Block-size: To measure IOPS, use a block-size which matches the disk sector size as this will give the highest count of efficient transactions. This value should be 4kb. Picking a smaller block-size will be inefficient as the whole 4k disk sector will still have to be read/written to; picking a larger block size won’t provide a valid measurement of IOPS.
Queue-depth: Provide at least 64 outstanding IO’s (also known as QD). Scale the queue-depth through 96, 128 and 256, re-running the test each time to see where the performance flattens off.
Test file size: choose a test file size which will be bigger than the first-level cache. On a PERC H700 and H710 this is either 512MB or 1GB, on PERC H710P this is 1GB. Smaller file sizes will allow the controller to carry out all IO operations in cache, providing an invalid result.
Cache warm-up: CacheCade caches read operations only. So you need to warm-up the cache by running the same benchmark to create significant numbers of reads from the test file before launching a performance test. The larger the data-set (test file), the longer the warm-up. Some performance tools such as fio in Linux provide a ramp-time option to allow for this.
We also have a more detailed description of the above available. Visit and ask for support in our Communities Create an online support Request
Need more help?
Find additional PowerEdge and PowerVault articles
Visit and ask for support in our Communities
Create an online support Request
上次修改日期： 07/27/2018 05:43 AM