Storage Performance Metrics for the Cloud – Throughput, Seek Time, and Latency
Storage benchmarks released by data storage vendors are often taken as ideal benchmarks. However, we need to remember that these benchmarks may or may not match real-world storage metrics. It is, therefore, important to have an understanding of storage performance metrics. This kind of understanding is even more important when it comes to cloud backup and storage, the issue becomes all the more urgent and demanding.
Data is stored on disks in the cloud and is transferred from disks on the server over the Internet. Consequently, the most basic backup and storage measure is the “throughput.” The throughput is measured in megabytes per second (Mbps). When data moves from source to destination, the source read speed and the subsequent disk write speed, taken together, are described as the throughput. Buffers and buses that intervene (and are considered necessary) may slow down the actual read or write process. The geography of the spinning disks and the movement of the write head may further retard the speed of read-write across the disks. This brings us to the concept of “seek time” and “latency.”
The spinning speed of the disk is measured as “seek time.” This is noted in “milliseconds per rotation” for each type of hard disk drive.
Latency is the measure of experienced delays. Latency creates limitations to the maximum amount of data that can be transmitted over a communication channel. This can affect the amount of data that can be backed up or restored or accessed by an end user.
When flash drives and solid-state drives are used, latency and seek times are absent for read operations. However, write operations can be time consuming and latency can be unpredictable. The benchmark must take into consideration the latency of the write operations and must judiciously average out the times. Where read operations are more frequent than write operations, the measure can be skewed towards read operations.
A practical way of approaching the problem of benchmarking for cloud computing would be to use application-based benchmarks. This will give users a realistic view of transmission times and latencies and allow apple-to-apple comparisons. The performance expectations can be based on actual observations and not on synthetic and theoretical benchmarks.
However, it should be remembered that applications are frequently updated and changed. This means observed benchmarks may have to be revised upward or downward with every change, or at least revised at periodic and predetermined intervals. Alternately, benchmarks given out by the Standard Performance Evaluation Corporation (SPEC) may be used as a starting point. SPEC creates benchmarks by stimulating input/output operations for a given application; its resulting benchmarks are fairly accurate for a given workload.
When it comes to cloud backup and storage services, developing storage performance metrics is very important. Prospective users of the cloud backup service should not shy away from asking the vendor for performance metrics that list all possible performance indicators. Vendors should be more than happy to show you a side-by-side comparison with their competitors. However, check to make sure these are actual apple-to-apple comparisons, and look out for signs that metrics were selected to favor a particular vendor.