Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ Search Prices:

Random Access of data on a hard disc is a combination of track-to-track or seek latency and rotational latencies. If the accesses are truly random, the average rotational latency should be exactly ½ of a full rotation. In the case of a 7200 rpm drive, there will be a random distribution between 0 and 8.33 ms with an average of ½ of 8.33msec or 4.17 msec. In the case of a 10k rpm drive, the average rotational latency will be 3 msec.

In addition to the rotational latency, there will be the seek latency. The seek latency will increase with the lateral distance the head has to cover moving between the OD and the ID. Keep in mind, however, that there are only two positions on the drive, namely the OD and the ID from which the full span of the lateral movement will even be possible. If all relevant parameters, that is the starting position and the random displacement distance within the possible range (depending on the starting position) are plotted, the result is not what one expect in form of a Gaussian Bell distribution but a triangular probability distribution function (PDF) with the highest probability for single-track displacements. As a result, the average seek length will be 1/3 of the maximum length.

Seek Length Probability Distribution Function: the probability for a full "swing" across the entire platter is almost zero, most seeks (~45%) will only go to the nearest neighbor track and the average of all seeks is going to be less than 30% of the total platter length.

So far, so good and even though we'll show later that there is little relevance in day to day use, it looks like a valid and reproducible test. Unfortunately, this is not the case, since there are three different standards for the execution of a seek command. The reason is that WRITEs require a higher precision positioning of the head over the desired track than a read. That is, READs require less energy than WRITEs and, moreover, the higher WRITE power can easily corrupt data on neighboring tracks if the head position is not dead center on the target track but somewhere in between two adjacent cylinders. In addition, on WRITEs, the head needs to be lowered more than on READs.

In other words, the different parameters that can be used for the execution of a "Seek" command are:

• READ precision head positioning
• WRITE precision head positioning
• Don't care (see next paragraph)

For seeks, it therefore depends on the algorithm used by the drive manufacturer, namely whether it is WRITE or READ precision that is implemented, how fast the command will be executed. A third possibility is the "Don't Care" algorithm that allows the drive to signal back that the command was executed regardless of what actually happened. Bottom line is that there is no way of telling which parameters are used and, therefore, comparing Random Access Latencies from one drive to another can be worse than apples and oranges.

next page:    => Seek Optimizations =>