On X2, X3, and X4 systems, high performance disks can execute about 300 small I/Os per second (IOPS) without a
large increase in response time (peak performance is actually above 400 IOPS), or 50,000 IOPS on a full rack. A
large I/O is roughly 3 times as expensive as a small IO. You can determine approximate disk utilization by counting
the total small I/Os and adding the total large I/Os multiplied by 3. Then compare this count to the 300 IOPS
threshold to determine utilization. For 4TB high capacity disks, the IOPS are around 120 and the multiplier for large
I/Os is about 2.
4TB high capacity disks have IOPS around 190 or 32,000 IOPS for a full rack and should also use a 2x multiplier for
large I/Os. For additional Exadata capacity details please reference the Oracle Exadata Database Machine Data
Sheets.
High disk latencies are not necessarily a problem – it depends on how the application is impacted. For a Data
Warehouse it may be perfectly fine for the disks to be running at maximum throughput and latency when processing
queries.