Should you care about whether you can upgrade hardware?

November 18, 2006

An online discussion I was in today touched tangentially on the question of how important it was to be able to upgrade the hardware on non-server machines (and how upgradeable various sorts of them are). My view is more and more that it probably isn't all that important.

(To be clear, I'm talking about post-purchase expansion down the road, not being to add necessary stuff at purchase time. If the machine doesn't do what you need at the start, you just don't buy it.)

Historically, we've almost never expanded or upgraded machines. We buy them in the configuration we want, usually with some extra margin in things like RAM and CPU power, run them for BIGNUM years, and then replace them wholesale. By the time a hardware upgrade seems necessary and can be sold to the powers that be, you usually can't get the necessary hardware bits any more.

(And even if you can, the assumptions that you built the machine on may seem quaint and outdated, like 'SCSI drives are the way to go' or 'it's too early to trust SATA, we'll stay with IDE'.)

So, how often are real machines actually expanded or upgraded? My own suspicion is that most machines that actually get upgraded were underconfigured when they were bought, for whatever reason. (Price is one obvious one.)

(Thinking about it more, part of my lack of interest in upgrading machines is because I like to keep old machines in an operable state, either as backups or to be used for various undemanding things.)

Comments on this page:

From at 2006-11-19 10:28:53:

The company I work for typically upgrades servers every 3-years. The word on the street is that they upgrade at three years because failure rates start to increase, and the maintenance costs for the machines shoots through the roof (in some cases, a server upgrade is actually cheaped than continuing to pay for hardware maintenance on the old server). In these cases it's not that the hardware was sized wrong, but that the server has outlived it's "useful" life (I don't agree with this, but that is how the bean counters think).

- Ryan

From at 2006-11-19 12:28:12:

I don't think that saying "underconfigured because of price" is a useful way to look at things. And the reason is, we would always buy more if we had more money, so it does not give us any useful insight.

The real decision we are making here deals with the tradeoff between the current capability and potential expansion in the future. Since some amount of money is wasted on supporting the expansion, the alternative is to buy a cheaper non-expandable system with the same parameters, or more capable system, if we agree that we are constrained by the price. So, the sunk money plays against the expense of whole-system replacement later (it may be easier to play off those systems which have to be replaced today). If one knows his labor expenses and knows what it takes to bring the new system online and reprocess the old one, he can calculate the tradeoff and make a rational decision.

The article (unlike the first comment) dealt with non-server systems, where outside of the box components play a role, specifically desktop monitors. I encounter people having monitors and system boxes on different replacement schedules. Printers were decoupled so long ago that nobody even remembers times when they were bought together with the main system.

-- Pete

By cks at 2006-11-19 22:33:55:

I don't think that saying "underconfigured because of price" is a useful way to look at things. And the reason is, we would always buy more if we had more money, so it does not give us any useful insight.

I've generally had a somewhat different experience; I haven't felt that I was underbuying a machine for price reasons (beyond display size, which I consider a separate issue). Spending more would have been getting too much computer for the job (especially as most machines are used for undemanding stuff). It takes years for software growth and requirements to creep up to the point where the old hardware becomes underpowered for the new stuff we want.

(Partly this may be because we have historically bought towards the high end of current hardware, because we know that we're not going to get to replace it for years, no matter what.)

I hadn't thought about the monitor issue before, mostly because (again) we tend to buy high when we get the chance at all and then run stuff into the ground and only replace it when it explodes.

Written on 18 November 2006.
« The good old days of Unix
Link: Golden Rules for Bad User Interfaces »

Page tools: View Source, View Normal, Add Comment.
Login: Password:
Atom Syndication: Recent Comments.

Last modified: Sat Nov 18 22:51:15 2006
This dinky wiki is brought to you by the Insane Hackers Guild, Python sub-branch.