Matti,
     #1, 2.1.5.x is the SMP code, 2.1.0.x is the single processor code.
     #2, spread your adsm data base files out over different controllers
     and for sure different volumes. AND run your logs on different volumes
     than your data base files.
     #3, I have a 591 (with only 256 MB core) that handles @ 50 concurrent
     sessions, 4.5 TeraBytes/Month (made up of @ 1 million files) and it is
     coasting.
     #4, I have a 59H (with 512 MB core) that I WISH would handle its 50
     concurrent sessions, 4.5 terabytes/month (made up of 10 million
     files). It is a dog!
     BUT.... the 591's DB is 3 GB and the 59H's DB is 13 GB, the 591 (as
     you can see) deals with LARGE files, the 59H deals with little files
     (in the xx KB range)  the 591 has realistic retention, the 59H is
     trying to keep files until all the uranium on the earth is depleted.
     MY CONCLUSION is that these smaller processors are bound by (A) file
     quantity of between 5 and 7 million files per month OR (B) how ever
     much data you can shove in them in an 8-10 hour time frame depending
     on how many network interface cards you have, how many tape drives you
     have, and/or how much disk buffer you have.

     BIG DOG FACTOR #1 : Novell servers... (insider trick (a) have a
     scheduled event that runs a xxx.ncf that kicks off a LOAD DSMC INC
     afilesystemlogicalvolume  for each logical volume in the server. If
     you have a SYS: APPS: DATA1...DATA3 you could kick off 5 LOAD DSMC's
     but 6 seems to be the max our netframes can drive. )

     You can also use this multi-dsmc in unix environments and 9-12
     concurrent sessions seems to be about max we can drive. (each pushes @
     1.5 GB / hr) Oh that is FROM a single box, then we have 2, 3, or 4
     boxes backing up at the same time...

     tune, tune, tune, .....   network MTU size is a biggie outside ADSM,
     ALL the ADSM parameters, etc.

     It also SEEMS archives are easier on the servers than backups... I'm
     guessing less overhead.  (so production data base files whose copies
     are known to be worthless a couple weeks after they are made are made
     with archives.  This might be my twisted vision but it seems to be the
     case)

     Hope this helps
     later
          DWight



______________________________ Reply Separator _________________________________
Subject: How to Scale ADSM
Author:  ADSM-L (ADSM-L AT VM.MARIST DOT EDU) at unix,mime
Date:    3/10/97 3:47 AM


  Hello !

I am running my ADSM server 2.1.5.12 in RS/6000 J40 and I have problems
with performance.

When I run "dsmserv audit db -fix" it takes 15 hours (my db has 10 500
000 entries).

I also I need to run more than 20 "dsmc i" sessions parallell within this
server machine (adsm client and server in same machine). And the server
get's slow.

So would some more memory or additional cpu's help, currently I have two
CPU's and 256 Mem. Can ADSM use SMP advantages ?

What could I do to make performance better.

  Matti Harvala
  Unix System Manager (Computing Services)

  Nokia Mobile Phones    Phone:  +358 (0)10 505 2254
  P.O. Box 86            Fax:    +358 (0)10 505 2899
  24101 SALO FINLAND     Mobile: +358 (0)40 556 5931
  E-mail: matti.harvala AT nmp.nokia DOT com