6.3.9 Monitoring Cell Disk I/O
When database performance issues are related to I/O load on the Exadata storage servers, typically there will be increased latencies in the I/O-related wait events, and increased database time in the User I/O or System I/O wait classes. If the increased database latencies are due to the performance of the Exadata storage servers, then the increased latencies will also be visible in the cell-side statistics. If you have comprehensive baseline statistics for periods when the system is performing well, then you can compare the baseline statistics with other observations to identify differences and explain the situation.
For example, if Oracle Database reports increased cell single block physical
read latencies, you can then check the statistics from the storage servers
and compare them to a baseline to determine if the cause is increased latency on flash
devices, more disk I/O, or some other cause. If the statistics show an increase in disk
I/O requests, that may be related to a change in Exadata Smart Flash Cache, which would prompt you to review the Exadata Smart Flash Cache statistics. On the other
hand, if the statistics show a latency increase for small reads from flash, then this
may be caused by a change in the I/O pattern, and understanding the type of I/O that has
increased (small reads, small writes, large reads, or large writes) can help to drive
further investigation.
Typically, in an Exadata environment, an even workload distribution is expected across all cells and all disks. If one cell or one disk is doing more work than the others, then you should investigate further, as that cell or disk has the potential of slowing down the entire system.