For swappiness it seems that it would be set by each system and what they are doing from what I am reading..
For the cache I have found that you MUST enter the command sync first then it all works fine and a free display shows lower after .. without entering the sync command first they system just hangs up ... very true ... Thanks, Paul -----Original Message----- From: Linux on 390 Port [mailto:linux-...@vm.marist.edu] On Behalf Of Rob van der Heij Sent: Wednesday, April 01, 2009 3:43 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: swappiness & drop_caches ? On Wed, Apr 1, 2009 at 9:20 PM, Ayer, Paul W <pwa...@statestreet.com> wrote: > Good afternoon all, > > Just wondering if anyone has some input (good, bad, warnings ...) or > has had to used the following two items .... we are running VM5.4 and > RHEL4.x and 5.x sles 9 and 10 systems > > > > 1) Setting swappiness to other than the default of 60 ? > > Echo nn > /proc/sys/vm/swappiness This really is a "desktop thingy" but may turn out useful for us as well. It defines an amount of page cache to retain even when that means Linux would need to swap things. You would want it large enough to hold relevant program binaries and shared libraries, etc. The nasty part is that it is expressed as a percentage of total memory resources rather than a fixed amount. So you need to come up with a right setting each time you change the virtual machine size. In theory, for largish virtual machines you would want to lower the swappiness. However, when the application does shared memory that lives in page cache (for example the Oracle SGA) then you want to make sure you also leave room for that. Setting it too low will not leave room for the good stuff. > 2) dropping caches ? > > > Echo 1 or 2 or 3 > /proc/sys/vm/drop_caches The Linux server died when the customer tried this. I told him it should return after completion, but did not in 5 minutes. Does that count as bad? ;-) Obviously, if it works well then this is just temporary relief. Linux will immediately start to load stuff in page cache again. The other problem is that z/VM is not aware that the pages have been freed and will be re-used, so they will still be backed by z/VM real page frames or paging space. And by just touching and re-arrange of the pages you may actually make things worse. Instead, you could use CMM-1 and inflate the balloon by the amount that you want to drop from the cache. It will use the same criteria to select pages, but this time it *will* tell z/VM to drop the corresponding real storage. Although the amount may be a bit harder to determine, the advantage is that you don't disturb the usage patterns of the portion that you want to retain. Rob -- Rob van der Heij Velocity Software http://www.velocitysoftware.com/ ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
