I don’t really need swap until I need to compile blender… Need somewhere around 64Gb (with compression) to deal with the peaks
Once tortured myself with having to go through…
The Android build system. Booo, scary.
I only wanted a LineageOS boot image built. The recommended specs list 64GiB RAM. Figured wouldn’t apply to me until I tried to build a target literally called “nothing”. It just used up all of the RAM and OOMed about 2 minutes in, building the “build cache” or whatever the fuck it needs to do to “warm up”.
After a few fucktons of swapspace, turning off some of my daemons, some sysctl fuckery (there were options for limiting process virtual memory size IIRC? I put it all the way up) and
taskset -p 0-15:1(use logical CPUs 0, 2, 4…), I had my laptop allocating 50 something GiB. I have 16GB (decimal) physical RAM installed and I consider that already way too much for my usual needs.After about half an hour of the build system build systeming, it just stopped doing that then built the kernel in 4 minutes with everything else in some time too.
Yes, I have tried just running make directly. That just resulted in an image that did nothing but toggle on and off an empty (no percentage) charging circle via power button.
“I’m not a friggin bear, I don’t hibernate!”
Swap and zram saved my R and Python Dana analysis script a couple of times.
Everyone defending swap like crazy in the comments… so let me give a counter argument specifically for desktop:
Many years ago I stopped using swap on all my computers (I exclusively use linux) and the reason is: every time anything regarding interactivity went to swap, my system would freeze and literally become unusable for a really long time until I killed something. Since I got tired if that behaviour I decided I prefer for random shit to die but my computer still be usable.
It wasn’t very often, but I don’t get OOM killer often either (as in I don’t remember the last time it happened and I use linux every day for both work and fun), and the freezing situation was way more annoying, to the point I sometimes ended up cold switching off the computer. I’m surprised on how many people run out of memory constantly. I guess you all are just using linux for servers and have crazy processing of infinite data…
Or shotcut. I once ran it on 8GB, minimal arch with no swap. Well.
And I’m not even trying to build android on my new relatively modern laptop. It won’t work.
Were you using a HDD for swap? It’s nearly unnoticeable on an NVMe drive.
yep it was HDD, but with the low amount of times I have any OOM issue, I’m way more worried about NVMe flash burning than a random process being killed once every blue moon.
In fact I didn’t mention, but since the advent of flash drives what I’ve done is trying to locate heavy disk writing apps and find alternatives for them, amid disabling many logs.
For swap, it hunts me to think some automatic process I don’t have control over will burn those sweet erase cycles without me noticing just because someone programmed it poorly. I’m way more happy knowing I have some control over it.
My built-in Devil’s Advocate has to ask if that’s really true? Have you got a citation newer than 2 decades old? Frankly, i find it difficult to believe that anyone wouldn’t like to use electron software, for the simple reason that making your SSD suffer is fun.
Edit: in 30 years of administering Linux and (previously) Solaris sytems i’ve never once had to change HDDs or SSDs, so claims of torture are, i opine, exaggerated. i’ll also claim that keeping electron software out of the filesystem is not good for your hygiene.
Also, while Linux does run without Swap, it’s like running gears without grease.
That’s some really great luck you have, over the 15 years or so I have been running my Linux server I have had to replace the SSD twice(likely due to gentoo and compiling everything all the time) and had 2 hdds in my raid cluster need replacement. Here’s hoping my NAS has your luck
To be clear: what i wrote is not actually something real, it was a joke i made by modifying a comment from a thread on github gist, i am actually surprised it wad credible enough that someone would reply seriously xD
https://gist.github.com/joshenders/c4960cec9c63a7b7d68ffa9543356c43
Definitely bit the onion on this one, it did seem remarkable lol though I just pulled a 16 year old drive out of my Son’s PC(that was technically still working however slowly) hence the semi plausibility on my end lol
lmao, 16 years is a lot! That drive can drive
I manage a cluster that makes heavy use of SSDs in RAID10 configuration. Those things are constantly degrading and having to be swapped out. I’ve started a drawer of faulted drives that I don’t have the heart you throw out.
swapfile gang
.rw------- root root 4.0 GB Mon Jun 1 04:10:37 2026 /var/swapDo you hibernate? 🐻
No, I have a 13-second boot time
Bah, 56 seconds ought to be enough for anyone
Yes, swap partitions are shit.
The real hotness is swap drives.
Just the use that your old 64GB SSDs were waiting for!
(Fun fact: in Linux, if you have multiple swap partitions mounted, the kernel will automatically use striping (like RAID) in order to read and write from them as fast as possible. If you have multiple small SSDs used for swap, the read/write speed from swap can actually become pretty fast.)
do you encrypt your swap drive? would this work with LUKS?
It works fine. The only partition that needs to be unencrypted on a modern hardrive is the esp/boot partition.
I do not. No idea how this interacts with swap encryption.
How many SSDs do you need for it to match the speed of RAM? Can I use this to make cheap RAM?
You’re going to run into issues with your motherboard’s SATA or NVME (or PCIE) controller throughput well before you get close to actual RAM speeds.
Once you reach that bottleneck, the speed won’t increase anymore when you add more drives. The motherboard architecture just doesn’t support accessing any kind of storage device at that speed.
I just download more ram. Know your tech!!
I did a (very) quick search and they should just sum up, so for example a raid of 2 sata SSD with 64gb that can reach 500MB/s will have a total speed of 1GB/s,
DDR3 ram can get up to 17GB/s so we would need 34 drives to get DDR3 speeds (not very fast BUT you would be the person with most ““RAM”” in the city without being a data center)
And here I am using swap for my self hosted services, some of which use ridiculous amount of ram.
> linuxmemes
> macos screenshot
ಠ_ಠ
Shit did I botch it? I made this on my phone and the toot I copied the script from was just text. Wanted to make it look more interesting so I found this site that let you make the windows. I just went with what I thought looked best.
That makes it funnier ngl
Is this tongue in cheek i swear I cannot tell any longer.
Some swap is beneficial
Like a lot of memes, they exaggerate for humor. This meme was originally about math.
It’s based on a common meme template. Yeah it’s tongue in cheek and I guess a lot of commenters can’t tell given this is a memes comm
I hope the phone is running PostmarketOS ;)
TBH I’m sure you could configure a linux terminal to look like that, but yeah it looks very default macos to me
Idk. KDE has themes like this but yeah I get it
i’m sure someone has made a KDE theme or a Gnome theme that replicates MacOS, let’s pretend that’s what’s happened.
I used to be this guy. We have huge amount of memory in our storage server. Why would we need swap? Well, once the system started killing programs I read up on swap and how the system handles its caches, especially ZFS ARC.
Short version is that while the system can evict its caches syncronously, it can only async notify ZFS that it should also shrink its caches, but the out-of-memory killer will start killing before ZFS and its kernel counterpart can start to free up memory.
A tiny bit of swap completely solved this, since swapping is sync for the kernel and the OOM killer will wait for it.
I found that with swap the applications would slow to a crawl and then basically lock up entirely until I hauled my ass over and manually power cycled the machine.
As well as oom situations, swapping allows the memory to be used more effectively by swapping out anonymous pages that aren’t likely to be used but can’t be dropped.
I sometimes cannot do hibernate when I did a scrub on my ZFS. But the first try always frees up the swap and so the second attempts works every time.
No
It turns out the computer mostly freezes once you go past 200% RAM used.
what is using so much ram?
16GB are my usual breakfast, I’ll fill that quickly after booting… I’m happy that I got 64GB, though even that is often used up, thus I sometimes wish for 128Gb though (and for running somewhat capable local LLMs 512GB would be even better…).
Bottomline: ypu can basically never have enough RAM…
Trying to ray-trace a 512,000x256,000 image using Chunky. I had to split it into 16 parts to reduce the RAM to a sane amount (in 2018, it may have been optimised since).
I was trying to find the limit for a single png, and I think I found it.
I’d guess that there are probably utilities to export the world file to POV-Ray or Blender or something. They might be able to scale better.
There are good use cases for swap. I work with very performance sensitive memory hogging applications that tend to process in bursts as incoming batched data arrives. We also have a lot of required background “security” applications.
Having the idle applications move out to swap so that a significant amount of free RAM to allow for the active, real-time applications to expand into is a major boon. RAM is expensive now, so it’s become even more critical to not waste it.
But even in self-hosting at home I take advantage of it since I have a limited number of servers with limited RAM and a bunch of applications and/or docker containers running, many of which don’t need to be active at all times, but if they were all active, would have trouble with the limited RAM. Swap allows for inactive applications to free up RAM for the active ones and leave enough available at all times to allow an application to quickly spin up new jobs.
Anyway, sure, for a simple desktop computer without a bunch of junk background applications running for “security” or whatever, and an overabundance of RAM installed, it’s probably not useful. But for high performance, memory hungry applications with lots of background stuff running, it can be a big help.
That said, there are use cases where the opposite is useful and using a RAM drive is better than writing to disk for cache and temporary logs when you have enough RAM. My opnsense router for instance uses a RAM drive to store cache and logs (logs are mirrored to the NAS for long term storage, so just the local copy is lost on reboot). That system has enough RAM and I’d rather reduce disk wear, and I don’t plan to add too many other applications to that server that it would run low.
Actually, there are more than these. For example, hibernation and/or hybrid suspension. It’s, AFAIK, mandatory to have a swap partition to be able to hibernate and/or hybrid suspend your device.
Its a meme built on stuff like this:
https://gist.github.com/joshenders/c4960cec9c63a7b7d68ffa9543356c43
I also found it useful to do some one-off batch processing of a big file without thinking of chunking and doing too complicated things. Imagine something like a 60GB file in a custom format that need to be deduplicated. It’s easier to just load it all up in memory and do run the algorithm 2-3x slower using a 50GB swapfile than spend hours to complicate the algorithm.
Yep. That’s why it can be nice to just have like an old 1TB drive completely used for swap. Things might get slow once it exceeds your physical memory, but at least it will eventually finish successfully – no out of memory crashes here!
People still use swap partitions instead of swap files?
This. Swap files on LUKS encrypted partition. Also, the swapspace daemon!
I’ve seen so many setups where people encrypt their data but forget to encrypt their swap.
what’s the problem? it’s not like you dump the entire contents of your system’s ram to disk any time use hibernate /s
Even without hibernation, data in apps you have open will end up in the swap if your system is ever RAM-constrained.
Generally it only happens to programs you keep running in the background and might not use for a long time that also don’t have much background activity. For example password managers.
It depends on what you set your swappiness to, too. I think Linux distros still have pretty aggressive defaults that swap more than they probably need to.
Don’t you need swap partitions to be able to use hibernation?
Been using a swap file in a luks partition for years. I hibernate very regularly.
Nah, you can use
resume_offsetinstead.Here’s an example: https://forum.endeavouros.com/t/how-to-hibernate-with-btrfs-swapfile-luks2-systemd-boot-dracut/55620/2
Nope, I hibernate with a swap file and it’s in a LUKS partition. Although IIRC that whole setup requires me to use systemd boot.
I do the same and use ugrd. I think dracut would also work.
You just set the resume kernel parameter to the decrypted memory device with the swapfile offset.
As long as the initrd attempts a resume after it decrypted all volumes it’ll work, and the initrd doesn’t even need to understand why and what was resumed from
It’s doable with a swap file instead. But some filesystems are more supportive than others.
For example BTRFS doesn’t support swapfiles over multiple devices (so my raid1c3 spanning 3 differently sized disks for example) and also swapfiles must be completely pre-allocated.
And no matter the filesystem you need to manually tell the kernel device and physical offset for the file via kernel parameters. Which can again be easier on some than on others (also causing the no-multiple-device filesystem problem mentioned above.
btrfs’s handling makes sense. The kernel talks to the disk directly, the file is just a pointer for what range it uses, and a placeholder for the filesystem. Something like raid, balancing, cow, all won’t work because the kernel doesn’t even go through btrfs to change the swap"file".
If you want a proper file as swapfile, with the corresponding overhead, you can make a regular file and mount it via loopback then use it as a swap device.
I don’t use hibernation so I’m not sure.
So you shut your pc down like a caveman when you do hardware upgrades or repairs or move apartments?
Yeah I shut down every day. It boots up fast enough that it’s really not an issue.
that’s what trixie’s netinst defaults to when i set up new desktops using encrypted lvm… root and swap volumes inside that, unlocking together at boot. my use doesn’t hit swap much, so it’s nbd. i just leave it like it is. only where an installer doesn’t support encrypted swap partitions do i use a file or ram-based swap instead.
It’s not as bad with LVM since LVM is more flexible. A swap file does still make it easier to change the size as needed though. If you need more swap, you can spin up a new swap file pretty much instantly while the system is running, without having to repartition. You can have both a swap partition and a swap file active at the same time though.
On systems where the data is encrypted, also encrypting the swap makes sense so I’m glad Debian does that.
Definitely configure swap, but I don’t think there is much benefit of a swap partition over a swap file these days and swap files are trivial to configure.
