IAM Tuning Guidelines
Overview
For the great majority of data sets converted to IAM, there is no real need to perform any tuning. This is because IAM, with its Real Time Tuning capability and Turbo buffering mode, can generally provide an outstanding level of performance, without the need for any manual intervention. The tuning guide is being provided for those installations that want to make sure that they are getting the best possible performance from IAM. It is also being written to address those few files that just seem to require a bit of extra effort. The last reason is to provide an aid for installations that are having a resource constraint, particularly with real or virtual storage.
What do we mean by performance? With an access method, such as IAM, performance is retrieving and storing data faster while using less computing system resources. Nothing is free however, and frequently there are trade-offs involved, such as using more of one resource to use less of another. These resources include the utilization of the processor, referred to as CPU time, DASD space utilization, and utilization of channels, control units, and physical devices to move the data between processor storage and the device, and the use of both virtual and real storage. An access method needs to use a portion of all of these resources to provide the service. Frequently tuning involves adjusting resource use from a constrained resource to an unconstrained resource.
The longest portion of any I/O operation is the time it takes to transfer data from the storage device into processor storage. It therefore follows that the fastest logical I/O is one where there is no physical I/O. IAM’s goal is to satisfy as many logical I/O requests as possible without performing any physical I/O. IAM generally utilizes virtual storage, and of course the underlying real storage, to obtain the high performance. The two primary storage areas to accomplish that are the index area and the buffers. IAM retains the entire index, normally in an internally compressed format, for the data set in virtual storage while the data set is open. This eliminates any need for physical I/O to read the index. Any record can be retrieved with no more than 1 physical I/O. With VSAM, assuming the required control intervals are not within the buffer pool, a typical random read will generally require at least 2 physical I/O’s, one for the index and the other for the data. Depending on the VSAM index structure, more I/O’s may be required for the index component.
The other way IAM reduces physical I/O’s is through its proven Real Time Tuning buffer management algorithms. Using dynamic buffer management techniques, IAM is able to reduce physical I/O to a level that is quite difficult to achieve with VSAM, even with extensive manual tuning. IAM will dynamically adjust the number of buffers, the buffer management algorithms, and physical I/O techniques used to match the application programs current requirements. The result with IAM is better performance, resulting in reduced elapsed times for batch jobs, reduced response times for online transactions, and less time investment required on tuning to achieve those goals.
This section provides more information about the following topics: