[comp.os.vms] Allin1 performance

lpz%ocu.DECnet@ORNL-STC10.ARPA ("OCU::LPZ") (07/16/87)

         Hi:
         
         We heard all of the complaints about Allin1 performance, but
         it looked light-years ahead of any other office automation
         software, so we bought it.  We use the following account limits:
         
             Maxjobs:         0  Fillm:       255  Bytlm:        36000
             Maxacctjobs:     0  Shrfillm:      0  Pbytlm:           0
             Maxdetach:       0  BIOlm:        50  JTquota:       1024
             Prclm:           6  DIOlm:        18  WSdef:         1024
             Prio:            4  ASTlm:        80  WSquo:         2048
             Queprio:         0  TQElm:        50  WSextent:      3072
             CPU:        (none)  Enqlm:       200  Pgflquo:      20000

         We are running it on a pair of 8600s.  Right now we have some bad
         memory and have 72 meg on one machine and 52 in the other. I will
         probably raise the WsExtent when we get the memory back, but not
         because of Allin1.  It seems quite happy with 3072 pages.  Right
         now we have 50 users and 50% Null time.  This seems to imply that
         we could support at least 80 concurrent Allin1 users, given enough
         memory. I really don't know.  Now, when other people use SAS or
         ISSCO or compile large programs, the CPU gets totally loaded down.
         We also get real slow when the cluster loses a node and 100 people
         are in Allin1 and others are doing SAS, etc. 
         
         You need lots of memory to run Allin1, and that makes the uVAX a
         poor choice, unless you only need to support ~10 concurrent users
         (I wouldn't recommend more, but I have never used Allin1 on a
         uVAX, so it might work anyway). A better idea might be PC Allin1
         for the users and use the uVAX as a file server.  This assumes
         that you have PCs.  Of course, you can get IBM compatibles for
         around 2k each, so this is also cost effective.  Just realize that
         you will need big disks on the uVAX.
         
         I have found that ALL of the complaints about Allin1 speed
         come from one of two sources:
         
         People who listened to the original DEC Allin1 performance
         estimates and now expect it to work with small memory machines;
         
         And those who have slow terminal multiplexers (e.g. DCA, like we
         do, on some of our machines) and don't like waiting for screen
         repaints.  There is some help here, since you can customize
         the menus, but it will still be slow.  But that isn't Allin1's
         fault.  It was designed for use with optimum hardware (large
         memory, fast terminal support (e.g. terminal servers, DMF32),
         etc., and if you try to get around that, you lose.  Most of
         the competing products use disks for everything, and you lose
         on big systems, because you are limited by disk speed.  Memory
         is both cheaper and faster than disks, after all.
         
                                     Lawrence
                                         ~

 -----------------------------------------------------------------------------
 
       ARPA: LPZ@ORNL-STC10.ARPA
        MFE: MACINTYRE@ORN
     BITNET: LPZ@ORNLSTC
 Bell-South: 615.576.0824
   US-Snail: Lawrence MacIntyre
             Martin Marietta Energy Systems
             Bldg 9201-5  MS 8
             P O Box Y
------
------

Peck@RADC-MULTICS.ARPA (Rodney) (07/21/87)

41