As with the other types of VSAM clusters, entry-sequenced data sets generally perform better with larger control interval sizes, both in terms of read/write performance and DASD utilization. As with the other types also, this will vary depending on the primary access mode used to process the data set — either direct or sequential. Since not many applications process entry-sequenced data sets in direct mode by RBA address, the larger CI size will probably be more effective. This cannot be an alternative, however, if the data set is used by IMS or CICS. Your installation can have requirements limiting the CI size to decrease the system resources required for VSAM buffers. If this is the case, do not change CI size without checking first (with IMS and CICS databases it can cause problems if you change the CI size anyway, regardless of whether the buffer space is a restriction — it depends on the use of the ESDS cluster).
A very important set of values is printed if the data set is defined with the SPANNED attribute and contains at least one spanned record. Because of the tremendous overhead involved with spanned records (both performance and DASD), they should definitely be the exception and not the rule in a cluster. Check the percentage of spanned records in the data set. Depending on the cluster, even 1 percent can be too high for performance reasons. Check the minimum, maximum, and average spanned record size. It can be possible to redefine the cluster with the next-larger control interval size and eliminate the spanned records altogether. We must again, however, raise the caveat concerning IMS and CICS clusters: Don't change the CI size of the cluster unless the cluster can tolerate being reorganized with a new CI size, and unless the buffer space is available for your online applications.
|
Copyright © 2015 CA Technologies.
All rights reserved.
|
|