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:

IN THE PREVIOUS ARTICLE, I covered some of the basic parameters of hard disk drive internal performance, that is, how the geometry of the data distribution on the HDD media translates into actually measurable performance in the form of sequential transfers. Sequential transfers shown were, however, only based on READs rather than on READs and WRITEs. Are there differences between the two processes and if so, what is it that one can expect? In addition, there is a so-called average sequential performance, probably the most misunderstood and misused hard disc parameter out there and a good place to start:

Average Sequential Transfer

A common way to determine the effective internal performance of hard discs is to run sequential reads across the entire length of the platter, that is, from the fastest tracks at the OD to the slowest tracks at the ID. If this is done correctly, the benchmark will measure every LBA on the platter. This, however, does take quite a bit of time, depending on the capacity of the drive. To give an idea, measuring a 20 GB drive at 40 MB/sec average will take 500 sec, a 120 GB drive will take about 1 hour to run. Hence, many benchmarks use shortcuts, that is H2Dbench uses 1000 distinct spot check points in regular intervals across the lenght of the platter, HDTach divides the entire platter into blocks of 65532 kBytes size and runs a brief test sequence within each. The inner cylinders are slower than the outer ones and somebody came up with the clever idea of using the average across the entire length as a metric to define a given drive's media performance. Unfortunately, there are some serious problems with this approach.

One good example is the Seagate Barracuda V which comes in different capacity flavors. That is, one version is 80 GB total drive capacity, another one is 120 GB with the first using three heads and the latter using four heads. What this comes down to is that the 80 GB drive is using 26 GB per head, whereas the 120 GB version is using 30 GB per head. The platter geometry in this case is identical, however, the platters of the 80 GB version are shorter, meaning that the inner cylinders are not being used.

If we look at the WB99 disk inspection test (and for simplicity reason I ran it on a single head only), it is quite obvious that the slowest tracks are just beyond the 26 GB cutoff. This means that the 120 GB drive has, in addition to the faster tracks, a number of slower tracks at the inner diameter of the platters. If we now measure the sequential performance across all zones, the "extended" platters will return a lower average than the shorter platter, however, for what it is worth, the actual drive performance at any given point will be the same.

Sequential performance across the entire platter length from OD to ID on a single head Seagate Barracuda SATA-V with a 30 GB/platter capacity. The same platter length is used in the 120 GB model whereas the 80 GB version only uses 26.6 GB per platter and "discards" the innermost tracks. The capacity of either platter configuration is shown as exclusive / inclusive the hilighted area on the right (underneath the graph). The average sequential performance of the two drives is given as the red (30GB/platter) or the yellow (26.6GB/platter) and it is clear that the yellow line characterizes a much faster drive ... Well. actually not, the two drives will perform identically -- it is only the metric that is misleading, showing one drive roughly 8% faster.

Bottom-line is that the average performance is a weighed arithmetic mean across all zones that is interesting for statistical and drive architectural purposes but has little bearing on real drive performance and absolutely nothing to do with e.g. controller performance. Even within the same model of the same manufacturer, different capacity drives will have different averages, which solely reflects the use or non-use of the innermost tracks and nothing else.

next page:    => Hard Disc Drive Architecture II: READ Vs. WRITE =>