I can live with 4 hour downtime, but
it will not be comfortable. Am I
better off buying or building? I see
pro's and con's both ways.
The costs you save on parts will most likely be cheaper due to quality and/or poor support. From RAID cards to memory and power supplies, saving a few bucks here and there is usually a losing battle in the end. If that end happens to be 7 months away to 7 years, people's opinion of "the end" tends to be more of a matter of luck when rolling your own servers.
I only really need one server (I'll
use VMware esxi), but if I build I
will build two identical machines to
make up for lack of a support
Well, you haven't addressed shared storage so having two ESXi servers (w/o shared stored) is no different than having one server. How is this a benefit? Personally, I'd rather have one beefer server with quality hardware vs. two cheap machines and no shared storage.
Slightly off-topic: if you're going the VMWare route, in my experience VMWare is much more 'picky' when it comes to nVidia chipsets for AMD Opterons. Granted you could be going with Intel but if cost is such a factor, I would highly recommend AMD from a cost perspective. I don't know if you're dead-set on ESX/ESXi, but if not I recommend Citrix XenServer 5.5. With more features for their free version right out of the box and overall less cost in terms of pricing for enterprise-y features, give them a look. Especially for education where every damn dollar matters. I would avoid going the VMWare route for cost and cost alone. Just my two cents.
What do I need to be considering most,
and what am I potentially missing?
I've answered a similar question to what you're posting (here). I'll cover my 3 main points I brought up in terms of what you're missing from server level hardware. Womble covered a lot of points that I'm reiterating so feel free to skip to the bottom.
Durability - Granted not everyone
has an identical experience but
server grade equipment seems to last
a bit longer than their desktop
counterparts. But even for prolonged
(and stressful) usage, server
hardware tends to hold up to the
demands under the specifications
provided by the manufacturer.
Usually better driver support for the appropriate OS. Desktop
equipment may or may not have solid
drivers but server hardware wouldn't
sell without some serious attention
to detail. Driver support is
critical as well as serious testing
of equipment. I find that once I pay
for the server hardware, I worry
less about this issue. If an issue
does arise, manufacturers usually
Scalable - The majority of most core
server hardware (motherboards, CPU,
RAM) anticipate upgrades to one
degree or another. Most desktops do
not anticipate a larger quantity of
resources. This is usually a
function of the motherboard chipset
more than anything, but server
chipsets are significantly different
What you're missing
You're missing out on better hardware/software (drivers) from vendors with tons of experience and lessons learned baked into their products. Not to mention the typically seamless upgrade paths (scaling-up) to management features, server vendors have tremendous added value that you just simply can't build by yourself.
You're missing out on components that when combined together tilts the probability of failure away from your server. Whether you realize it or not, you're adding more workload to your responsibilities that really aren't furthering the school district needs and under-utilizing your time to the fullest. And that means that the school district is missing out on your attention for more important tasks.
What you need to consider
I have built servers over the years for myself and a handful of clients. These days, I find myself recommending/emphasizing that my clients go to a vendor (HP, Dell, Sun, etc.) for the quality of parts that I cannot obtain or for the support contracts. Support contracts seem wasteful/meaningless but to me they're kind of like insurance: you only appreciate it when you need it.