发明名称 Computer system backup performance optimization through performance analytics
摘要 Embodiments in accordance with the present invention disclose a method, computer program product, and system for optimizing performance of a computer backup solution that includes at least two data movers. The automated method includes measuring data mover performance during operation of a backup cycle, and optimizing the performance of data movers by increasing or decreasing the number of threads operating concurrently in the data movers. The method further includes computation of performance rankings of the data movers and shifting workload among the data movers in accordance with their respective performance rankings, such that the computer backup solution converges toward an optimized configuration.
申请公布号 US9542275(B2) 申请公布日期 2017.01.10
申请号 US201615052926 申请日期 2016.02.25
申请人 International Business Machines Corporation 发明人 Damgar James E.;Wolfe Daniel C.
分类号 G06F9/455;G06F11/14 主分类号 G06F9/455
代理机构 代理人 Woycechowsky David B.
主权项 1. A method for optimizing performance of a computer backup solution that includes at least a first data mover and a second data mover, the method comprising: measuring performance statistics of the first data mover with respect to at least one virtual machine class processed by the first data mover, to produce performance statistics, by virtual machine class, of the first data mover; measuring performance statistics of the second data mover with respect to at least one virtual machine class processed by the second data mover, to produce performance statistics, by virtual machine class, of the second data mover; optimizing, by one or more processors, performance of the data mover; operating, by one or more processors, a performance model operationally coupled with a data mover wherein the performance model includes at least performance statistics of the first data mover and performance statistics of the second data mover; initiating, by one or more processors, communications between the first data mover and the second data mover during at least a backup cycle, to exchange at least performance statistics between the first data mover and the second data mover; generating, by one or more processors, a peer-to-peer model wherein the first data mover has access to the performance model of the second data mover and the second data mover has access to the performance model of the first data mover; computing, by one or more processors, performance rankings with respect to virtual machine class, of the first data mover and the second data mover, based at least in part on performance statistics of the first data mover and performance statistics of the second data mover; analyzing, by one or more processors, respective performance statistics of the first data mover and the second data mover, with respect to virtual machine class, to produce performance rankings, by virtual machine class, of the first data mover and the second data mover; shifting, by one or more processors, some workload from the first data mover to the second data mover, in accordance with their respective performance rankings, with respect to virtual machine class, such that the computer backup solution converges toward an optimized configuration; shifting, by one or more processors, some workload from virtual machines of a class, from the first data mover to the second data mover, based at least in part, on the performance rankings of the first data mover and the second data mover with respect to workload from virtual machines of the class; operating concurrently, one or more producer threads in the data mover; operating concurrently, one or more consumer threads in the data mover; changing the number of producer threads or consumer threads operating concurrently in the data mover; measuring a change of performance of the data mover, in response to changing the number of producer threads or consumer threads operating concurrently in the data mover, to produce performance statistics on which to base, at least in part, a subsequent change in the number of producer threads or consumer threads operating concurrently in the data mover, such that the data mover converges toward an optimized performance state; storing at least performance statistics relative to the first data mover, in the performance model associated with the first data mover; storing at least performance statistics relative to the first data mover, in the performance model associated with the second data mover; storing at least performance statistics relative to the second data mover, in the performance model associated with the first data mover; and storing at least performance statistics relative to the second data mover, in the performance model associated with the second data mover.
地址 Armonk NY US