Stardate 20021024.0116

(On Screen): I could begin this with another putdown of Mac users, but I'll resist the urge. Brian Tiemann, my favorite Mac fan, points to a benchmark done by XBit Labs which compared the performance of a USB2 hard disk, a Firewire disk, and an ATA/100 disk. It uses several different benchmarks, and on every test Firewire came up with a higher score than USB2 did.

I use USB2 for hard disks. I currently have three of them, in fact. They were purchased at different times, and since each time I went back to Fry's they no longer stocked the previous one, it turns out that they're all different brands, which is convenient in this case. So I decided to try to see what kinds of numbers I'd get.

Brian may or may not know about XBit. I've been a reader of their site off and on for years, and they're honest and conscientious about testing. They're also located in Moscow, and because of that, and because none of them are wealthy, and because they don't tend to get a lot of freebies, they're using a lot of obsolescent equipment. As proof of their honesty, they provide the specifications for the computer used to do the testing. A "Coppermine" is a Pentium III, and a 600 MHz Coppermine is a processor which was state-of-the-art about four years ago. The mobo in it came out in 1999; it's based on the now-legendary, now-obsolete i440BX chipset from Intel. (Those were the days!) The computer that XBit used is long since obsolescent, and I fear that it may have affected their results.

They ran four different benchmarks, and I downloaded them all, but three of the four turned out to be useless for my purposes. HDTach refuses to run on NT unless I register (and pay) and I'm not that dedicated. Intel IOMeter and XBit's own home-grown FC-Test require extensive setup, and there was too much of a chance I would not set them up properly and thus that my numbers would not be comparable to theirs.

That left WinBench99, which is a canned test: you push a couple of buttons and it runs and provides you some numbers. There are two versions of that (1.2 and 2.0) and XBit's page isn't totally clear on which they were using. I ended up finding both versions and downloaded them both. First I installed the 2.0 and ran the same tests as XBit, and then I uninstalled it and ran them again using the 1.2 version.

In addition to my three USB2 drives, my workstation uses an Ultra160 SCSI RAID card to talk to two pairs of drives (total 4) mounted inside the chassis. They're running in RAID 1 ("mirror") mode which means that each disk in a pair holds exactly the same information as the other does. I'm certain that costs me a certain amount of performance but it isn't terrible. I ran the benchmarks on a representative partition on each pair of drives, just for reference purposes.

The first USB2 drive I got, way back when, was an 80G drive from ACom, and all I know about it is that it's a 5400 RPM drive. The second drive is 120G and carries the Maxtor brandname, and is also a 5400 RPM drive. The third is 120G and came from Western Digital, and it's 7200 RPM.

The drive that XBit tested was actually a hybrid. They bought an adaptor and put their own drive into it. The adaptor provides power and support and interface electronics, but it was a no-name, and it's not clear who designed the electronics in it. It's entirely possible that they loused up the performance. My three units are all closed-case, and given that the latter two are produced by name-brand vendors, there's more reason to believe that they put more attention into the interface electronics and the software which drives it.

So I ran my tests and then compared them to XBit's results. Problem is, I can't make sense of XBit's numbers, and I think that the problem is that the timer on their motherboard isn't as granular, or else as accurate, as mine is. Here are the results of my testing. It's also possible that their slow CPU aliased the test, or that there's something else going on that I don't understand. So it doesn't seem to be possible for me to compare my USB2 numbers to XBit's Firewire numbers to learn anything.

Nonetheless, it's possible to learn something from my tests of my three units, because the ACOM unit is drastically slower than either of the other two. In particular, the 5400RPM Maxtor unit is much faster than the 5400RPM ACOM drive, with the 7200RPM Western Digital drive being marginally faster than the Maxtor unit.

What that suggests is that differences in implementation are the predominant factor in determining the effective speed of the drive. I think that the controller chip in the ACOM unit must be substantially inferior to whatever it is that Maxtor and Western Digital used.

XBit's test involved exactly one firewire drive and one USB2 drive, and there's no way of knowing what other factors may have influenced the test so that the USB2 drive ended up slower. In particular, we don't have the slightest idea what kind of controller chip was in the no-name adaptor. XBit's test proves nothing about the merits of USB2 relative to Firewire for this application.