Generic PCs For Corporate Use?

porkThreeWays writes "I work for a government agency supporting about 1000 PCs. The economy has hit us just like everyone else and we are looking at ways to save money. We currently buy Dell computers and even with our government discounts end up spending about $1,000 for a pretty mediocre computer. I had the idea of building our own PCs for considerably less. We'd spec out a standard configuration that we'd use for 18 months. CPU speeds and RAM sizes may change during that time, but socket types, memory standards, hard drive interfaces standards, etc, etc would be required to stay the same. We have Dell warranties right now, but I could see just keeping spare parts on the shelf and building that into the cost of the PC. We'd also be able to transfer Windows licenses because the Dell installs are non-transferable. However, I couldn't find anyone on the large scale doing this. Is anyone on Slashdot using PCs they built themselves on the large scale?"

Comments

$1000 a PC?

[FlyHelicopters]

What is on them, a Core i7 with 12GB of RAM and an SSD?

Me thinks you're overpaying... Dell isn't that expensive, really it isn't...

Re:$1000 a PC?

[MoonBuggy]

Dell has better volume discounts than you ever will, both with Microsoft and the hardware manufacturers. They further offset this by bundling in a whole load of crapware on the default OS install.

Even after accounting for their profit margin and your time spent re-imaging the machines with a clean version of Windows, the cost from Dell compared to DIY for standard beige-box business machines should be somewhere between slightly cheaper and slightly more expensive; if it's the latter, a single point of contact for warranty issues is still perhaps worth the money. If it's the former, you win on all counts.

Re:$1000 a PC?

[CAIMLAS]

They are loathe to cut out a middle man unless it means a substantial guaranteed return on investment.

That's part of the picture. Consider:

* I buy 100 Dell systems, then leave. My stupid manager barely knows how to plug in a keyboard, but she can rely upon Dell for support.
* I build 100 PCs and save 20-30%. My time is a sunk cost, and I'd have spent the same/similar time rolling out the Dells, if a bit more. I leave. My clueless manager has to find someone with the skillset to directly support these PCs; she is now reliant upon others down the food chain, instead of someone external. Why would she put herself in this bind?

Since most managers seem to see their subordinates as cogs, this is never a sane option for them.

Re:$1000 a PC?

[Smallpond]

By going with Newegg and building it myself over a weekend, the price was cut in a little more than half.

Whats your chargeout rate for weekend work?

Also factor in returning and replacing parts that don't work, test and burn-in, loading software, and managing licenses.

Re:$1000 a PC?

[DJRumpy]

Not that I agree with 'never', given that the industry standard is to go with manufacturer deals, I would think that out of the millions of businesses out there, it's been tried and found not to be cost effective under normal use cases to 'build and support your own'. Given that current IT shops can either simply box up a problem machine and ship it off and swap it out with another, it removes any need for advanced troubleshooting skills, the need to stock replacement parts, and the associated storage costs for such replacement parts. It simply doesn't sound like it would be profitable, and it could easily cost more in all of those areas. Seems the cheaper cost of 'home built' could easily be outweighed by the supporting costs.

Sweeten the pot with the deals that vendors cut with larger companies, including discounts on the OS software and the OS itself, and it just doesn't seem cost effective to take on those additional costs to spare a few hundred bucks and do those things yourself.

For instance, in our company, up front cost is not always the deciding factor. If that were the case, we'd all be using FOSS. TCO should always be factored in, and not only in parts and supplies but support/infrastructure costs.

Don't do it

[tftp]

I had the idea of building our own PCs for considerably less.

This is an awful idea. I had some experience with such an experiment; it didn't work. The computers were failing left and right, and the vendor distanced itself from the situaton. You will first be forced to maintain all that herd, and eventually you will become a scapegoat.

Business is all about using money to make other money. It is a legitimate expense to buy a computer; it's tax-deductible on corporate level, so you don't need to squint too hard at the prices. Buy good computers with a warranty and on-site support and be happy.

Why not start a company?

[SpaghettiPattern]

It sounds like you could compete with Dell and that you should start a company. Maybe then you realise that 1kUS$ isn't that much for a system.

Don't make the mistake of not calculating the effort it costs you to assemble the systems yourself. Say you cost a modest 100US$ per hour to your employer and redo the maths.

You seem to know about hardware. Now consider how you will train co-workers to attain your level of expertise. Will you now be teaching as well? Think of what will happen when you'll leave the company. Don't worry, you eventually will move on to other challenges.

I myself build the systems for my own small business. It's costs me significant amounts of effort which I could put towards paying customers. I only do it because I like it and because I take the liberty to do so. But really, I probably shouldn't.

Re:Why not start a company?

[kiwimate]

You've made all the points I would make except for one. There's the economic concept of comparative advantage. Basically...how good are you at other tasks? What are you not doing while you're building these PCs? Not building a new file server cluster? Not updating the firmware on your SAN?

Now, who does get to do those tasks? Are they as good at it as you?

The point is - yes, you can build computers. But is it the best use of your time? Forget if it's fun (like for the parent) - what would your employer say is the best use of your time? When figuring this out, don't forget that your hourly rate is only a part of what your employer pays. If you're making $40 an hour, the true cost to your employer is closer to $70 or $80 for an hour of your time.

Don't do it yourself

[uglyduckling]

You're crazy if you think it's worthwhile building the PCs yourself. You can easily find an off-the-shelf PC for considerably less than $1000, probably less than $500, and unless you have a team of at least 6 people sitting around with nothing better to do then you won't save money building them yourself, and you'll just cause yourself a massive headache. Simply commissioning 100 pre-built PCs (presuming you're planning to replace 10% of them at a time) is plenty of work for a support dept., even if you're not making massive software changes.

You've never run a big network I'd guess

[Sycraft-fu]

Little more complicated than you make it out to be. To be truly redundant that a switch going down won't take out a bunch of systems you have to have the systems themselves plugged in to two switches, and then every switch down the line. That can work but not only takes a lot more switches, but more complex clients. Most thin clients aren't going to do that. You need not only 2 NICs but the understanding of how to handle failover. Also if the failover is to be fast and reliable you need expensive switches. Maybe not a problem, maybe you use those anyhow, but something that has to be considered. At every level the switches need to be high end such as Cisco to be able to quickly, reliably, handle rebuilding the span. No Linksys stuff that may freak out and create a switching loop (which they do with RSTP sometimes, as I've seen).

Bandwidth needs will also go up substantially. If you go a little heavy on the oversubscription in a normal office setup, no big deal all it means is file transfers to the servers are slow. Do it in a thin client environment, and you are talking interface lag which is really bothersome. So you'll need to have plenty of bandwidth to the switches, probably 10gb instead of gig, and maybe more to the distribution switches.

Then you also have to do redundant power for the switches. If both switches on a floor are on the same breaker it doesn't help much, you need separate circuits, all the way out to the grid/generator if you want real independence.

Of course there's the servers also. If one server runs 50 machines, well then its failure is a major outage. So you'll need backup servers. How many depends on how much depth you think you need, but you need to have servers ready to take over if one goes down. Probably fairly beefy servers too. While you can stack low-impact servers (like DNS or license servers) pretty heavy on a VM, you have to be more careful with interactive systems. Get too many, they'll get sluggish. You'll want lots of CPU, lots of RAM, and still won't want to load clients on them too heavy. You'd have to test your specific setup to find what works but I'd bet no more than 2 clients per server core and probably less.

That also means everything has to be on a separate, high speed, disk system. You can't use local storage or they can't be migrated to new servers. So something like a NetApp. Disks need to be high performance too, since they are going to have a lot of random access put on them. IO is also the biggest problem for multiple VMs. No large cheap SATA arrays, you'll need 15k SAS most likely and SSDs would be a good idea, except real expensive.

Well that needs to be backed up too. If everything is riding on one NetApp, reliable though it is, that's a massive failure point. So you need two of them, running in sync, so that if there's a failure there's no problem.

Ok this is all doable, no question about it. I could design an implement such a system... However I'd have real questions as to if it would save any money. You weigh all that high end gear with service contracts against the cost of a bunch of reasonable desktops. Is it really worth it? My guess is not.

Also remember you aren't saving any money on other server costs. You still need all your other server infrastructure. Maybe you could get rid of your central storage and just use the storage the VMs are on, but I would have to see that in action to be convinced the performance would be ok.

The thin client idea isn't a money saver I don't think, unless low performance/reliability is ok. Maybe a school lab situation or the like. I think it is more the sort of thing you'd do when you need portability (like no matter where someone physically is in the building, they can get to "their" computer) or for security (for whatever reason you want all systems physically in a secure room).