APPLICATIONS

From the Desk of Geoff Barrall, CTO of BlueArc Corporation -- When talking with suppliers or customers, I find people are very interested in disk performance as it relates to their enterprise applications. The most-often quoted metric measuring disk performance is that of the rotational speed of the disk, which for most major disk vendors defines the difference between high end and lower-end disk lines. Today, these rotational speeds are typically 5,400 RPM, 7,200RPM, 10,000PRM and 15,000RPM. However, in applications where non-sequential data access is required (the vast majority), I would suggest that access time becomes a much more important, yet usually ignored, metric for determining performance. Access time is the combination of three important disk metrics: controller overhead (the time taken by the controller embedded on the disk to process the operation), seek time (the time taken for the disk head to move to the correct disk track) and rotational latency (the time taken for the disk to rotate far enough so the head reaches the data required). For example, consider the below chart: This chart shows both the access time and the data transfer time for the retrieval of an 8k block of data (a very typical operation) from the disk. What is immediately obvious from the chart is that the access time overwhelms the data transfer time in terms of its impact on the overall operation time. One can also clearly see that the key factor in disk performance for this kind of operation is seek time, unaffected by the rotational speed of the disk. For a concrete example of these factors, I visited the Western Digital Web site (www.wdc.com). Western Digital is a major provider of hard drives to personal computers, personal video recorders and set-top boxes. The company's high performance ATA 120GB disk is listed as having a seek time of 8.9ms while having a rotational speed of 7200RPM, while their lower-priced 120GB disk, with a rotational speed of only 5400RPM has almost the same seek time of 9ms. For these two different disks, overall data transfer time may be very similar for most application areas. For the above chart, I used a Fibre Channel disk, yet in the ATA world (see Should ATA Disks Become a Standard for Online Data?) seek times are much higher, often in the 9ms range. More than once, I have heard people comment that there is little difference between an ATA and FC disk at a technical level and while this may be true, with the seek time presenting a difference of 100% between the two technologies there will continue to be a clear difference in overall performance in most applications. But this data does not imply the rotational speed of the disk is a factor that can be completely ignored. Rotational speed has the second-largest impact on a disk's overall access time as well as on the speed of data transfer. Average rotational latency for a 10,000 RPM Seagate drive is reported as 2.99ms and as 2.00ms for a 15,000 RPM disk. Despite this, it is worth noting that the overall access times between the two disks vary by only 10 to 20 percent in real terms even though their rotational speeds vary by 50 percent because seek time (at 4-5ms for Fibre Channel disks and 8-9ms for ATA disks) remains the dominant factor. When considering disk selection for applications where performance is a consideration, I would recommend serious research of all of the factors contributing to overall access time, in addition to rotational speeds and latency. This is especially true when considering new disk technologies such as ATA and SATA in comparison with currently deployed Fibre Channel or SCSI disk systems, or when comparing disk drives across multiple vendors.