Query optimization considering virtual machine mirroring costs

摘要

Techniques are disclosed for selecting between query execution plans in part based on VM mirroring costs. In one embodiment, a query optimizer determines runtimes for multiple query execution plans, and modifies these runtimes based on flashing cost, which is itself determined based on the amount of memory which needs to be flashed to the secondary memory if each execution plan is executed. In so doing, the query optimizer may select a execution plan which takes a longer to run, but incurs less flashing cost. By considering both the query runtime itself and the flashing cost, the query optimizer may be better able to determine the fastest query execution plan to execute.

主权项

1. A computer-implemented method for optimizing query execution, comprising:
determining an original runtime for each of multiple execution plans for a query; and if execution of the query overlaps with one or more virtual machine (VM) mirroring operations:
modifying each of the original runtimes of the execution plans based on time required to perform the mirroring operations if the query were executed according to the execution plan, wherein the time required to perform the mirroring operations is determined based on an estimated number of memory pages that would become dirty and a cost to flash the dirty memory pages,selecting one of the execution plans based on the modified runtimes, andexecuting the query according to the selected execution plan; and if execution of the query does not overlap with any VM mirroring operations:
executing the query according to the execution plan with the fastest original runtime.