDs suggests a benchmarked read bandwidth of 110Gbps.

As does context :-) We were discussing IPoAC !!!

Cheers,
Wol


Copyright © 2026, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds