Would you buy a computer that you had no intention of using?
A friend of mine does IT work for a company and asked me for my opinion regarding spare parts pools. I told her it was a good idea to have a few spares available of the most common components, especially when you have a homogeneous environment.
But she told me she already knew that. What she wanted was my opinion on spare servers.
Right now my friend is working setting up a blade server architecture, buying a half-populated blade chasis that would handle approximately six servers. The salesperson recommended adding one blade to the chasis as an emergency spare.
By using a spare blade, it can be immediately put into service when another server fails. That reduces the likelyhood of a critical service going down. I understood the argument made by the salesperson, but I didn’t feel 100% comfortable with it.
I keep four spare disk drives (two hot, two cold) for my NAS appliance. They don’t cost much, they’re reliable, and the data maintained on the NAS is very important to my company’s infrastructure. I also have one or two spare components, like NIC cards or DIMMs. To this day, only one of those parts has been put in use (knock on wood), but it provides a sense of relief to know they’re available.
But a spare server sounds like an unnecessary cost to me. I told my friend that she would be buying a server that she has no intention of ever using until something goes wrong. If it’s kept in the blade chasis, it will consume power, and populate a blade slot that could have been occupied by an actual production server.
When she upgrades the servers, she would have to upgrade the spare unit as well. Unlike a spare hard drive, the cost of a server blade is a little more noticeable on a capital expenditure form. There’s also the risk that someone may choose to put the spare into service, giving the usual non-apologetic sysadmin excuse, “I needed a server, and that one was doing nothing.”
I told her that if the blades are running a distributed application, (mail farm, web servers, etc), and the loss of one host will not affect the performance of that application; then a spare unit is unneeded.
A spare server should only be utilized if it can protect an application or service that cannot run in an active/active cluster environment. In that case, the spare unit should be provisioned expressly for the service intended, and not just for the benefit of a spare parts pool.
In either situation, a spare server is no replacement for a good service contract. If you decide to use a spare server, make sure you can fix the original server within a reasonable amount of time.
I run many, many active-passive server pairs. That is, effectively, a spare server. I also run setups much like this with four running servers and a fifth just in case one of the others dies, we need to test something quickly, or whatever.
They're handy for all sorts of things; upgrades as well as failures.
Blades are even cheaper than full servers, so having an extra few shouldn't be a big problem. Why troubleshoot all night and run at degraded performance in your cluster while your wait for hardware when you can just fail over and work on it the following day?
Hardware is cheap compared to software and people anyway. (especially hourly people on call pay, and you usually don't have to pay software licenses for truly passive nodes where the software isn't running)
With blades, usually you're using some sort of SAN or shared storage, and they tend to die quite often. In that case keeping them around is a good thing(tm), especially with the management features that's availiable on them.