First, I agree with Bill, that it usually is not a good idea to break up an account across user volume sets. This is because the rename command will fail and usually this will be in some critical month end or quarter end job.

The way that I like to think of the transaction manager (XM) is a dirty page manager. It processes the pages for attached files that have been dirtied by modification. The dirty pages are buffered in memory and then in a log on the volume master. At checkpoint time the log is then posted to the corresponding physical page.

Thus, the first priority should be to dirty as few pages as possible. In Turbo Image terms this means minimizing secondaries, minimizing the number of paths, using CIUPDATE, and adding detail data at the eof.

The second priority is to move the head as little as possible to reduce the seek times. This can be translated into isolating the volume master as much as possible from the busiest user files, isolating master/detail pairs, and isolating work files.

Many applications have a single master detail pair that accounts for over 80% of all of the physical IO. Many batch applications have self defeating IO, serial read of the master, find, then chained read of the detail, writing qualified entries to a work file. 

Another side benefit of isolating work files is that then they can be more easily excluded from backups. It is not unusual to find month end work files among the largest files on the system. In most cases the work files are purged at the start of the job and it is quite easy to copy the purges to the end. 

Finally the main value of user volumes is that they break up the dreaded re-install/restore process and therefore reduce the associated downtime, although with mirroring and RAID implementations this is becoming less of a factor. When can we expect a solution for the MPE system volume set?

Mike