Need clarifications on IOMeter Burstiness Settings (Burst Length + Transfer Delay)
I am trying to understand more clearly how should the Burstiness Settings (Burst Length + Transfer Delay) affect my storage disk performance?
When I do my test (100% Read, 100% Random, 8KB Transfer Size, 16 Oustanding I/Os), my storage gives me 1000 IOPS with an average response time of 12ms.
If I use the same settings but I change the Burst Length to 1000 and the Transfer Delay stays at 0ms, my storage performance drops to 700 IOPS and the average response time increases to 14ms.
Since I kept my transfer delay to 0ms, the workload should be continuous (no burst) and thus the IOPS should not be affected? Am I right?
I didn't find much details in the IOMeter Userguide for the Burstiness setting but I found the following:
"If the Transfer Delay value is 0, the Burst Length is not significant because there is no delay between bursts."
So based on this information and my understanding of the Burstiness settings, why do I not have the same performance for those two tests?
If someone has more knowledge on those settings, I would gladly appreciate your sharing.
(Be aware that all other IOmeter settings were kept to their default values. The same logical device was used for both tests and no other workloads were generated on this storage subsystem during those tests.)