[SOLVED] swap or swapfile?

[xanderal]

Hi, I'm thinking of giving gentoo a go but had a question first:
I have 16g ram and a 512 GB SSD M.2 PCIe NVMe Opal2 on my new laptop.
I haven't needed swap yet but always created a swap partition anyways.
But now I'm thinking of instead creating a swapfile after the installation or not doing either.
Any thoughts?

[krinn]

i think that you should try another distro ; gentoo do compile and takes cpu time, something laptop are at best medium at doing the task with power and heat issue.
while many gentoo users do use gentoo with a laptop, i think only hardcore users has start learning gentoo with one.
you could get the max out of a laptop because of gentoo tunning, but using one to "play/discover" gentoo won't be the best first experience for a new user.

in case you want only "throughts" about the swapfile, the answer is : don't care, you can do all you want from partition to file anytime you wish or need.

[NeddySeagoon]

xan73r4l,

As its a laptop, make a swap partition equal to your RAM size.
Its used for suspend to disk. If that's not a feature you will use, you can add a swapfile late.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[Irre]

I have no swap partition, but I attach an external disk when I need to rebuild gcc, llvm etc.

[fedeliallalinea]

I also repeat what NeddySeagoon said, creates a swap partition of the same size as the ram for suspend to disk
_________________
Questions are guaranteed in life; Answers aren't.

[axl]

Personally I do not use swap to suspend to disk, but sometimes, some of the systems actually need swap because they run out of memory. PS3 for instance only has 256 MB of memory. You can get decent performance out of a combination of zswap and zram, both provided by kernel.

[1clue]

I have an 8-core atom system. It has 16gb and an ssd, and also a spinner. So the same type of thing in some ways? Only my atom is server hardware and does not support suspend to RAM or sleep.

I usually have a swap partition the size of RAM on each drive. In this case it's a small ssd and I would rather prevent wear, so I just have swap on the spinner. That said, I NEVER hit swap, and if I notice that I do, I get more RAM if possible. The idea of swap was to prevent system failure in an out-of-memory situation. Not to be a regular part of every day, like some people seem to think.

That said, and the reason I use swap at all, is because I use a lot of tmpfs partitions. /tmp and /var/tmp for example. /var/tmp is 4g in my case because I don't use any GUI items. If you use a GUI you will want it to be larger. If you want to build Firefox or such it should be significantly larger.

tmpfs makes your system much faster by using RAM as disk for those locations. /tmp is a system-wide general purpose temp directory. Everything that goes in there can be "forgotten" on reboot without harm. /var/tmp (specifically /var/tmp/portage) is used by portage as compile space, and likewise can be forgotten between boots or even between builds.

The tmpfs system uses the same hardware on your CPU that is used for swap. In the event of your regular RAM + your tmpfs mounts being greater than physical RAM, the VM hardware decides what is least likely to be used and swaps it out. It could be some part of RAM or it could be some part of your tmpfs mounts, or whatever else is using the VM mechanism. The thing is, this mechanism is very good at making these decisions, so it makes it less likely that your system actually wants back whatever was swapped out.

Right at the moment, with a week or so of uptime, my gentoo atom system has 9.3 GiB free, 0B used on swap.

My swap partition is serving the same function as a motorcycle helmet. You really should have it but you hope you never have to use it. If you must use it, it can save you a lot of pain.

In some cases I have had a system in service so long that I needed my swap partition and even had to build another. That's when I use the regular file on another partition, because in those situations it may not be feasible to add storage or modify existing storage. Also note that in this situation I set up the swap and immediately begin working on a physical upgrade to a new system.

[Fitzcarraldo]

xan73r4l,

Just for information, it is also possible to suspend to disk (a.k.a. hibernate) using a swapfile instead of a swap partition:

https://wiki.gentoo.org/wiki/Suspend_and_hibernate#Suspend_to_disk_with_swapfile

(Ubuntu, for example, installs with a swapfile rather than a swap partition, so I configured my family's PC running Lubuntu to be able to hibernate to the swapfile.)
_________________
Clevo W230SS: amd64, VIDEO_CARDS="intel modesetting nvidia".
Compal NBLB2: ~amd64, xf86-video-ati. Dual boot Win 7 Pro 64-bit.
OpenRC udev elogind & KDE on both.
Fitzcarraldo's blog

[NeddySeagoon]

Fitzcarraldo,

Thank you. I was aware that suspend to a file was a work in progress.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[xanderal]

Sorry for taking so long to come around again.
Thank you for the replies, that helped a lot, especially @1clue

[1clue]

Sorry to be pedantic about this.

My habit of having a swap partition on every disk doesn't mean I use that much swap. As I said before, the existence and "use" of swap is for me sort of like a motorcycle helmet.

Also having a swap partition on every disk which is the size of RAM is way overkill. You should never turn all that on, or at least you should never plan to use it all at the same time.

One thing that has happened to me is that I needed disk space for some reason, and at that point I'd delete the swap partition and then repurpose that space. Or I'll just leave it and let the disk firmware use that space to patch bad sectors. This is less important as disks get bigger, because the tiny bit of space for a swap partition is inconsequential compared to the size of the disk.

The point is that you never know when you need an actual physical partition, or when the conditions of your use case will change. In my opinion it's better to have something reserved.

[axl]

I love the idea of swap as a motorcycle helmet. I get what you are saying perfectly. But did you know they invented compressed ram as swap? So instead of using a disk to create more space out of nothing, you get to try to compress some of the ram u're using, into compressed ram. and if don't leverage your arch too much it's way more elegant than swap on disk. and WAY faster. I mean, how hard can it be for the kernel to run a gzip in ram instead of a cp on a very slow disk? think about it.

[1clue]

I've heard of compressed RAM as disk.

RAM is cheaper than it ever has been. Absolutely every scheme you can come up with to store swap somewhere, it won't be as fast as just not needing swap in the first place. It won't be as simple either.

What I'm saying is that using compressed RAM as a disk for swap may be legitimate, but in my case it makes no difference. A plain-old spinner is fine for me. Because literally, if I find out that my system has swapped even once, I will go out and buy more RAM. It just takes one time, unless it was a programming error that gobbled up all my RAM and kept going. I'm sticking with my motorcycle helmet.

Linux does something interesting with free memory. It caches things that you've read in, it keeps track of that information. As long as you haven't officially allocated that memory for an app, that information is there. A system with large amounts of RAM will get faster the longer it's up, because eventually Linux knows every part of the disk you've been hitting and short-circuits a disk read using the cache.

[Lars]

I have also 16GB or RAM and an 128GB SSD for system. I don't use swap for something like hibernate. Also I have no swap partition nor swap file.

I have actived zram for swap, /tmp and also /var/tmp/portage. Sounds stupid, but believe me, it works like a RAM doubler. With a fast CPU like my i7 6700k it works like a charm.
Swap use some compressed space in RAM, therefore more space for other stuff. It writes bytes faster to swap-RAM with compression than to every other disk a disk. Also /tmp and /var/tmp/portage are good for that.

So IMHO the SSD is under less stress to write things to disk, which we never use for longer.

There is only two stupid things, first zram is not so dynamic as expected. I need to give a upper limit for space. Second, never store some important data to /tmp nor /var/tmp/portage, because after reboot everything here is gone.

zram swap 4gb
zram /tmp 8gb
zram /var/tmp/portage 8gb

These 20gb need less real RAM as the real 16gb most the time. (algo: lz4)

One last point: Every byte you write to ssd disk, kills it a little bit.
_________________

[pjp]

Is there a practical point where zram makes sense? Given 16GB of actual RAM, reserving 4GB for zram means the system would start swapping when RAM usage hits 12GB, no? Presuming that to be true, then that same system without zram would likely not have gone to swap at all or as soon. That approach sounds like zram being useful as a canary in a coal mine to warn when it might be time to buy more RAM, or reconsider how the system is used.

How is over subscribing handled (as in Lars' example of 16GB of RAM, 20GB of subscribed zram)? In my experience, over subscribing tends to be a dangerous practice where data integrity and system availability are important.
_________________
Quis separabit? Quo animo?

[Ant P.]

[pjp]

Interesting. I'll have to look into it, thanks. The "worse under pressure" seems important to understand regarding use / don't use.
_________________
Quis separabit? Quo animo?

[axl]

2 things.

first, dont use zram devices to write stuff. because they are slower then the ram counter devices. use shm or ramfs instead.

and the second thing, dont think of compressed ram as you would have imagine old type swap. it's different. first of all you use zram but also zswap together. And second of all, don't think of the swap as something smaller than your huge ram. but think of it as double or triple the size of your ram. zswap sometimes manages to compress stuff 3 times over. So, in an ideal situation, when you compress all your ram, your final swap uncompressed space will be 3 times bigger. you can define it a hundred times bigger, if it can be compressed it will fit in there, if not, the kernel will complain that it ran out of memory. But the point is, the swap space has to be considerably larger than your actual ram. because ram is now compressed pages and swap is the real ram where things are uncompressed in fact. its a little bit counter-intuitive, I know.

Lets say my computer has 24 Gb of ram. I make 3 zram partitions of 24Gb each, activate as ram. I ran a script, that inits 64 variables equal to 1Gb in zeros. My script shows up in top with like 99% percent memory usage and 40Gb of swap. I don't have 64Gb of memory in my system. But my system is able to compress those zeroes very fast into compressed pages. In the real world, maybe the stuff in your memory is not as easy to compress as zeroes, however, there are real life applications where this design works very very well. I find VM's work very well in this type of environment.

[1clue]

[axl]

I used 24 Gb arbitrary to make a point. The reason I don't buy more ram is that I cannot fit more inside

And no, I'm clearly not saying that whatever quantity of compressed ram is better than actual ram. That would be silly. What I am saying is that certain applications could benefit more from compressed swap rather than swap on disk. This stuff allows you multi-thread compress/decompress, it's using ram as backend, you can even customize the compressor to suit your needs.

Just give it a go, instead of making suppositions. You could test the stuff with your workload.

[axl]

KiB Mem : 30878088 total, 17478356 free, 12890592 used, 509140 buff/cache
KiB Swap: 94371824 total, 79174280 free, 15197548 used. 17557476 avail Mem


kernel reports 13.5 Gb used or in cache. and 15 Gb of swap. maybe my math is wrong. or this script. I'm sorry I use mostly php for testing things. but here is the script:

[axl]

to be honest.

on the other hand I am considering and should be testing the idea of double compression. zram is one level of compression. zswap is another. I honestly dont know if I need both, or if they know about each other and compliment each other... I kept my eye on the idle cpu %. when you want it to be 90% or upwards it is ok.

and again to be honest, not all data is just zeroes. if you really want to see what happens just read files from distfiles as variables into memory until something crashes. it's one thing to read zeroes, whole another thing to read stuff that was previously compressed and cannot be compressed anymore. and put that forcefully into ram an see what happens.

I cannot honestly do that to my systems... just to test. It's like abusing on purpose when you know you don't have to.

I'm sure different people have different ideas about how to treat a system as a whole.

[axl]

KiB Mem : 30878088 total, 17232164 free, 13192144 used, 453780 buff/cache
KiB Swap: 94371824 total, 79794848 free, 14576980 used. 17264384 avail Mem


I hour later. Do I trust my kernel to manage it all? I do!