Does Size Really Matter Or Is It Performance That Counts?

MARK: I've been trying to figure out the most cost-effective ways to improve system performance, when I saw a presentation given in June by Maxtor. They concluded that a faster HDD is a more cost-effective way to increase total system performance than using a faster processor or adding more RAM.

HAL: I might disagree about RAM. That's your scratch pad and so many applications make demands on it now that you just can't have too much RAM. Even 128MB, which seemed enormous a few years ago, could be too little if you have to work with image files or graphics all the time.

MARK: What about the processor?

HAL: I think that's less important, unless your application is very computationally intensive. If you're modeling the weather or locating pockets of oil underground--

MARK: Or playing the latest generation of video games or using a speech recognition program...

HAL: Right. Then, you need as many "mips" as you can afford. The rest of us can get by with 233MHz or 300MHz, or whatever the clone-maker ships with the system.

MARK: Unless that system is a server, in which case you probably need one of the new 500MHz or 550MHz, or even 600MHZ CPUs, or maybe even dual processor systems. What about the hard drive?

HAL: Gotta go with the flow. Get as big a drive as you can afford. It can't be too big.

MARK: I've got to disagree. Putting too much data on any one spindle means you'll be looking into that drive far too often, no matter how many platters are inside. I'll bet that two 9GB hard disk drives give you faster access than one 18GB hard disk drive--especially if your data is located on many different parts of the drive. I'll bet that streaming video streams a lot smoother when it comes from more spindles, too.

HAL: That raises the issue of hard-disk partitions, which are a problem for older Windows systems, but I guess Linux and the high-end Oss doesn't care how big your logical drives are. In any case, I'd still have to say that, capacity-wise, bigger is better.

MARK: More total storage capacity is always best. But I'm sticking with my argument for putting your data onto more spindles. If you use only one hard disk drive, even one of the new 47GB drives, you've got to load all your apps and read all your data off that one HDD, and I think that's asking for trouble. In my opinion, throughput is more important than capacity. I happen to know this from bitter experience.

HAL: Do tell.

MARK: I've got a big hard drive now--22GB--that's divided into six partitions.

HAL: Wow! I've never heard of so many in one drive. How'd you get into such a configuration?

MARK: Actually, I copied over the data from four old, smaller hard disk drives onto the one big one.

HAL: Maybe you shouldn't have.

MARK: Maybe you're right. But it was all IDE/ATAPI. I went down from four connections to one, which opened up three new connections. That was good. I've already used those connections for a DVD-ROM drive, a CD-RW drive, and a LS-120 drive. I couldn't have done that if I hadn't consolidated my four HDD's worth of files into one big drive.

HAL: It sounds like more work than it was worth.

MARK: I might not do it again. But here's something else I learned from experience: putting so much stuff onto a single drive is like putting all your eggs into one basket. If the drive fails, you lose a LOT more data. Further, because there can be so much on a partition, it takes a lot longer to defragment the drive--and you may not want to wait while it's being done. This can result in a huge and really badly fragmented drive and significant drops in performance.

HAL: Sorry. But can you spell "SCSI"?

MARK: I have a SCSI bus too.

HAL: Why didn't you use it? Did you think IDE was going to be faster?

MARK: It's not faster, of course. But at the time, the BIOS on my motherboard couldn't boot from a SCSI drive. Since then, though, I got a new motherboard with a BIOS that does boot from SCSI.

HAL: Okay, but getting back to your first point about HDD capacity, you definitely traded capacity for throughput. I hope you're happy.

MARK: I am now--now that my system performance is optimized all around. And, actually, I got both--a much larger hard drive and a much faster hard drive to boot. The one fast drive probably had much better throughput than the older drives, even if they were all moving data simultaneously.

HAL: You probably did what most people do when they upgrade their HDD: You focused on capacity, which is easier to measure, rather than on throughput. It depends on what you need most, at the time, I guess. It's a paradox: you probably can't measure throughput improvement until your system is running under the new configuration, at which point you've lost your basis for comparison.

MARK: What Maxtor was saying is that going to a faster rotational speed--in this case 7,200rpm--is better than 5,400rpm. It gives you more performance-improvement per dollar than upping the CPU speed or boosting RAM.

HAL: I didn't see the test results, but I guess they have a point, especially because it encourages people to go out and buy Maxtor's 7,200rpm drive.

MARK: You're missing the point, although it's possible that the price delta in moving from a 5400rpm Maxtor drive to a 7200rpm Maxtor drive may be lower than the delta for the equivalent offerings from other manufacturers.

HAL: As the expression goes: "Actual mileage may vary." If you think adding hard-disk capacity is the most effective way to boost system performance, e-mail me at [e-mail].

MARK: And if you think that a speedier hard drive, however you want to measure it, makes the bigger difference, e-mail me at [e-mail].