The Net's most popular PC hardware websites, seem to have forgotten about accurate benchmarking of Hard Disks, so I decided to test some of my own HDs.

I selected the following devices:
- Maxtor UDMA100 DiamondMax 60 GB (5400 rpm)
- Maxtor UDMA100 DiamondMax 40 GB (7200 rpm)
- Western Digital UWSCSI2 LVD 18 GB (10000 rpm)
- Quantum SE SCSI2 8 GB (5400 rpm)

The IDE devices have been tested as "masters" and as "slaves". There are some interesting results, but first let me describe what is relevant for the test.

The test workstation runs two IDE buses and two SCSI buses.

Each IDE bus supports two devices, one configured as "master", the other configured as "slaves". "Master" devices have priority, when requesting the processor's attention, so HDs configured like that, are expected to have better performance - but how much better? This is one of the question that gets solved by the end of the test.

Each SCSI bus supports up to seven devices, including the SCSI controller; but there are many SCSI flavors...

I am running a "regular" SCSI2 bus, and a less usual UWSCSI2 LVD bus (Ultra Wide SCSI2 with Low Voltage Differential support).

The old Quantum 8 GB HD is a "regular" SCSI2 device and is installed on a fully loaded bus with six other friends. How does the SCSI bus load affect a single device's performance? This test does *not* answer that question.

The more recent ten thousand rotations-per-minute (10000 rpm) Western Digital HD, is alone on the UWSCSI2 LVD bus. Again, this test won't answer if the LVD HD's performance would be different, on a different bus load.

The tests were performed on a Dual P3@733 Mhz machine, with 1 GB RAM, on a TYAN S1833D motherboard (VIA VT82C694X chipset). The SCSI controller is an Adaptec AHA-2940U2W, capable of controlling the two SCSI buses.

The 7200 rpm 40 GB IDE HDs were/are installed on the same (primary IDE) bus; the 5400 rpm 60 GB IDE HDs were/are installed on the same (secondary IDE) bus.

Here are the benchmarking graphics (click on the links):

IDE HD11 (bus IDE1, primary); IDE HD12 (bus IDE1, secondary); IDE HD21; IDE HD22; Western Digital 10K RPM UWSCSI2 LVD; Quantum SE SCSI2.

The charts show that:

#1) the fastest HD, with both the lower random access times and the higher read data bursting, is the Western Digital, running on the UWSCSI2 LVD bus. This HD has everything going for it: it is alone in its bus, it spins very fast (and very hot) at 10k rpm, and it has small capacity plates, that do translate in lower access times.

#2) size seem to matter. Notice how the access times grow, as any HD needs accessing data that lives on its outside (top capacity) - as a consequence, HD benchmarking usually states that bigger drives are "slower", on average. But this is NOT accurate - it is highly dependent on the data size and on the data contiguity.

For example, if you have a large contiguous file, accessing it will probably be faster, from a large HD, than from a smaller one, having the exact same technology, just because the reading head runs across a higher density surface (ie it needs to cross a "smaller field").

The above statement is suggested by the read burst speed for the 60 GB HDs: despite being benchmarked as twice slower on the average access times, they nearly match the data speed of drives that spin 50% faster! How can it be? Answer: contiguous data and higher logical data density per physical inch.

#3) "Master" or "slave" doesn't seem to matter. When the drives are identical, they'll perform the same.

#4) UWSCSI2 LVD is much better than IDE, featuring nearly 50% faster access times, and slightly better read burst speeds, even on relatively low density HDs. However, in practical terms, most users won't feel the performance difference, even after forking twice the money :). My advice is to invest in one LVD for storing crucial files (the operating system, for example), and on IDE for relatively passive storage.

That's all. For short: the best value-for-money are the very large IDE HDs.