The basis for achieving performance is the ability of TSAFS to predict the backup job, and read-ahead thereby reducing the average read service time.
To achieve higher performance, provide TSAFS with as large a set of data items as possible. This can be in the form of specifying multiple resources or a container resource that is likely to contain a large number of data sets.
The TSAs are capable of traversing the file system tree. In this new model, this is used effectively to build the cache. It is recommended that backup engines provide the highest level resource to be backed up and do not start a job at each intermediate level in the file system tree
Implementing selection lists in the backup engine reduces the effectiveness of the read-ahead as a subset of the data sets read by TSAFS will be discarded by the engine. These unnecessary read-aheads impact the average read service time for data sets that are actually backed up. This can be mitigated by either providing the TSA with a filtered list of data sets to be backed up or using the TSA selection list.
It is advisable not to change the open mode during the course of the backup, if possible. The open mode provided to TSAFS is used to build a cache of open files. If the open mode is changed, all open and read-ahead operations have to be stopped, the caches destroyed and rebuilt. This will affect the average service time.
In the new model, the open operation triggers a chain of events such as open and read-aheads. Given this, it is recommended that open is used only during the backup and not during a scan for any reason.
It is recommended that NWSMTSScanDataSetEnd can be used to terminate the job prematurely. This assumes more importance in the current model. If this API is not used, scan, open and read-ahead continue impacting the overall system performance.
In this model, NWSMTSReturnToParent is likely to degrade system performance as all open and read-ahead operations have to be stopped, and the caches should be destroyed and rebuilt. While this is the case, the API is useful in certain circumstances. Hence, the benefits should be understood before its usage.
It is recommended that Close is called immediately on all data sets that were successfully read. This allows TSAFS to effectively use its resources and also prevent resource starvation in extreme cases.