(ZOS) SMF record splitting

Since most of the WebSphere Application Server SMF records are used to describe variable-length data structures (for example, there might be hundreds of classes by container and hundreds of methods by class), the SMF records may be larger than the maximum record size supported by SMF (32KB). In this case, the logical records need to be split into several physical records.

Each of those physical records needs to be self-describing and self-contained. Self-describing indicates what we described in the paragraph on triplets before; it is a purely mechanical structure to help read a record. Self-contained indicates that, even if we have only a subset of the physical records at hand that together describe the original logical record, we need to be able to evaluate these records, combine the information stored in them, and set an 'incomplete' flag. This is required since, as we break up a logical record into physical records and write them to SMF one after the other, SMF might decide that only the first few physical records fit into the primary SMF dump dataset whereas the remaining physical records are written into an alternate SMF dump dataset. At the time when a formatted SMF dump dataset is evaluated, we may not assume that all physical records that make up one logical record are present. For example, self-containedness of a physical container activity record means that it contains the description of the container, but not necessarily all of its classes.

We use a similar splitting mechanism like the one that is currently used in the RMF product. Note that in the case of container records (subtypes 5, 6,7, and 8), we cannot assume that records will be split at a class boundary, but we must consider the case when the methods that belong to one class also need to be split over multiple physical records.

The section length numbers used throughout the following diagrams are only for demonstrative purposes. In particular, the arrows indicating 32K boundaries or the total length of the records are placed at random. We can fit many more classes and methods into a physical record than suggested by the diagrams.

Collecting job-related information with the System Management Facility (SMF)

Triplets

Header/self-defining section