Tommy_Ericson_QZ@QZCOM.MAILNET (07/25/86)
Is there anybody out there who has done some investigations about the a) overall system effects b) user response when having PFRATL=0 and PFRATL>0 (e.g. 1)? We have the feeling that PFRATL=0 is to be preferred only if primary memory is not fully committed, but that PFRATL=1 gives positive effects on both a) and b) when memory is overcommitted. While on the topic: having PFRATL=1 and plenty of free memory, VMS still reduces phys mem for processes, thus forcing them to unnecessary page-faults (although in core). I would suggest that there was a SYSGEN parameter (unit: Pages) with the semantics "do not reduce process memory allocation if free memory is higher than this value", i.e. in effect PFRATL=0. Isn't this actually how the swapper is supposed to work? I cannot see any drawback with this.
sasaki@HARVARD.HARVARD.EDU.UUCP (07/26/86)
One thing that I noticed about version 4, and I was glad to see others confirm this, is that the swapper doesn't do working set trimming as soon as it used to. On version 3 of VMS you could set PFRATL to 0 and leave it there, trimming would work and you would see working sets get smaller and rarely see swapping. I usually run the VMS systems with PFRATL=0 as long as the free page list is large. If the system gets really loaded, then I move PFRATL to 1 to 3 until the overall response improves. ---------------- Marty Sasaki uucp: harvard!sasaki Harvard University Science Center arpa: sasaki@harvard.harvard.edu One Oxford Street bitnet: sasaki@harvunxh Cambridge, MA 02138 phone: 617-495-1270