Virtual SCSI latency

Find information about Virtual SCSI latency.

I/O latency is the amount of time that passes between the initiation and completion of a disk I/O operation. For example, consider a program that performs 1000 random disk I/O operations, one at a time. If the time to complete an average operation is 6 milliseconds, the program runs in no fewer than 6 seconds. However, if the average response time is reduced to 3 milliseconds, the run time might be reduced by 3 seconds. Applications that are multithreaded or use asynchronous I/O might be less sensitive to latency, but in most circumstances, lower latency can help improve performance.

Because Virtual SCSI is implemented as a client and server model, there is some latency overhead that does not exist with directly attached storage. The overhead might range from 0.03 to 0.06 milliseconds per I/O operation depending primarily on the block size of the request. The average latency overhead is comparable for both physical disk and logical volume-backed virtual drives. The latency experienced when using a Virtual I/O Server in a shared-processor partition can be higher and more variable than using a Virtual I/O Server in a dedicated partition. For additional information about the performance differences between dedicated partitions and shared-processor partitions, see Virtual SCSI sizing considerations.

The following table identifies latency overheads for different block-size transmissions on both physical disk and logical-volume-backed Virtual SCSI disks.

The average disk-response time increases as the block size increases. The latency increases for a Virtual SCSI operation are relatively greater on smaller block sizes because of their shorter response time.