Welcome to the Linux Foundation Forum!

Swapping on SSD, and heavy swapping

Hello fellow future Linux admins ;-)

Is there something to take care of when setting up swap space on an SSD?
In particular concerning the reliability of the system?

Below are my thoughts; what I am interested in is:

  • do you have own experiences in that regard? (the failures, the recoveries)
  • links you know of to articles that discuss these issues

Thanks, and happy studying! Christian


I heard some reasons why swapping is good or required over the past years (without formal Linux education) :

  1. it allows the kernel to move unused pages (from unused/sleeping code or unused/irrelevant data) out of the hot RAM, and thus to utilize the RAM better (for buffering, for caching, for more space for actually active code/data)
    • I think this was important when free RAM was scarce; these days, I have some systems running stable without any swap partition
  2. it was/is used as backing space for hibernation
    • I think this is only relevant for laptops and desktop machines, not for server machines (or server VMs)

In the ages of HDD and insufficient RAM there often were times when heavy swapping set in. With regard to SSD, I see two differences that make me skeptical:

  1. the wear was on the mechanics lifetime (HDD), not on the flash lifetime (SSD)
    • HDD mechanics could be (probably were) designed to life with high-cycle-fatigue, whereas SSD flash has a nominal limit for write/erase-cycles
    • so I expect SSD swapping will eventually fail, depending to some degree on the load pattern and on the quality of the SSD
  2. the sluggish nature of mechanics meant that HDD-swapping throttles the system independently of age (of remaining lifetime)
    • that works like a counter measure; when the system is overloaded, instead of just burning through, it will also throttle down
    • you get a performance feedback always and instantly, regardless of the remaining lifetime of the HDD/SSD component, so you are triggered to react long before the HDD/SSD is actually about to fail

Regarding that, what I wonder is:

  1. if frequent swapping is expected, would it still be best practice to put the Linux, the user software/data, and the swap on one SSD device? Or is the best practice to invest in a separate device for swap, even if this will raise the price?
  2. are there soft measures?
    • (maybe to partition SSD flash, so the wear of one partition will not harm the life expectation of another partition - I don't think that is technically implemented, I think each SSD is just one black box, unlike the magnetic tracks of the HDD you cannot tell which flash cells will be used)
  3. what is the hard impact?
    • I'd expect that, when all three - OS, user software/data, and swap - are one one partition, sudden death due to heavy swapping will then also destroy the OS and user software/data
    • or are there ways to soften the failure of the swap device

Comments

  • coop
    coop Posts: 915

    It is not that complicated. If you are swapping all the time, either get more RAM or don't use the machine this way, or make sure the swap device is not SSD. Or you can use zswap which uses compressed memory for swap -- fedora now does this by default, but it can be minor league tricky to set up.

    Any system which depends on using swap often is broken to begin with. A lot of folks, including me, tend to just not use swap anymore.

Categories

Upcoming Training