Thanks for not hurling insults in this reply as the first two posters. The first article is dated (i.e., not update with 5.8 changes) and the second article is equally wrong due to its narrow, also dated thinking.
If I over-provision RAM (to the point that the system never swaps even with high swappiness), then the matter is entirely moot. Good grief … is that not obvious. No matter what benefits arise under moderate or high memory pressure, there is zero benefit when there is zero swap. Period. Period. Period. Exclamation mark! And from what I can tell over-provisioned systems are common nowadays.
Now, lets look at the narrow statement, “The main incentive for swap nowadays is not to magically turn 16GiB of RAM into 32 GiB, but to make more efficient use of the installed, available RAM.” The only excuse for that statement is being woefully out-of-date. Regarding magic, I’m most familiar with Chromebooks and zRAM where Chromebooks configure zRAM as 2xRAM; then 4GB systems behave as if they have 8GB (or more), 8GB systems that behave as if they have 16GB (or more), etc. And the fact that Chromebooks employ swap as effective memory multiplier has much evidence; to be sure, Google does not think swap is merely “about making memory reclamation egalitarian and efficient.”
I think, in most cases, swap needs to be considered for the case in hand. For too-much-RAM-to-swap-anyhow systems, swap is moot. Duh. For very memory constrained systems with slow disks and a modern-ish CPU, then you want zRAM (slam dunk IF you are observing good compression). So, I think the references you cite are quaint. And if you look at the number of installed Linux’s (allowing that Chromebooks run Linux), the vast majority use of swap is for the memory multiplier of zRAM, not improving “memory reclamation”. Moreover, when the other Linux distros catch up to Fedora, those devices that zRAM benefits will more often be using swap for the magic RAM multiplier.
In the dark(er) ages of Linux swap, if you swapped more than a little bit, it was typical for the system to go into the toilet and never (or very slowly) come back (exactly how true now, I don’t know). Basically, what you could successfully swap was a handful of almost never needed anonymous pages.
I think claiming the purpose of swap is "making memory reclamation egalitarian and efficient” is reverse engineering a holy purpose for swap that matched its very limited practical use. In other words, it is pseudo-intellectual BS. The purpose of swap is expand the virtual memory of the machine … a purpose that was never realized very well on Linux (well until improved more recently).
The changes in Linux 5.8 (and possibly other changes) allowed zRAM configs of 2xRAM to be practical. I don’t test every use case in Linux, but from anecdotal evidence (for certain work loads), NVMe swap can, in effect, be a memory multiplier, too (i.e., another contradiction of the reverse engineered “purpose” of swap).
Now, on my system, w/o swap:
$ free -h
total used free shared buff/cache available
Mem: 31Gi 10Gi 3.9Gi 525Mi 19Gi 20Gi
It has been up quite a while, “free” has not exhausted, I’ve used vscode, firefox, chrome, vlc, etc., … that is, all that I ever do or will do (w/o a change in purpose in the machine), and what does it actually matter if a few dribbles of anonymous memory are swapped? It matters not.
The spirit of “have some amount of swap” (IMHO) is have just enough swap to cover the a dribble of almost never used anonymous pages. Since that advice was born from pseudo-intellectual BS, the advice is pseudo-intellectual BS.
Sometimes, no swap is OK. Sometimes, a lot of swap is best. And sometimes “some” swap is best (like a broken clock is right twice a day).
