Volumes and MDisks performance

The Volumes and MDisks graphs on the Performance panel show four metrics: Read, Write, Read latency, and Write latency. You can use these metrics to help determine the overall performance health of the volumes and MDisks on your system. Consistent unexpected results can indicate errors in configuration, system faults, or connectivity issues.

Both the volumes and MDisk graphs contain the same metrics to compare and use to evaluate performance; however, the data points for these metrics can be quite different due to the impact of system cache, RAID overhead, and Copy Services functions. You can either choose to display data points in megabytes per second (MBps) or I/O per second (IOPS).

If you select the read metric, data points on the graph indicate the average amount in MBps or IOPS for read operations that have been processed over the sample period. Read metrics represent how much data the system is processing (the bandwidth).

Write metrics display data points on the graphs that show the average MBps or IOPS for write operations that have been processed during the collection period. These metrics indicate how effectively the system is writing data to volumes and MDisks.

The read latency metric for volumes measures the average amount of time in milliseconds that the system takes to respond to read requests over the sample period. The MDisks latency metrics measures the response time for back-end storage.

Increased workload or error recovery can cause spikes in read latency. Values that are consistently higher than expected can indicate a fault situation or an overloaded system. Volume write latency tends to be lower than the MDisk write latency because of volume caching. If the volume write latency is equal to or greater than MDisk write latency, it suggests that the drives might be overloaded and more drives might be necessary to accommodate the increased workload.

Volume latency may be lower than MDisk latency because of the volume cache. Alternatively volume latency may be higher than MDisk latency due to hosts using large I/O sizes. Neither of these conditions indicates a problem with the system.

The difference between read and write IOPS shows the mixture of workload that the system is executing. You can determine the average transfer sizes of data that the system is experiencing by dividing the read and write operations in MBps by the read and write operations in IOPS. This information can be used for validating and predicting disk configuration for the system or input to a disk provisioning application.

Write latency is the average time (in milliseconds) that the system writes data to volumes or MDisksbut does not include the time for write operations that are used to keep volumes in Global Mirror relationships synchronized. As with read latency, MDisk write latency tends to be higher than volumes because of write caching and RAID overheads. For example, a write operation to a volume can result in additional read and write operations on the MDisk depending on the RAID type for the array.