This is a response to the eWeek editorial Bring DIY Systems to Work. Nice theory. Unfortunately, lab theory and the real world don’t always mesh.
I like building PCs. I built my first PC in early 1994, back when everything was on a separate card and you had to set interrupts and DMA channels using jumpers and DIP switches and in most cases you had to tell the BIOS exactly what size drive was in it–it wouldn’t detect anything for you. I built my main PC at home myself. I built my secondary and tertiary PCs at home myself too. And my girlfriend’s PC, and my mom’s PC, and my sister’s PC.
Get the idea?
For the first couple of years of my career, I did DIY at work. It made sense then. IBM was selling us one-size-fits-all PCs with a lot more capability than most secretaries needed. As this was a state university, our budget was being cut, and it was 1998 and the PC on most people’s desk was still a 486–sometimes the crippled IBM 486SLC2 which was really just a 386SX with the 486 instruction set added, clock doubled to 66 MHz and given a supersized L1 cache, but with a 16 megabyte address space–so we were stuck.
My proposal was that since we couldn’t afford $1,500 PCs for everyone, we build $500 PCs. I’d build $500 PCs by cutting corners where appropriate. In my estimation, there was no reason for a secretary to have a sound card. Or a fancy video card. So I’d skip the sound card, buy the cheapest PCI video card I could find that still had some GUI acceleration, drop in an AMD or Cyrix CPU instead of a Pentium–these computers were for running Word, not Quake–and, since this was business use and we wanted the computers to last as long as possible, I’d splurge on the hard drive, buying the fastest model I could find and stay within budget.
At the time, this made all kinds of sense. Emachines didn’t exist yet, so there was a whole lot of nothin’ in the sub-$500 space. The computers made fantastic productivity boxes but lousy game systems. We wouldn’t have to worry about people loading games illegally on the systems–chances were the games wouldn’t install anyway, and if they would they’d run so poorly no one would bother.
Then we learned the downside the hard way. Support was all us. If a component failed, we had to find a spare to get the system back up and running, ship it off, convince the vendor it was bad, who might replace it, or they might refer us to the manufacturer, who would want to know why we thought it was bad and might want us to run some tests. In a home environment where you have two PCs, this is a minor hassle. In a business environment where you have a few hundred PCs, you’re going to have some failures–even the big boys have them–and it might take half an FTE just to take care of this stuff.
Unfortunately, we didn’t really have the budget to keep much in the way of spare parts. An awful lot of components from the old 486s ended up getting recycled, even though they had no business being recycled. But when you have a dead hard drive and nothing but a 500-meg drive to replace it, what do you do?
I put it in and listened to a lot of complaints.
Overall I don’t regret it. At the time it was the only way to accomplish what we needed to accomplish.
Times changed quickly, however. The $1,000 PC existed in 1998. Late in 1998, eMachines came along and rocked everyone’s world. Those early eMachines were underpowered, but they launched a price war. Today, you can get a business PC for what would cost to build it. If you buy in any kind of quantity, you can usually get it cheaper. You get to deal with one vendor instead of five (assuming the best-case scenario of a system consisting of an integrated motherboard, memory, hard drive, case, and monitor). And when a computer built by someone else breaks, some of the blame goes to you and some of it goes to the company who built it. Having had it both ways, I like to be able to share the blame.
Additionally, the corners I cut in 1998 can’t really be cut anymore. Integrated motherboards with video, sound, networking, and basically everything you need exist today, and they cost 50 bucks. AMD and Intel sell cheap CPUs and VIA sells really slow and cool-running CPUs. In 1998, AMD, Cyrix, and IDT (WinChip) were all willing to sell cheap CPUs, and a fourth company, Rise, was going to release one as well. (If Rise ever did release its cheap CPU, I never saw one. But the threat was there.)
By building your own, you might be able to save 50 bucks. But you’ll spend more than 50 bucks in labor to put the thing together and burn it in.
As far as the suggestion of using DIY servers, forget it. There is no benefit to upgrading a server the way you would a desktop PC. In anything bigger than a small business, people howl when you take the server down to patch Internet Exploiter. Do you think they’ll let you take it down to replace a motherboard? No way.
Would you do it anyway? No. We still have a couple of servers from 1996 running. They’re slow, but they’re still getting the job done.
I agree with the author that for the price of a service contract you can keep a lot of spare parts. But that means you have to have a place to store them. You also really need to verify that they work before you store them, because when a server is down, the replacement part needs to work. I’ve had three hardware failures on servers within the past week. (It sounds like a lot, but when you have 125 of them and can’t really remember when the last one was, that’s not nearly as bad.) It’s worth the money on that service contract for it to have been someone else’s problem.
Besides, only really cheap servers use desktop components. When you can find server-grade motherboards, they aren’t cheap. You might save 50 bucks by building your own server. But again, you’ll spend 50 bucks in labor.
Even when we were building our own PCs, we still bought our servers from IBM. I remember one horrible weekend when one of the servers failed. Between IBM’s mighty resources and ours, we were able to get it back up and going over the course of a weekend. Without IBM behind us, it might have taken us a week. A very long week.
If we’d ever had any thoughts of building our own servers, they evaporated during that grueling 72-hour time frame.
If building PCs is something you enjoy, great. Make it your hobby. Unless your business only has a dozen or two PCs in it, you don’t have time to be doing it at work too.