On Wed, Aug 26, 2020 at 7:38 PM Chris Murphy <lists@xxxxxxxxxxxxxxxxx> wrote: > > cc: Zbigniew and Alexey for opinions. ---------- Forwarded message --------- From: Zbigniew Jędrzejewski-Szmek <zbyszek@xxxxxxxxx> Date: Thu, Aug 27, 2020 at 1:44 AM On Wed, Aug 26, 2020 at 07:38:37PM -0600, Chris Murphy wrote: > On Wed, Aug 26, 2020 at 6:02 PM Ed Greshko <ed.greshko@xxxxxxxxxxx> wrote: > > > > On 2020-08-27 06:28, Chris Murphy wrote: > > >>> We are considering bumping the default size for swaponzram. So (b) is > > >>> helpful to see if there are any negative side effects long term. I > > >>> never saw any, but I am not a sufficient sample size. > > >>> > > >>> Once the installation is finished, you can copy that same .conf file > > >>> to /mnt/sysimage/etc/systemd/ in the installation environment, so it > > >>> takes effect upon reboot. > > >> I'll do this soon. It may be delayed a day or so due to previous commitments. Is that OK? > > > Yeah no hurry. > > > > > > > My morning unexpectedly opened up. > > > > Doing the above allowed the installation to complete with 1248mb of memory assigned to the VM. > > The system booted and is running fine. > > > > [egreshko@localhost ~]$ zramctl > > NAME ALGORITHM DISKSIZE DATA COMPR TOTAL STREAMS MOUNTPOINT > > /dev/zram0 lzo-rle 1.2G 510.6M 159.7M 167.6M 1 [SWAP] [I think this will be rejected from test@ because I'm not subscribed...] > Effective compression ratio is 3:1. Also, that's at the low end of > what I regularly see. High end 4:1. Yes, that's what I see too. On my laptop: 0.27, on a VM here: 0.28, on a rpi3 that's doing dnf upgrade: 0.32. And if we have usual ratios like that, then we could crank up the zram partition size. With 1/3 * 0.5 we're really only using 1/6th of RAM for swap. We could crank it up to 1.4 and not exhaust all RAM. (Controlling zram through the uncompressed size is not very convenient. It'd be much easier to specify the compressed size, i.e. the backing size. An RFE for the kernel people.) I think we should ask people to report their compression fraction in various scenarios and make a decision based on that. > If we had more data, 1:1 might be the safer and more conservative > option. Thing is, I don't know that. I can't prove it. But there's a > lot of historical data suggesting swap should be 50% of RAM. It only > became common to use 100% and higher, to support the hibernation use > case. > > A compromise for F33 that has already been suggested, and is > compelling, is bump the ratio to 75%. And the cap from 4G to 6G. We > have so many use cases we don't know a ton about, on a variety of > archs, is the main reluctance. Things have been going extremely > smoothly so far. > > But if we were to change it, a Beta FE might be a good time to do it, > so the new values are in beta. Because we could then back off for > final. > > cc: Zbigniew and Alexey for opinions. In general, I'm for increasing. Zbyszek -- Chris Murphy _______________________________________________ test mailing list -- test@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to test-leave@xxxxxxxxxxxxxxxxxxxxxxx Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/test@xxxxxxxxxxxxxxxxxxxxxxx
