My Debian stretch system has a lot of RAM. Currently, it uses an encrypted swap partition on the SSD, but under normal conditions I don't ever expect lack of memory to cause it to be used.

But turning off swap completely is a bad idea according to the comments of Snakedoc here:

Just note, it's never "safe" to fully disable swap. Even with systems
that have 96+ GB of ram, running at least 512MB of swap is a good
safety measure. It gives your system time to kill unnecessary
processes if something runs amuck and eats up all the ram
unexpectedly... the alternative with swap disabled is to kernel panic.

How to safely turn off swap permanently and reclaim the space? (on Debian Jessie)

So I thought I might reserve some of the ram, say 0.5G for a 'RAM disk', and use that as a swap file, saving writes to the SSD.

How do I do that? Is it a bad idea for some reason?

I wouldn't say its reason is stability, it is a common misconception. If a system ram + swap is full, you will find actually the same symptoms as without swap:

1. Unexplained, big slowdowns (the system locks the processes asking for ram up for a while, maybe somebody frees the ram required for them)


2. Processes killed by the OOM (out of memory) killer.

The only difference that without swap, these symptoms happen suddenly, while with swap, they happen only after the swap is filled out. Thus, you have a much bigger delay time before your system begins to be perfectly unresponsive.

This is the first reason, why is it useful to have swap.

The second is lesser known, but actually it is much more important:

In general, having swap doesn't slow your machine down, it accelerates it, even in you have far than enough RAM.

Its reason is the following: what does your system with the RAM actually not used by your processes? They are used to cache your hard disks, making your system faster with it.

If you have swap, you essentially allows your system to swap out the really rarely memory pages, and use their space for disk cache. How fasten do they happen, it depends on your VM settings and heuristics, but they are really good (as far I know, linux is far the best in this sense).

If your hard disks are on SSD, doesn't really influnece these both arguments (although having much faster and zero seek-time disks of course accelerates everything).

I mostly use the old golden rule for swap partitions: I use swap 2x more as my RAM. This is for to avoid swap fragmentation (blocks on the swap file can just so fragment as files on a filesystem). In your case, where you have a tremendous amount of RAM, but costly hard disk with zero seek time, swap fragmentation is not an issue, thus maybe also lesser would be enough. I don't know, how many SSD do you have, maybe 1x or even 0.5x times of your RAM site would be enough.

In my opinion, this "have at least 512M swap" advice is bad, the amount of swap you use have to depend on your system parameters (disk cost, RAM amount, disk amount, hard disk seek time, etc). This 512M is not a "golden rule", it is a "magic number", which is meaningless if we use it in general.

Using an HDD for swap on an SSD system is a bad idea. Not only it's slower in speed, it has much longer access time, is far more power hungry and possibly less reliable than a modern SSD. Nowadays SSDs have 5 years warranty and can easily last for 8-10 years even if you write hundreds of GBs to it every day

But it'd be even much better to use zram, possibly along with a small backup swap partition on the SSD. Various Linux distros as well as ChromeOS and Android and have enabled zram by default for years. There's not even the conventional swap by default on Android. Even Windows and macOS nowadays uses compressed RAM before resorting to swap

Since the CPU can decompress data much faster than the HDD, you can clearly feel the system responsiveness once the system begins to swap. The difference is not that inherent on SSDs, but according to the benchmark results here zram with the default lzo algorithm still seems faster. You can easily change to lz4 algorithm which is nearly 5 times faster than lzo with a slightly lower compression ratio

There's also zswap which uses a different mechanism, but still stores compressed data on RAM

Further reading

• https://wiki.debian.org/ZRam


• https://wiki.archlinux.org/index.php/improving_performance#RAM_and_swap


• zram vs zswap vs zcache Ultimate guide: when to use which one


• Linux, SSD and swap


• Difference between ZRAM and ZSWAP


• https://www.kernel.org/doc/Documentation/blockdev/zram.txt


• https://wiki.gentoo.org/wiki/Zram

Given the answer by @peterh, probably a good solution is to stick any old HDD in the machine and use it as scratch space, including swap. That way all ram can be available for the system, even the last 0.5G. A laptop HDD likely would be more energy efficient than a 3.5" drive.