Sizing and measuring.
WHAT TO MEASURE
Operations per second Throughput Latency Queue depth
WHAT MATTERS MOST FOR MOST WORKLOADS
Latency and operations per second, not throughput.
WHY
Typical workloads perform many small operations rather than few large transfers.
WHAT DETERMINES OPERATIONS PER SECOND
The drive technology, primarily.
WHAT THE DIFFERENCE IS
Mechanical drives manage a modest number; solid-state drives manage orders of magnitude more.
WHAT THAT MEANS
Replacing mechanical with solid-state transforms most workloads.
WHAT TO PLAN FOR
Capacity growth, which is relentless Performance degradation as arrays fill
WHY THAT SECOND POINT
Many systems slow substantially when nearly full.
WHAT MARGIN TO KEEP
A meaningful proportion free.
WHAT TO MONITOR
Capacity trend, with alerts well before exhaustion.
WHY WELL BEFORE
Adding storage takes time, and running out causes an outage.
WHAT TO SEPARATE
Workloads with different characteristics, where they interfere.
WHAT TO TEST
Performance under realistic load, not synthetic benchmarks.
