Using Cache to Improve Performance for Directly-Accessed Data Sets

Dynamic cache management can be used to improve performance for data sets that are primarily accessed directly.

To use enhanced dynamic cache management, you need cache-capable 3990 storage controls with the extended platform.

If you use dynamic cache management in a storage environment that includes cache-capable 3990 storage controls, you can establish a performance hierarchy for data sets that are primarily accessed using direct access methods.

System-managed data sets can assume the following three states with 3990 cache and DASD fast write services:

Must-cache data sets if the data set's storage class performance objective recorded as the direct millisecond response or sequential millisecond response demands cache facilities.
Never-cache data sets if the data set's direct millisecond response or sequential millisecond response is specified as 999.
May-cache data sets if direct millisecond response and sequential millisecond response values do not require the use of cache. These data sets can use cache facilities only if the must-cache data sets do not fully use cache and non-volatile storage required for DASD fast write.

The enhanced dynamic cache management of DFSMS ensures that must-cache data sets have a priority on 3990 cache and DASD fast write services, and that the may-cache data sets that benefit most from cache and DASD fast write receive these specialized performance services. You can get the best performance by assigning most data to the may-cache category. The enhanced dynamic cache management then supports performance management automation, but lets you designate selected data as must-cache data sets.

The system selects cache candidates so that the components of a cache-capable 3990 storage control are not over-committed. Data set level statistics are recorded in SMF so you can monitor the effect of cache services on I/O performance. MVS/ESA SML: Managing Data describes the data that is collected.

The cacheability of data sets also depends on the applications. Some applications could access the same data several times, while others (for example, sequential access) might not.