When it comes to increasing spindle speed, Seagate Technology has undoubtedly been the pioneer. Announced over ten years ago, the company's original Elite was the first to reach a then-scorching 5400 rpm spindle speed. Four years later (1993), Seagate introduced the Barracuda, the first hard disk to feature 7200 rpm operation. Most recently, in late 1996, the manufacturer introduced the Cheetah, the first drive to feature 10k rpm speeds. Two years ago, Hitachi Ltd caught the industry's attention with its Pegasus, a drive featuring a 12k rpm spindle speed. As an incremental upgrade, however, Hitachi's initiative never quite caught on... and 10k operation still anchored the high-end. Finally, however, in the year 2000, it has been left once again to Seagate to ascend towards the next level of performance. Enter the Cheetah X15.
The X15 is the first hard drive to feature a 15,000 rpm spindle speed. Unlike Hitachi's attempt at 12k operation, 15k speeds are a significant improvement, along the lines of going from 5400 rpm to 7200 rpm or from 7200 to 10k speeds. With other manufacturers gearing up for eventual migration, 15k seems here to stay.
What exactly do spindle speed increases deliver? As many veteran StorageReview.com readers are surely aware, sending and retrieving data to and from the hard disk consists of two primary stages:
- Getting the read/write heads into position to read or write the data ("Positioning," as Charles Kozierok of the PC Guide calls it)
- Reading the data from or writing the data to the platter ("Transfer")
While there are some notable exceptions, in this day and age, taking into account the current relative speeds of hard disk mechanisms and the disk access patterns of contemporary operating systems, the former (positioning) is clearly the dominant bottleneck in the vast majority of applications. It is thus of paramount importance to minimize the time spent positioning the heads. Such reductions in many cases will result in directly proportional increases in overall hard disk speeds.
One can further divide the act of "positioning" into two discrete sections:
- Moving the actuator (on which the heads are mounted) into place over the correct track
- Waiting for the correct sectors (where the desired data is to be read from or written to) to rotate under the heads
The former is commonly known as "seek time
," the later, "rotational latency
." The sum of the two figures is "access time
." As we move from generation to generation, at least in the SCSI
hard disk domain, we've been blessed with a downward trend in seek times. Witness, for example, the evolution of the 10k rpm drive. Seagate introduced the first-generation Cheetah 4LP
with a seek time of 7.7 milliseconds. The second-generation Cheetah 9LP
featured a seek time of 5.2 milliseconds. The third-generation-class IBM Ultrastar 18LZX
shaved its average seek time to 4.9 milliseconds. Finally, the fourth-generation equivalent Quantum Atlas 10k II
has reached a svelte 4.7ms seek.
However, each of the four drives, whether we're talking the 1997 Cheetah 4LP or the 2000 Quantum Atlas 10k II, features a rotational latency of 3 milliseconds. This is, of course, due to the models' consistent spindle speeds. Rotational latency, after all, is a direct derivative of spindle speed. 10000 rotations per minute translates into 6 milliseconds per rotation. The best-case latency scenario would involve the required sectors (either containing the data to be read or the space for data to be written to) passing under the heads as soon as the actuator arrives at the proper track. Conversely, a worst case scenario would have the same sectors just past the heads, requiring a full 6 millisecond rotation before the correct sectors re-arrived under the heads. The average case would be the average between the best and worst cases... 3ms.
Decreases in rotational latency are therefore much more rare, occurring only when spindle speeds are increased. A 15k spindle speed results in the first significant improvement we've seen to rotational latency in four years, yielding a figure of 2 milliseconds. Combining this with the average seek time in today's top drives of 5 milliseconds creates an average access time of 7 milliseconds. When access times are this low, a 1 millisecond improvement is significant... it results in 14% faster accesses.
Of course, Seagate would never let a revolutionary (no pun intended) spindle speed increase to go unaccompanied by anything less than a state-of-the-art seek time. The manufacturer has thus reduced the platter size of the Cheetah X15 to about 2.6 inches, down from the 10k rpm standard of 3.0 inches (which in itself is a reduction in size when compared to the 3.5" platters found in most 7200rpm and 5400rpm drives). As a result, the X15 sports a seek time of just 3.9 milliseconds. Take this seek time and combine it with the drive's low rotational latency and the result is an access time of 6 milliseconds... an improvement of 30% or so over current 10k rpm models!
Some skeptics may wonder why Seagate chose to further reduce platter size to achieve its decrease in seek time rather than applying more power to the actuator and creating faster seeks. After all, larger platters would result in the dual benefit of larger capacities and faster sequential transfer rates.
Through interviews with many of its top customers, the company found that its drives were being used with partitions that spanned only one-half or even one-third of total capacity. The reason? Clients wanted to improve on drive seek times by restricting seeks to only a fraction of the platter's span.
Combine this with Seagate's desire to have the drive integrate into all situations (noise and heat wise) where the third-generation Cheetah 18LP worked. All other things being equal, a higher spindle speed will require a more powerful motor, which will result in both more noise and more heat. Smaller platters, however, mean less mass that must be spun at the higher rate, thus evening things out quite a bit. Summarized:
- Lower Seek Times
- Lower Noise Levels
- Lower Heat Levels
- Lower Transfer Rates
- Lower Capacities
Let's take a moment to examine the sequential transfer rate (STR) situation. As we've stated before, only a small handful of applications are bottlenecked by STR. The vast majority are access time dependent. Secondly, the smaller sector-per-track count (one of two major factors in determining STR) in the outer zones of the reduced diameter platters in the X15 is countered by the drive's 15k rpm operation (the other major factor). Finally, keep in mind that the minimum STR, that of the innermost zone, should be well above what we've seen in the past. After all, the X15's inner zone benefits from a linear data density that is state-of-the-art combined with the higher spindle speed... i.e., both the benefits and none of the disadvantages.
The Ultra160/m Cheetah X15 is available in just one size. Its five platters, each storing 3.7 gigs of data, total up for 18.4 gigs of capacity. No, not huge by today's standards, but certainly enough capacity to satisfy the drive's target market: e-commerce servers, transaction processing, data mining, etc. Not to mention rip-roaring power users. After all, no power user in his right mind would have the OS, apps, data, and swapfile all on the same drive, right?
A 4 megabyte buffer rounds out the package. As is standard for Seagate's offerings, an "A/V Ready" version equipped with a 16 meg buffer will also be available. The drive features a standard five-year warranty. Despite the novelty of its design, Seagate maintains through its MTBF figures that the X15 is every bit as reliable as the 10k rpm workhorses that preceded it.
Now then, 'tis the moment that we've been waiting for: How does the Cheetah X15 perform? Come with us as we examine!
WB99/Win2k Low-Level Measurements