From: CRDGW2::CRDGW2::MRGATE::"SMTP::CRVAX.SRI.COM::%GIZMO.SRI.COM:RELAY-INFO-VAX" 20-APR-1989 21:14 To: MRGATE::"ARISIA::EVERHART" Subj: VMS V5.0.2 MPW revisited... Received: From KL.SRI.COM by CRVAX.SRI.COM with TCP; Thu, 20 APR 89 17:28:51 PDT Received: from fermat.Mayo.edu by KL.SRI.COM with TCP; Thu, 20 Apr 89 17:11:17 PDT Received: from RCF.DECnet MAIL11D_V3 by fermat.Mayo.edu (5.57/Ultrix2.4-C) id AA05859; Thu, 20 Apr 89 18:55:19 CDT Date: Thu, 20 Apr 89 18:55:18 CDT Message-Id: <8904202355.AA05859@fermat.Mayo.edu> From: brunkhorst@Mayo.edu (Geoffrey Brunkhorst) To: "info-vax@kl.sri.com"%FERMAT.dnet@fermat Cc: BRUNKHORST@fermat.Mayo.edu Subject: VMS V5.0.2 MPW revisited... I am becoming acutely aware of VMS V5's MPW's (modified page writer) new "feature" of slowly becoming filled and fragmented. During some network storms, one of my cluster nodes (an 8600) was failing over all the users to my 8810. While I have 48mb of mem on the system, and 4 page files (6000, 100,000, 60,000, and 60,000 pages respectively), I genned about 5 balance set slots too few for the total load incurred during this time. To make matters worse, it seemed the swapper was always choosing COM batch jobs to swap (even if I had 10 HIB jobs sitting idle, like evl and configure), causing a lot of swapper trimming. Performance was hell. I learned quick that over-genning balsetcnt may not be such a bad thing afterall. Anyway, I have noted during this time that I am 'losing' about 1000 pages a day in the 'free' column of SHOW/MEMORY/FILE display. 'Page file [xxx] fragmented... continueing' messages start to show up during real bad times. But as I write, I have three batch jobs running, each of which is taking up less than 10 k Virtual memory, nothing swapped, no user jobs other than my own, and I got over 60,000 pages 'not free'. Why? I don't know? Does Anyone know? A month ago, I would have almost no page space in use with the same load. Well, I remember about a month ago, someone flamed a little on this new feature of the MPW. Has there been any fallout? Does 5.1 (or for you beta sites, does 5.2) fix this problem? From the release notes of 5.1, I see no glaring admittance of a problem. If memory serves me, the problem was fixed by rebooting the system. This is out of the question for me most of the time, as I run 2-4 week simulations on my system, often times overlapping in duration. (It has been 35 days since my last reboot.... I could have prefaced that with "Forgive me father, for I have sinned..." ). Comments? - Geoff Brunkhorst@mayo.edu 507-285-6941 Mayo Foundation, Rochester MN