SQL Anywhere supports two different kinds of parallelism for query execution: inter-query, and intra-query. Inter-query parallelism involves executing different requests simultaneously on separate CPUs. Each request (task) runs on a single thread and executes on a single processor.
Intra-query parallelism involves having more than one CPU handle a single request simultaneously, so that portions of the query are computed in parallel on multi-processor hardware. Processing of these portions is handled by the Exchange algorithm.
Intra-query parallelism can benefit a workload where the number of simultaneously-executing queries is usually less than the number of available processors. The maximum degree of parallelism is controlled by the setting of the max_query_tasks option.
The optimizer estimates the extra cost of parallelism (extra copying of rows, extra costs for co-ordination of effort) and chooses parallel plans only if they are expected to improve performance.
Intra-query parallelism is not used for connections with the priority option set to background.
Intra-query parallelism is not used if the number of server threads that are currently handling a request (ActiveReq server property) recently exceeded the number of CPU cores on the computer that the database server is licensed to use. The exact period of time is decided by the server and is normally a few seconds.
Discuter à propos de cette page dans DocCommentXchange.
|Copyright © 2013, SAP AG ou société affiliée SAP - SAP Sybase SQL Anywhere 16.0|