To swappiness or not to swappiness, that is the question...

[trytip]

[ off-topic discussion split from support request viewtopic.php?f=46&t=330146 and moved here ]

Gotta fixing now I hope, do I need to update grub after the fstab fix?

i just find out swapiness for SSD is better set to 1 instead of 5 or 10
https://sites.google.com/site/guidetoop ... ring-a-ssd

[pbear]

Gotta fixing now I hope, do I need to update grub after the fstab fix?

i just find out swapiness for SSD is better set to 1 instead of 5 or 10
https://sites.google.com/site/guidetoop ... ring-a-ssd

That's nonsense. SSDs don't need to be coddled. They're now more durable than spinning drives.
All choking swappiness to the bare minimum does is hamstring the system if it needs to use swap.

For more on swappiness, see gm10's general discussion of the issue.

[Pjotr]

i just find out swapiness for SSD is better set to 1 instead of 5 or 10
https://sites.google.com/site/guidetoop ... ring-a-ssd

That's nonsense. SSDs don't need to be coddled. They're now more durable than spinning drives.
All choking swappiness to the bare minimum does is hamstring the system if it needs to use swap.

For more on swappiness, see gm10's general discussion of the issue.

As you can see in gm10's contribution on this issue, to which you refer, you see that his viewpoint on decreasing swappiness is ultimately this:

for the average user, I won't make any statements as to what is best for I honestly do not know what the average Linux user does on their computer.
(.....) Either way, the best advice to give is to simply add enough RAM to your device and thus avoid the question from even posing itself.

viewtopic.php?p=1699701#p1699701

In my personal experience, decreasing swappiness noticeably increases responsiveness. More or less as promised by the author of this article:
https://rudd-o.com/linux-and-free-softw ... o-fix-that

Now, what's wrong with taking some of the wear load off an SSD, whilst at the same time increasing overall system responsiveness? A win-win situation if there ever was one, if you ask me.

[LanceM]

Now, what's wrong with taking some of the wear load off an SSD, whilst at the same time increasing overall system responsiveness? A win-win situation if there ever was one, if you ask me.

With all due respect, I ran several extended write tests to SSDs with swappiness set at default, 10, 5, and 1 for a week each with full use on SSD and monitored writes using Micron Storage Executive for Total Writes. There was no statistical difference in any setting. I'm just not convinced that swappiness or noatime make any real difference.

[Pjotr]

Now, what's wrong with taking some of the wear load off an SSD, whilst at the same time increasing overall system responsiveness? A win-win situation if there ever was one, if you ask me.

With all due respect, I ran several extended write tests to SSDs with swappiness set at default, 10, 5, and 1 for a week each with full use on SSD and monitored writes using Micron Storage Executive for Total Writes. There was no statistical difference in any setting. I'm just not convinced that swappiness or noatime make any real difference.

Yes, well, my statement should have been more qualified.... I meant to say that I've noticed increased system responsiveness because of reduced swappiness, on computers with relatively little RAM.

[LanceM]

Yes, well, my statement should have been more qualified.... I meant to say that I've noticed increased system responsiveness because of reduced swappiness, on computers with relatively little RAM.

I should of added, that I purposely used a RAM starved system with just 2GB. As soon a swap started being used the systems slowed way down.. Even the mouse quit responding. Swappiness settings seemed to make no difference to that. This test was done on Cinnamon 19.3. With the newer kernels, in 20 there have been improvements in swap use. It doesn't engage nearly as soon, nor does it impact the system as much. Certainly one system and a few tests doesn't invalidate the entire swappiness setting, but makes me skeptical of it. One could try it and see. To just blindly change swappiness, irregardless of how much RAM is in the PC and not first monitor if swap is even being used, seems foolish to me. Enough RAM positively, eliminates the issue. 8GB for most users is adequate to have swap never kick in. It can be there for emergency overload if the need arises. If it does, I think default may well be the best setting.

[Pjotr]

Yes, well, my statement should have been more qualified.... I meant to say that I've noticed increased system responsiveness because of reduced swappiness, on computers with relatively little RAM.

I should of added, that I purposely used a RAM starved system with just 2GB. As soon a swap started being used the systems slowed way down.. Even the mouse quit responding. Swappiness settings seemed to make no difference to that.

That's peculiar, because a decreased swappiness should reduce the number of times that the swap kicks in, thus keeping your RAM starved system generally more responsive...

This test was done on Cinnamon 19.3. With the newer kernels, in 20 there have been improvements in swap use. It doesn't engage nearly as soon, nor does it impact the system as much. Certainly one system and a few tests doesn't invalidate the entire swappiness setting, but makes me skeptical of it. One could try it and see. To just blindly change swappiness, irregardless of how much RAM is in the PC and not first monitor if swap is even being used, seems foolish to me. Enough RAM positively, eliminates the issue. 8GB for most users is adequate to have swap never kick in. It can be there for emergency overload if the need arises. If it does, I think default may well be the best setting.

Well, I do it on all of my computers as a matter of course. Been doing it for years: if I remember correctly, since 2009/2010 or something, around the time that I bought my first SSD. Regardless of RAM amount. Fire and forget. Never seen any negative effects because of it, in over a decade.

[Portreve]

Book of Pjotr, Chapter 2, v. 5-7:

Now shalt thou set thy swappiness to 5, lest thy SSD be o'er-wrought with wear. Typeth thee xed admin:///etc/sysctl.conf in thy terminal and presseth [ENTER]. Then mayest thou scroll to the end of thine sysctl.conf file and add the following thereunto:
# Sharply reduce the inclination to swap
vm.swappiness=5

This is what I follow religiously in setting up fresh installs of LM.

[absque fenestris]

Well, Portreve has just emphatically reminded me of the Book of Pjotr.

My goodness - I installed a new system without consulting the book ...

[LanceM]

A reference, albeit not from Mt Sinai https://linuxhint.com/understanding_vm_swappiness/

[LanceM]

Anyone seriously wanting to reduce writes to their SSDs should install profile-sync daemon. For browser writes it will reduce the writes to about half. It's easy to configure. With it, the option to use overlayfs is available and greatly enhances performance. It's in the Software Manager. Documentation:https://wiki.archlinux.org/index.php/pr ... ync-daemon

Code: Select all

lance@LM:~$ psd p
Profile-sync-daemon v6.34 on Linux Mint 20

 Systemd service is currently active.
 Systemd resync-timer is currently active.
 Overlayfs v23 is currently active.

Psd will manage the following per /home/lance/.config/psd/.psd.conf:

 browser/psname:  google-chrome/chrome
 owner/group id:  lance/1000
 sync target:     /home/lance/.config/google-chrome
 tmpfs dir:       /run/user/1000/lance-google-chrome
 profile size:    198M
 overlayfs size:  34M
 recovery dirs:   1 <- delete with the c option

[pbear]

Notice the recommendation to minimize swappiness is based on a single article, which has no cites and gives no credentials for the author (identified only by nom de plume). Meanwhile, my research turned up plenty of users with as much apparent expertise arguing the other way. See, e.g., askubuntu, Linux & Unix, and Linux.com. The last quotes a kernel developer (Andrew Morton) who argues for using more swap, not less.

In any event, there's no reason to reduce swappiness even further to coddle SSDs.

[AZgl1800]

a couple of years ago, I had to deal with Swapiness....

I solved the problem by tripling the amount of RAM in the 1st laptop, and when I ordered this current laptop, it came with 16gB RAM.

I have never felt it slow down unless the Internet got in the way.

In fact, when I installed 19.3 Cinnamon with the Defaults, it did not create a Swap partition at all.

[LanceM]

Swappiness and noatime are two things that are hard to get out of people's heads. It's become ingrained and propagates like medusa. Reminds me of advising to cut snake bites with a razor. That myth has largely dissipated but those old kits are still available online. So I say put mindlessly tweaking swappines and adding noatime to the sword. Do our part in killing the myth. Maybe MythBusters should....

[LanceM]

a couple of years ago, I had to deal with Swapiness....

I solved the problem by tripling the amount of RAM in the 1st laptop, and when I ordered this current laptop, it came with 16gB RAM.

I have never felt it slow down unless the Internet got in the way.

In fact, when I installed 19.3 Cinnamon with the Defaults, it did not create a Swap partition at all.

I'll bet it installed a swapfile. That's what does now instead of a partition.

[AZgl1800]

a couple of years ago, I had to deal with Swapiness....

I solved the problem by tripling the amount of RAM in the 1st laptop, and when I ordered this current laptop, it came with 16gB RAM.

I have never felt it slow down unless the Internet got in the way.

In fact, when I installed 19.3 Cinnamon with the Defaults, it did not create a Swap partition at all.

I'll bet it installed a swapfile. That's what does now instead of a partition.

yup, it did

[antikythera]

I still set swappiness and vfs_cache_pressure, it's a habit I have fallen into and it's not hobbling my systems. My systems are only hobbled by the idiot behind the keyboard deliberately loading too much stuff at times. The CPU bottlenecks a long time before the RAM because they are fairly low end. One is an Athlon 5350 4C 4T with single channel RAM controller and hardly any L2 cache, the other a Core i3-3110M 2C 4T mobile CPU with dual channel RAM (faster than the Athlon 5350). If I apply common sense, I never struggle during normal usage of either system for speed or resources.

swap is activated when necessary on them, I have checked by opening a ridiculous amount of chrome tabs, vlc and libreoffice at the same time to overload the RAM. It took a fair while with 16GB but was easier in the system with 8GB admittedly. I am thinking of upgrading the 8GB to 16GB and doing away with swap altogether.

[Pjotr]

It's an interesting topic, and well worth discussing it again. One should always be critical towards one's own engrained convictions: time passes and things change. And there is such a thing as progressive insight: what looked right yesterday, may well prove to be wrong today.

That said, let's take a look:

Notice the recommendation to minimize swappiness is based on a single article, which has no cites and gives no credentials for the author (identified only by nom de plume). Meanwhile, my research turned up plenty of users with as much apparent expertise arguing the other way. See, e.g., askubuntu,

Now that askubuntu item has this interesting opposing view in its thread:
https://askubuntu.com/a/475465

Linux & Unix, and Linux.com. The last quotes a kernel developer (Andrew Morton) who argues for using more swap, not less.

Note that neither of those articles is very conclusive either way.

In any event, there's no reason to reduce swappiness even further to coddle SSDs.

Maybe that's so for modern SSD's. But if you're aiming for extreme "mileage" and reliability, it makes sense to reduce wear and tear anyway, just to be on the safe side.

Overall, I'm more interested in "best practice" than in scientific double-blind peer-reviewed experiments. So in view of the differing opinions among the knowledgeable, it might be a good idea to move away a bit from the lower extremes of swappiness. Even though those lower extremes have never caused any problems for me, in over a decade.

Therefore I've decided to change my recommendation for swappiness to 20 in all cases (both for SSD's and for spinners). Not because I fear controversy (quite on the contrary: I love a good fight. ). But because we're dealing with a technical performance compromise here, and shifting away a bit from the lower extremes might be a better technical compromise.

[antikythera]

That's the issue, there's no right or wrong setting as such because there are many factors at play. Before even considering the CPU and RAM, or storage controller capabilities there's a silly amount of storage types, some examples - SSD (TLC, QLC), eMMC, M.2 (nvme PCI-E gen 3 or 4, or slower SSD style), HDD

Different sources state different recommended values, so I am sticking with what I know works for my systems for the time being. You could indeed soundly argue for not bothering with these settings changes at all with current generation SSD. BUT currently the best advice as far as I'm concerned is to experiment with the values and see what works best for your system. If performance is impacted whatever you put in, revert back to distribution defaults or simply leave it well alone in the first place if you don't know what you are doing.

[LanceM]

Maybe that's so for modern SSD's. But if you're aiming for extreme "mileage" and reliability, it makes sense to reduce wear and tear anyway, just to be on the safe side.

Then I suggest you promote the use of profile-sync daemon