| Gavin W. Burris on 12 Jul 2015 07:27:55 -0700 |
[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]
| Re: [PLUG] swappiness and ssd |
I tend to trust the documentation that Red Hat releases. Note that the latest 3.5+ kernels have different behavior. Where 0 was previously used, 1 is now recommended. This change has also been back-ported to 2.6 kernels. https://en.wikipedia.org/wiki/Swappiness#cite_note-2 Cheers. On Sat 07/11/15 12:08PM -0400, Keith C. Perry wrote: > Not sure I agree with Red Hat's assessment of swappiness = 0. The point is to aggressively not swap . This reminds me why Rich Freeman (errr, Prime) and I had similar ideas behind why we leave it off. For me, this sort of "tuning" doesn't directly command behavior or define a policy. So whether 0, 1, 5, or 30, I'm not really telling the kernel directly to, for example: > > swap only if I am about to run out of memory (e.g. under 1Mb while under memory pressure) > start swapping memory when system usage is under 50% > start swapping memory when user consumption is over 50% > stop swapping and return everything to main memory if I will have more that 30% free when done > > You get the idea. > > That said, I feel like I might have hijacked K.S.'s update in regards to swapping on SSD's. Putting aside log-structured file systems which are generally better for SSD use, shouldn't swappiness be turned down on SSD's generally? Doesn't a swappiness of 80 mean you are swapping more than when the system is set to 60? > > Also, if that is the case, does that imply you have to run fstrim more? > > I'm an XFS user and thankfully realtime discarding is available if I am ok with the performance penalty and I don't want to use batch mode. Something to remember for my next server build which will use an SSD for the system drive. > > http://xfs.org/index.php/FITRIM/discard > > > ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ > Keith C. Perry, MS E.E. > Owner, DAO Technologies LLC > (O) +1.215.525.4165 x2033 > (M) +1.215.432.5167 > www.daotechnologies.com > > > From: "Gavin W. Burris" <bug@wharton.upenn.edu> > To: "Philadelphia Linux User's Group Discussion List" <plug@lists.phillylinux.org> > Sent: Saturday, July 11, 2015 9:30:14 AM > Subject: Re: [PLUG] swappiness and ssd > > Is 0 really sane? It was my understanding that swappiness=0 was too > aggressive, almost disables swap all-together. Setting swappiness=1 may > be better, only swapping to avoid the OOM. This, and keeping just a few > gigs of swap as a last-ditch protection, provides a metric to monitor > and avoid. > > Big fat warning here, on the Performance Tuning Guide: > https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/html/Performance_Tuning_Guide/sect-Red_Hat_Enterprise_Linux-Performance_Tuning_Guide-Memory-Configuration_tools.html#sect-Red_Hat_Enterprise_Linux-Performance_Tuning_Guide-Configuration_tools-Configuring_system_memory_capacity > > Cheers. > > > On Fri 07/10/15 09:24PM -0400, Rich Mingin (PLUG) wrote: > > > Not at all. It's the default I had recommended to me by the Arch wiki, some > knowledgeable friends, and I think also Rich Prime. It works for me, though > I still have an issue where VMware stalls my whole machine for 5-30 seconds > every 3-5 minutes if I don't manually disable khugepaged defrag. Ah well. > > On Fri, Jul 10, 2015 at 9:10 PM, Keith C. Perry <kperry@daotechnologies.com> > wrote: > > > Ahh so swappiness=0 is not a weird setting. > > > > --- > > KP- > > > > On Jul 10, 2015 9:05 PM, "Rich Mingin (PLUG)" <plug@frags.us> wrote: > > > > > > I was declining to post, but I'll just chime in, I'm also swappiness=0 > > on my machines. The two that touch VMs have 32GB of ram, SSDs, and a small > > swap file on some spinning rust, so I get all of the benefits listed, > > without unneeded SSD wear. > > > > > > On Fri, Jul 10, 2015 at 8:46 PM, Rich Freeman <r-plug@thefreemanclan.net> > > wrote: > > >> > > >> On Fri, Jul 10, 2015 at 6:31 PM, Keith C. Perry > > >> <kperry@daotechnologies.com> wrote: > > >> > > > >> > Rich Freeman, you and I both leave off for similar reasons. Did you > > know > > >> > about this? Looks like a lower setting (like 0 LOL) might get swap > > to > > >> > perform in a "sane" way. > > >> > > > >> > > >> I've never really been satisfied with it, even with a swappiness of 0. > > >> > > >> -- > > >> Rich > > >> > > ___________________________________________________________________________ > > >> Philadelphia Linux Users Group -- > > http://www.phillylinux.org > > >> Announcements - > > http://lists.phillylinux.org/mailman/listinfo/plug-announce > > >> General Discussion -- > > http://lists.phillylinux.org/mailman/listinfo/plug > > > > > > > > ___________________________________________________________________________ > > Philadelphia Linux Users Group -- > > http://www.phillylinux.org > > Announcements - > > http://lists.phillylinux.org/mailman/listinfo/plug-announce > > General Discussion -- > > http://lists.phillylinux.org/mailman/listinfo/plug > > > > ___________________________________________________________________________ > Philadelphia Linux Users Group -- http://www.phillylinux.org > Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce > General Discussion -- http://lists.phillylinux.org/mailman/listinfo/plug > > > > > > -- > Gavin W. Burris > Senior Project Leader for Research Computing > The Wharton School > University of Pennsylvania > ___________________________________________________________________________ > Philadelphia Linux Users Group -- http://www.phillylinux.org > Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce > General Discussion -- http://lists.phillylinux.org/mailman/listinfo/plug > ___________________________________________________________________________ > Philadelphia Linux Users Group -- http://www.phillylinux.org > Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce > General Discussion -- http://lists.phillylinux.org/mailman/listinfo/plug -- Gavin W. Burris Senior Project Leader for Research Computing The Wharton School University of Pennsylvania ___________________________________________________________________________ Philadelphia Linux Users Group -- http://www.phillylinux.org Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce General Discussion -- http://lists.phillylinux.org/mailman/listinfo/plug