摘要 |
Monitoring the performance of a parallel database stored on a data storage device in a computer, and comprising a plurality of nodes (312-316, Fig. 3), each node containing one or more partitioned databases. Logical groups of one or more database nodes are identified (700-706, Fig. 7) and can be changed dynamically by the user. Within a predetermined collection time period, performance statistics from as many nodes in the group as possible are collected. The collected performance statistics are stored in memory ("snapshot buffer") and re-used when collecting performance statistics from one or more groups of nodes in a succeeding collection time period: performance statistics from different nodes in the group can thus be newly collected, and some of the oldest data from previously-monitored nodes can be refreshed as time allows. Thus after a number of runs, more precise results can be returned to the user. The monitoring statistics (which can include the row distribution in a set of nodes, the minimum, maximum or average number of SQL statements attempted in a set of nodes, or the number of deadlocks found in a set of nodes) can also be collected in parallel.
|