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 theory, there should be no difference between the sequential read and the sequential write performance of any HDD, the linear velocity of the particles on the tracks moving under the head is the only factor that really matters. In reality, there are a few additional issues that need to be factored in. The first factor is the question of how the performance is measured and, moreover, under which operating conditions.

If a true sequential read of large blocks is used, the only thing that matters will be the media performance and READ and WRITE performance will come out the same. If there is more randomness in the test file, the WRITE performance will usually be better than the read performance for the simple reason that the data are going through the cache first where they can be reordered for the most economic pattern of writing them to the platters. This reordering uses simple compare functions of the logical block addresses (LBAs) which have nothing to do with command queuing per se even though the effect is somewhat comparable.

HDTach sequential READ and WRITE performance of an IBM 120GXP (123 GB HDD). The sequential WRITE performance is shown as roughly 1/2 of the sequential READs. If this were true, the rotational speed of the HDD on READs would have to be twice as high as on WRITEs and we know for a fact that this is not the case. In other words, what we are looking at is not the sequential performance but an artifact showing the age of a benchmark used on drives it was never intended to be used on - we will get into a more detailed analysis later.

ATTO Disk Benchmark using a RAID Level0 configuration of two Seagate Barracuda 7200.7 on an Intel ICH5R (ASUS P4P800) controller. The relevant part of the figure is the lower half, showing the WRITEs pulling slightly ahead until the test block size compared to the "Command Overhead" increases sufficiently to saturate the media (effective internal) performance. Note that both READ and WRITE are returning the same performance numbers, which is exactly what it should be.

There is, however, the issue of the operating system that needs to be taken into account. WindowsXP, for example, uses a flag to disable the write cache and, instead, uses forced unit access (FUA) to prevent data loss in case of sudden power loss. In that case, some benchmarks that don't really measure sequential READs or WRITES will show a lower WRITE performance than the READ performance on the same drive. What it comes down to in most cases is the question of what benchmark is used but the take-home message is that in all cases, sequential READ and WRITE performance should be at least very similar. If not, something is severely wrong with either the benchmark or the drive.

next page:    => Seek and Random Access =>