Learn about the technologies behind the Internet with The TCP/IP Guide!|
NOTE: Using robot software to mass-download the site degrades the server and is prohibited. See here for more.
Find The PC Guide helpful? Please consider a donation to The PC Guide Tip Jar. Visa/MC/Paypal accepted.
|View over 750 of my fine art photos any time for free at DesktopScenes.com!|
[ The PC Guide | Systems and Components Reference Guide | Hard Disk Drives | Hard Disk Performance, Quality and Reliability | Hard Disk Performance | Hard Disk Performance Specifications | Transfer Performance Specifications ]
Internal Sustained Transfer Rate (STR)
The media transfer rate is the maximum rate that any particular track of the hard disk can have bits written to it or read from it. However, most transfers from the hard disk involve more than a single track (and the performance of accesses short enough to fit in a single track is typically dominated by positioning concerns more than transfer issues anyway). For real-world transfers of average files, what we are concerned with is the rate at which the drive can transfer data sequentially from multiple tracks and cylinders on the disk. This specification is the drive's sustained transfer rate (sometimes the sequential transfer rate), abbreviated STR.
Sustained transfer rates are most relevant for reflecting the drive's performance when dealing with largish files. It is based upon the drive's media transfer rate, but includes the overheads required for head switch time and cylinder switch time. Also, STR is normally measured in bytes, not bits like the media transfer rate, and includes only data, not the overhead portions of each sector or track. An example: let's say we want to read a 4 MB file from a hard disk that has 300 sectors per track in the zone where the file is located; that's about 0.15 MB per track. If the drive has three platters and six surfaces, this means that if this file is stored sequentially, it will on average occupy 26 tracks over some portion of 5 cylinders. Reading this file in its entirety would require (at least) 25 head switches and 4 cylinder switches.
STR can be calculated from various characteristics of a disk, but this isn't nearly as conceptually simple as calculating a media transfer rate on a single track. Rather than just provide a lengthy formula, I'll try to explain how the calculation is done. A transfer rate is of course data transferred per unit of time. So our equation will be a ratio of data transferred to the time taken to transfer it. Now, to represent a sustained transfer we need to cover an entire cylinder, so we include all the head switches while reading the cylinder, and one cylinder switch time as well (to get us to the next cylinder). The data that is transferred for an entire cylinder read is as follows:
where "number of surfaces" is identical to the number of tracks per cylinder, of course. Now, how much time is taken? First, we of course have to wait for the disk to make one complete revolution for each track read, as the data is read. Then we need to add a number of head switches equal to the number of surfaces less one, and finally, one cylinder switch. So the time taken to transfer an entire cylinder is as follows:
The easiest way to calculate platter revolution is to double the disk's latency specification. The final equation then looks like this:
The result is in bytes per second. Simple, right? ;^) Let's use the same IBM Deskstar 34GXP model that we discussed in the media transfer rate section. This drive has 452 sectors in its outermost zone, and a 7200 RPM spin speed (for latency of 4.17 ms). This family's head switch time is 1.5 ms and cylinder switch time is 2.0 ms. We'll consider the flagship drive that has five platters and hence ten surfaces:
The specification for maximum STR for this drive is in fact 23.4 MB/s. Out of curiosity, let's do the same calculation for the 13.6 GB version of this drive, which has only two platters:
The change is due entirely to the difference between head switch time and cylinder switch time: if they were identical the STRs would be as well. Since the drive with more platters performs a higher ratio of (faster) head switches compared to (slower) cylinder switches, its STR is a bit higher. Still, it's only a difference of less than 1% between the biggest and smallest members of the family.
An important question to consider is how meaningful the STR numbers really are: if you have the drive above, will it really let you read at a rate of about 23 MB/second? I'm sure you won't be shocked if I say "no". There are a number of issues involved. First, since STR is derived directly from the media transfer rate, its value also depends on what part of the disk is being read; larger outer cylinders have the highest STR, smaller inner cylinders have the lowest. Second, there's the matter of whether the access is really sequential. There is a big difference between a 10 MB file that is laid out contiguously on the disk, and one that is fragmented into a dozen pieces. Once you fragment the file, you aren't doing a consecutive data transfer any more. Each fragment of the file introduces the need for an additional positioning step to the location where the next piece starts, which slows the transfer and introduces other factors into the performance measurement. Finally, real-world transfers incur all sorts of penalties due to operating system overhead and other considerations. A good rule of thumb in the computer world is that you never get the theoretical maximum of anything. :^)
STR has in the last few years started to get more attention than it traditionally has--some would say too much. :^) It is important to those who do a lot of work with large files, but not as critical to those who work with a large number of smaller files, which includes many, if not most, Windows users. It is probably best to value it roughly equally with key positioning specifications such as access time.
Sustained transfer rate is affected by just about every internal performance factor you can name. :^) The number of platters influences it by changing the mix of head and cylinder switches; actuator design and controller circuitry affect the switch times; media issues and spindle speed influence the all-important underlying media transfer rates. There's probably no other performance specification that is affected by so many different design factors.
A final point about internal sustained transfer rates vs. external (interface) transfer rates. In order to get the most from the hard disk, the interface must be fast enough to be able to handle the maximum STR of the drive. This is usually not a problem because most disk interfaces have sufficient "headroom" to handle drives that run on them. However, many interfaces are also backward-compatible; for example, you can put the drive discussed above on an older IDE/ATA interface running at 16.6 MB/s. It will work, but clearly you will not get STR of 23 MB/s over that interface. The converse is that putting a drive on an interface much faster than it won't improve performance much; but that's a topic for another section.