Prefetching control mechanisms in a multiuser environment

When many simultaneous queries are prefetching large numbers of pages into a buffer pool, there is a risk that the buffers fetched for one query could be flushed from the pool before they are used.

Adaptive Server tracks the buffers brought into each pool by asynchronous prefetch and the number that are used. It maintains a per-pool count of prefetched but unused buffers. By default, Adaptive Server sets an asynchronous prefetch limit of 10 percent of each pool. In addition, the limit on the number of prefetched but unused buffers is configurable on a per-pool basis.

The pool limits and usage statistics act like a governor on asynchronous prefetch to keep the cache-hit ratio high and reduce unneeded I/O. Overall, the effect is to ensure that most queries experience a high cache-hit ratio and few stalls due to disk I/O sleeps.

The following sections describe how the look-ahead set is constructed for the activities and query types that use asynchronous prefetch. In some asynchronous prefetch optimizations, allocation pages are used to build the look-ahead set.

For information on how allocation pages record information about object storage, see “Allocation pages”.