Re: Performance on MVS

Hi Neil,

So far, mainframe (S/390 and S/370-XA) is the best solution for I/O bound DP. Since XA architecture, physical I/O handling has been delegated to separate processor, IOP. System processor or CPU just send a request (ORB) to IOP and do something else. When request is completed, IOP then interrupt system processor.

The way IOP manage I/O is also very good. Each disk actuator is connected to central storage by 4 or 8 channels, which is logically as a single path, known as subchannel. Balancing within subchannel is fully controled by IOP. Since this 4 or 8 channels are also connected to other disk actuator, IOP then only connect them to a disk when transferring data. Then within a set of 4 or 8 channels, only one subchannel active at a time.

However, how best it is, still depend on your software configuration. (1) At OS level, you shouldn't put any 2 of page datasets and spool dataset or catalog dataset in the same volume. Queueing on paging will delay memory request. Queueing on spooling will delay jobs execution. Queueing on catalog will delay data retrieval. Don't let conflict between paging and spooling and catalog. (2) If your spool is quite heavy loaded, separate into more than one volumes. Don't let queueing on spooling. (3) Dispatching and I/O priority or service class of service address spaces must higher than their application. For example, VTAM and JES2 must higher than CICS and TSO. DB2 subsystem and master must higher than their DB and stored proc. (4) At application level, heavily application datasets should be separated. Don't let 2 or more dataset heavily active in the same volume at the same time. Keep I/O balance for all volumes.

Hope this help.

Regards,

--
Deru Sudibyo


www.MainFrameForum.com USENET Gateway