High Performance

One of the key features of SAN storage systems is their high performance and low response time compared to other storage structures such as DAS. Ever increasing need for fast and reliable data access mandates efficient, high-performance SAN storage systems.

Performance metrics in data storage systems

The most important metrics for measuring the performance of data storage systems are as follows:

Throughput (Bandwidth)
Bandwidth (BW) is generally referred to the amount of data read/write storage system can handle in one second (in MB/s or GB/s units), demonstrating the capability of storage system in handling sequential requests. The storage bandwidth is a function of the bandwidth of disk array, internal interconnections (such as interconnections between controllers and disk subsystem), and external connections (such as connection between storage system and hosts). The highest bandwidth usage is observed when the requests are sequential with big size (as an example we can note multimedia applications). Storage manufacturers report the storage bandwidth for the sequential big requests that can use the maximum storage bandwidth. In practice, however, the bandwidth may be lower than the reported bandwidth depending on the workload characteristics.
Learn More

HDDs have low read/write performance due to their physical characteristics. This is not a big issue in home applications where the HDD is responsible to just one user. However, in the data storage systems that are responsible to handle multiple hundred thousands of requests per second, the low performance of HDDs becomes a major bottleneck. The enterprise HDDs can deliver 200 MBPS in the ideal situations, while a data storage system needs to provide over 100 GBPS throughput. SAN storage systems provide this high bandwidth using a combination of disk redundancies (such as RAID mechanisms), caching mechanisms, and data tiering mechanisms that mandate using many extra hardware/software modules such as DRAMs, SSDs, and controllers.

Input Output Per Second (IOPS)
Number of transactions per second, known as IOPS, stands for the number of IO operations the storage system can accomplish in one second. IOPS directly depends on workload characteristics such as request size and randomness. When using HDDs, workloads with small and random request result in comparably lower IOPS. An HDD can ideally handle 200 to 300 random requests per second. Using RAID configurations, the IOPS for random requests can increase to 500 and 600. Nowadays, many applications such as banking and social networks desire 30,000 to 50,000 IOPS that cannot be provided just by RAID mechanism, mandating more advanced mechanisms such as caching.
Learn More

IOPS is a function of various parameters such as workload characteristics and data access time (from disk subsystem). For example, a DSS can deliver higher IOPS when the size of requests is relatively small. We need to note that a DSS may deliver a great IOPS with a poor Response Time, when the data access time is high. Hence, IOPS is not enough for reporting the performance of DSS, while Response Time has also important impact on DSS performance. IOPS is typically reported for workloads with small request size (4 and 8 KB) with random access pattern. Following table reports the results of few tests on a single HDD:

IOPS زمان پاسخ (ms) درصد درخواست‌های خواندن درصد درخواست‌های متوالی اندازه درخواست (KB) ردیف
29736 0 100 100 4 1
236 4.2 100 60 4 2
140 7.1 100 30 4 3
100 10 100 0 4 4
293 3.4 60 100 4 5
168 5.9 30 100 4 6
7658 0.1 100 100 16 7
210 4.7 100 60 16 8
130 7.7 100 30 16 9
98 10.1 100 0 16 10
Average Response Time (Delay)
Average response time, usually reported in milliseconds (ms), is defined as the aggregation of time a request sent to SAN storage, queued, and finally served. It is notable that if the response delay surpasses a predefined value, the client system considers it as a request failure. Average response time is in direct relationship with the rate of requests and increases by the increase of request queue. A high average response time can decreases service quality and user satisfaction.
Contact Info
Unit 11,Eastern Part 1,Fatemi Dead End,Habibzadegan St.,Teymouri Square,Tehran,Iran
+98 (21) 66029164
info@hpds.ir

Support