Examples of metrics and measurements for storage efficiency and optimization include the following:
- Macro (e.g., facilities such as power usage effectiveness) and micro (device or component level)
- Time (performance or activity) vs. availability vs. space (capacity)
- Performance metrics, including IOPS, bandwidth, and response time or latency
- Additional performance metrics, including reads, writes, random, sequential or IO size
- Storage capacity metrics, including percent utilization as well as reduction ratios
- Other capacity metrics, including raw, formatted, free, allocated or allocated not used
Metrics can be obtained from in-house, third-party, or operating system and application-specific tools. Other metrics can be estimated or simulated; for example, benchmarks running specific workloads such as those from the Transaction Processing Performance Council (TPC), Storage Performance Council (SPC), Standard Performance Evaluation Corporation (SPEC) or Microsoft Exchange Solution Reviewed Program (ESRP).
Compound metrics, those made up of multiple metrics, include cost per GB and cost per IOP, along with capacity per watt or activity per watt, such as IOPS or bandwidth per watt of energy used.
A list of common storage performance metrics
Here is a list of common storage performance metrics:
- IOPS: I/O operations per second where the I/O can be of various size
- Latency: The response time where lower is better for time-sensitive applications
- MTBF: Mean time between failures indicates reliability or availability
- MTTR: Mean time to repair or replace a failed component or storage device
- Quality of Service (QoS): Refers to performance, availability or general service experience
- Recovery point objective (RPO): To what point in time is data saved or lost
- Recovery time objective (RTO): How quickly data or applications can be made available
- SPC: Storage Performance Council workload (IOP, bandwidth and others)
- TPC: Transaction Processing Council workload comparisons
Other metrics include uptime, planned or unplanned downtime, errors or defects, and missed windows for data protection or other infrastructure resource management tasks.
Remember to keep idle and active modes of operation in perspective when comparing tiered storage. Applications that rely on performance or data access need to be compared on an activity basis, while applications and data that are focused more on data retention should be compared on a cost per-capacity basis. For example, active, online and primary data that needs to provide performance should be looked at in terms of activity per-watt per-footprint cost, while inactive or idle data should be looked at on a capacity per-watt per-footprint cost basis.
Given that productivity is also a tenet of storage efficiency, metrics that shed light on how effectively resources are being used are important. For example, QoS, performance, transactions, IOPS, files serviced or other activity-based metrics should be looked at to determine how effective and productive storage resources are.
Tips for using data storage resource metrics
Here are three other storage efficiency tips to remember:
- Look beyond cost per-capacity comparisons
- Remember that GB per watt can mean capacity or performance bandwidth
- While hit rates may indicate good utilization, they may not necessarily mean effective performance
It can be easy to end up with an apples-to-oranges comparison when looking at different storage products optimized for idle or low activity that may have a good capacity per watt, but poor performance and low IOPS or bandwidth per watt. Likewise, a high-performance storage system may have good IOPS or bandwidth per watt, but may not be as attractive when compared on a capacity basis.
Remember that more information will have to be processed, stored and protected in multiple locations and at a lower cost in the future. Therefore, performance efficiency can enable more effective storage capacity at a given QoS level, for both active and idle storage.
This was first published in November 2009