With data storage a major part of any UK organisation's IT budget, storage pros need to ensure that they gain maximum...
efficiency per pound spent. Performance monitoring allows you to measure data storage efficiency, including data storage capacity, I/O and energy consumption.
You should first ensure that everything is running as it should be. Storage availability (and therefore performance) will be affected if a drive has failed, a controller has begun to rebuild a drive proactively after registering a failing component, or if a defective adapter has caused I/O path bandwidth to be restricted following a failover.
The amount, type and setup of ports is another important consideration. Do ports share backplanes or paths to switching devices or are they dedicated? What is the nature of any errors logged: are they re-tries, or dropped packets or frames? Did any ports renegotiate to a lower-than-optimal speed or have any components failed over to a different path or not been returned to their original path after failover?
Performance can be impacted by the type of drive used, the number of drives used as well as issues at the ports on the storage subsystem and the host server. Depending on the RAID level used, resources could be consumed by parity as well as by disk repair tasks or controllers that are not correctly load balanced.
Top performance monitoring tips
- Determine baseline performance during normal periods of operation.
- Compare baseline performance to that obtained when performance has dropped.
- Check RAID, I/O path and subsystem configuration for possible quick and easy changes that will improve performance characteristics.
- Higher performing servers will require faster storage subsystem and networking.
- Use tiered storage to achieve appropriate performance, availability and energy usage requirements.
- Maintain perspective: bear in mind that apparent bottlenecks in storage performance could arise from issues at the server, the pairing of components of disparate performance (such as a solid-state drive with a slow controller) and that high IOPS could mask poor latency. And don't forget to eliminate device failures as a cause of poor performance.
How to measure data storage performance
You can benchmark your storage by metrics associated with business-related operations, such as I/O performance. You can also measure performance in terms of, for example, energy consumption. Then there are compound metrics -- such as cost per GB of capacity, cost per I/O operation, GB capacity per watt and I/O operations per watt -- that determine performance per pound spent on energy and equipment costs.
Examples of data storage performance metrics include the following:
- Capacity utilisation: in terms of percent/GB of space used, as well as subcategories such as raw, formatted, free, allocated or allocated not used
- I/O per second (IOPS)
- Access time: read, write, random
- Energy usage: from macro (subsystem) to micro (device or component)
- Mean time between failure (MTBF)
- Mean time to repair or replace (MTTR) failed subsystems/components
- Recovery point objective (RPO): The point in time to which you want data restored
- Recovery time objective (RTO): The time period in which data to the point required by the RPO must be restored.
Storage performance tools
Storage performance monitoring software is available from third parties and as storage product-specific tools from vendors.
Benchmark test results allow you to see what products may be best suited to your environment and to compare your installed hardware and software with its known best possible performance.
Benchmark testing of storage and backup products is available from the following:
The Storage Performance Council (SPC), which develops benchmarks for products that range from disk and tape storage subsystems and storage network equipment to data backup and archiving software. SPC benchmarks test storage products in terms of performance in a range of computing environments – e.g., database, mail server, large file, video on-demand – in a number of configurations as well as in terms of energy usage.
The Microsoft Exchange Solution Reviewed Program (ESRP), which provides test results for storage products in Microsoft Exchange server environments of different sizes.
Points to note when measuring storage performance
Ensure the metric you use is appropriate to whether what you are measuring is an idle or active process. Storage for transactional data, for example, needs to measured as an active process and must be determined in terms of metrics such as activity per-watt per-footprint cost. Inactive data, such as backups or archived data, should be examined on a cost per-capacity basis.
Remember to compare like with like. Equipment that is optimised for data retention and low activity will demonstrate good capacity per watt metrics, but a low number of I/Os per watt. By the same token, media that is optimised for active, high I/O performance will show an I/O per-watt rating, but may not rate so highly when capacity is brought into the equation.
Bottlenecks can also occur at the server, and issues include inadequate CPU and memory to deal with I/O operations, bus contention and adapter setup. Application configuration and size and nature of the I/O – such as a large sequential transfer - should also be considered.
Performance benchmarks for server products can be obtained from the Transaction Processing Performance Council (TPC) and the Standard Performance Evaluation Corporation (SPEC).