CPU性能分析
CPU能力是制约存储控制器极限性能的最关键因素。因此在按照I/O路径排查性能问题时,需要首先对控制器CPU性能进行分析。
查看控制器CPU利用率
当CPU利用率较高时,系统调度时延会相应增大,从而造成I/O时延增大。
存储系统CPU消耗与I/O模型和组网方式有紧密的联系,在不同的I/O模型或组网方式下CPU利用率可能会相差很大,可以通过DeviceManager性能监控以及CLI命令查看当前控制器的CPU利用率。
- DeviceManager性能监控。
操作入口:“监控 > 性能监控 > 分析”。根据界面引导创建控制器的指标图表,在创建指标图表中勾选平均CPU利用率。创建成功后即可查看,创建指标图表的方法请参见创建指标图表。
- 执行show performance controller命令查看控制器CPU利用率。
admin:/>show performance controller controller_id=0A 0.Memory Usage(%) 1.Percentage of Cache Flushes to Write Requests(%) 2.Cache Flushing Bandwidth(MB/s) 3.Read Cache Hit Ratio(%) 4.Write Cache Hit Ratio(%) 5.Cache Read Usage(%) 6.Cache Write Usage(%) 7.% Hit 8.Cache Water(%) 9.The cumulative count of all read cache hits(Reads from Cache) 10.Cache page utilization(%) 11.Cache chunk utilization(%) 12.Max. Bandwidth(MB/s) 13.Queue Length 14.Bandwidth(MB/s) / Block Bandwidth(MB/s) 15.Throughput(IOPS)(IO/s) 16.Read Bandwidth(MB/s) 17.Average Read I/O Size(KB) 18.Read Throughput(IOPS)(IO/s) 19.Write Bandwidth(MB/s) 20.Average Write I/O Size(KB) 21.Write Throughput(IOPS)(IO/s) 22.Service Time(Excluding Queue Time)(us) 23.Read I/O Granularity Distribution: [0K,4K)(%) 24.Read I/O Granularity Distribution: [4K,8K)(%) 25.Read I/O Granularity Distribution: [8K,16K)(%) 26.Read I/O Granularity Distribution: [16K,32K)(%) 27.Read I/O Granularity Distribution: [32K,64K)(%) 28.Read I/O Granularity Distribution: [64K,128K)(%) 29.Read I/O Granularity Distribution: >= 128K(%) 30.Write I/O Granularity Distribution: [0K,4K)(%) 31.Write I/O Granularity Distribution: [4K,8K)(%) 32.Write I/O Granularity Distribution: [8K,16K)(%) 33.Write I/O Granularity Distribution: [16K,32K)(%) 34.Write I/O Granularity Distribution: [32K,64K)(%) 35.Write I/O Granularity Distribution: [64K,128K)(%) 36.Write I/O Granularity Distribution: >= 128K(%) 37.Average IO Size(KB) 38.% Read 39.% Write 40.Max IOPS(IO/s) 41.Max. I/O Size(KB) 42.Max. Read I/O Size(KB) 43.Max. Write I/O Size(KB) 44.The cumulative count of I/Os 45.The cumulative count of data transferred in Kbytes 46.The cumulative elapsed I/O time(ms) 47.The cumulative count of all reads 48.The cumulative count of data read in Kbytes(1024bytes = 1KByte) 49.The cumulative count of all writes 50.The cumulative count of data written in Kbytes 51.Max. I/O Latency(us) 52.Average I/O Latency(us) 53.Average Read I/O Latency(us) 54.Average Write I/O Latency(us) 55.CPU Usage(%) 56.SCSI IOPS (IO/s) 57.ISCSI IOPS (IO/s) 58.NFS operation count per second 59.CIFS operation count per second 60.Total Disk IOPS(IO/s) 61.READ Disk IOPS(IO/s) 62.WRITE Disk IOPS(IO/s) 63.Disk Max. Usage(%) 64.AI Cache Hit Ratio(%) 65.Unmap Command Bandwidth (MB/s) 66.Unmap Command IOPS (IO/s) 67.Avg. Unmap Command Size (KB) 68.Avg. Unmap Command Response Time (us) 69.WRITE SAME Command Bandwidth (MB/s) 70.WRITE SAME Command IOPS (IO/s) 71.Avg. WRITE SAME Command Size (KB) 72.Avg. WRITE SAME Command Response Time (us) 73.Full Copy Read Request Bandwidth (MB/s) 74.Full Copy Read Request IOPS (IO/s) 75.Avg. Full Copy Read Request Size (KB) 76.Avg. Full Copy Read Request Response Time (us) 77.Full Copy Write Request Bandwidth (MB/s) 78.Full Copy Write Request IOPS (IO/s) 79.Avg. Full Copy Write Request Size (KB) 80.Avg. Full Copy Write Request Response Time (us) Input item(s) number separated by comma:6 Cache Write Usage(%) : 0
当CPU整体利用率长时间处于较高水平时,说明已基本达到控制器的极限性能,建议将部分业务迁移到其他存储系统以降低业务压力。
存储系统支持对CPU利用率设置指标阈值,指标阈值默认为90%。当存储系统控制器的CPU利用率超过指标阈值时,会触发存储系统进行信息采集,采集的信息收集到“/OSM/coffer_data/omm/perf/exception_info/”目录下,该目录下的文件总的大小不能超过14MB,超过后将会覆盖之前的文件。收集到的信息用于后续性能调优或问题处理时具体的分析定位。