Swap across multiple storage hardware

Posted Apr 7, 2025 18:09 UTC (Mon) by aeden (subscriber, #116597)
Parent article: Three ways to rework the swap subsystem

Recently I posed the following question: suppose a system has three distinct NVME SSDs. Should I create a swap partition on each device and `swapon` all of them (i.e. /dev/nvme), or should I create a mdadm RAID0 across those three swap devices and then `swapon` _that_ (i.e. /dev/md0). It seems the current best strategy's is the latter, since it looks like the swap devices are handled in a dumb way.

to post comments

Swap across multiple storage hardware

Posted Apr 10, 2025 14:51 UTC (Thu) by Wol (subscriber, #4433) [Link] (2 responses)

To the best of my knowledge, (by default) swap is allocated a priority based on its order in fstab. So you should list devices in order of speed.

Or you can explicitly assign a priority to each, and the kernel will stripe (aka create a raid-0) across all devices of equal priority. So no, actually creating your own raid-0 is not a good idea.

The other thing is there's two types of raid-0 - there's the striping you're thinking of, and there's sequential (which iirc is what swap does by default) where raid will fill one device before moving on to the next.

The only time raid'ing swap makes sense is if you do a raid-1, so the system is hardened against swap-device-failure.

Cheers,
Wol

Swap across multiple storage hardware

Posted Apr 22, 2025 7:00 UTC (Tue) by mathstuf (subscriber, #69389) [Link] (1 responses)

> To the best of my knowledge, (by default) swap is allocated a priority based on its order in fstab. So you should list devices in order of speed.

Is this just "mount order"? How else are manually-mounted swap devices or systemd `.mount` swap units prioritized? It feels strange (to me) that the kernel would read `/etc/fstab` and make such policy decisions based on it.

Swap across multiple storage hardware

Posted Apr 22, 2025 18:52 UTC (Tue) by edgewood (subscriber, #1123) [Link]

"swapon" on the command line accepts a priority option, and passes that priority to the kernel. "pri=" in fstab does the same thing, and systemd presumably also does the same. So the kernel tracks the priority, no matter where it came from originally.

The swapon(2) man page describes what the kernel does with priorities if there is more than one swap device.