Slow storage is the most common cause of Oracle VMware and SQL Server on VMware performance problems in a vSphere environment.
There are two metrics to be considered to determine the average time for an I/O operation to complete:
- The Physical Device Read Latency
- Physical Device Write Latency
They combine to show the time total time necessary for an I/O operation from the time it is submitted to the hardware adapter until a completion notification is sent back to VMware ESX. These metrics therefore represent the performance capabilities of the storage infrastructure. The values for these metrics specified in the problem check for Slow Storage represent response-times beyond which performance problems may exist in the storage infrastructure. In order to understand whether these values represent an actual problem, however, it is necessary to understand the storage workload.
There are three main workload factors that affect the response-time of a storage subsystem:
- I/O arrival-rate
Any storage device configuration will have a finite or maximum rate at which it can handle specific combinations of I/O requests. When they exceed this maximum rate, they will often be buffered at spots along the I/O path. When this occurs, overall response time may be increased and VMware performance can be degraded.
- I/O Size
â€¨The speed at which I/O can be read and the rate of storage interconnects are fixed quantities that cannot be exceeded. This results in large I/O operations taking longer to complete.â€¨A response time that is slow for small transfers should be expected and considered normal for larger operations.
- I/O Locality
â€¨Sequentially stored data accessed by successive I/O request can be completed faster than those that are spread randomly across the array. Read requests to sequentially stored data are also more likely to be completed more out of high-speed data caches.
Most storage device OEMs provide monitoring tools that enable the collection of data so that administrators can monitor these and other workload factors. They will also provided data on recommended configurations and their expected performance levels. If the Slow Storage problem appears, use the monitoring tools provided to collect workload data. Storage response-times should also be compared to expectations for that specified workload. If you can determine that storage response times are unacceptably high, corrective action should be taken.
The solutions for slow storage are the same as those for overloaded storage. You may also want to review the solutions for Overloaded Storage.
Confio IgniteVM helps identify the impact of slow storage for sites running Oracle on VMware, SQL Server on VMware, and other virtual databases. IgniteVM helps DBAs maintain performance and availability on virtual servers. IgniteVM is the only virtualization-aware database monitoring solution.
Learn more about IgniteVM solutions for: