Oracle MVS CPU Utilization Nightmare

We are running Oracle for MVS, 7.2.3, but use it strictly as a mechanism to access our Unix instances of Oracle. For example, a CICS program will connect to Oracle/MVS, which then uses SQL*Net (2.2.3) and IBM's version of TCP/IP for MVS (V3) to route the request to a remote instance of Oracle on our AIX boxes. There are no Oracle instances on MVS.

While our applications work, we are paying a terribly heavy price for it in CPU consumption. We have a three processor Amdahl box, and Oracle, SQL*Net and TCP/IP are by far the largest consumers of CPU on it, to the point where little else is getting done. We are pumping a lot of transactions through (I don't have the exact numbers, but can get them), but this still seems irrational to me. The three MVS address spaces are basically doing not much more than routing the work to Unix. To my way of thinking, we are paying a lot and getting a little. I can contrast this to the DB2 work that is going on at the same time. It uses quite a bit less CPU than Oracle is using, and DB2 is actually managing data bases on MVS, not sending work somewhere else to be done.

Management is reluctant to upgrade the mainframe; moving work to Unix was supposed to reduce mainframe usage, not lead to upgrades. Also, it seems likely that there's enough latent demand that we'll expand to use up whatever we upgrade to, anyway.

I am not an Oracle DBA (my background is DB2), so I hope what's here makes sense. Oracle says that things are working as designed, but I don't buy it. We're in a pickle, and I hope someone can help. Received on Wed Oct 02 1996 - 00:00:00 CEST