z/VSE is a constant changing environment of software and hardware, and as technicians, we’re the conductors and agents of that change. One segment of that change that has been dramatically affected in the last 10 years is performance management and tuning.
Ten to 15 years ago, we tuned disk subsystems by measuring I/Os per file or device, channel busy percentages, and I/O service times. We would then move files to obtain the best performance.
Today, using those techniques with the current disk subsystems may net a performance improvement, but it would be through pure luck of the draw, not science. In today’s disk environments, we know what z/VSE device address a file resides on, but we usually don’t know which real devices comprise the data area for that z/VSE address.
To make those old techniques work, a z/VSE address-to-real device map is required. If you have that information, you can use those techniques to some advantage. However, it may not be worth the effort.
With modern disk subsystems it’s more important to focus on disk cache hit statistics and data in memory concepts. A modern, well-equipped disk subsystem should be directly resolving more than nine out of 10 I/Os from cache. Therefore, tuning using old techniques, if possible, would only improve performance for less than one I/O out of 10.
If your cache hit ratio isn’t 90 percent or greater, there are at least two solutions: more cache memory for the device or controller and/or move some highly accessed data to another cached disk subsystem that has space and cache capacity. Moving a copy of a highly accessed file to CPU main memory via VFBA would improve performance for that file and free up cache space for the remaining files, thus improving the cache hit percentage.
With today’s highly evolved disk subsystems tuning is easier, quicker, and more cost-effective.
Today, tuning CPU cycles and main memory usage is as important as ever. CPU tuning hasn’t changed much, as it still involves proper coding techniques to reduce overhead and a good job mix (CPU constrained and I/O constrained) to maintain a 100 percent busy CPU. With higher performance disk, keeping the CPU busy is easier today than ever.
What has changed is virtual and real memory management. Most z/VSE installations on relatively current hardware have an abundance of real main memory. The question is, are they using it? If VSIZE is less than available real main storage by definition, the system isn’t tuned.
There is no known advantage to running a Page Data Set (PDS), as it isn’t a performance option. And using the old technique of multiple drives for a PDS or partition-oriented multiple drives is a wrong solution. Instead, invoke NOPDS and gain these advantages: no paging (free up CPU cycles), free up disk space, reduce disk I/Os, and reduce load on disk cache.
By and large, measuring and improving tape performance has always meant buying faster tape drives, more connective paths, and programs using large data blocks. The strategies of pre-mounting tapes, use of auto loaders, and massive automated tape storage libraries aren’t new but are still valid.
Another tape performance item to check is to compare a large sequentially processed file on tape and disk. Then, if possible, use the faster device to process the data. As disk and tape environments are upgraded, at times sequential tape processing has actually been faster than disk. Use the device that’s fastest and meets the needs of the application.
Performance measurement is in most ways easier and less technical than it was in the past, but certainly no less important. A few of the old rules are still in effect, including to periodically review all performance statistics and re-tune the system as needed. Another is that if you don’t have time to evaluate and tune the system, then you must have time to suffer with problems of poor performance.
The fastest page-in or page-out is the one that isn’t done (NOPDS), and the fastest I/O is the one that isn’t done (VFBA). There are no substitutes for those two performance tips unless the laws of nature and science are repealed.
Good luck with performance and tuning in z/VSE. Thanks for reading this column, and see you all in the next issue!