Slashdot Mirror
Rolling Your Own Business Desktops?
mike asks: "I'm mulling the logic of my company building its own desktop computers. As the IT Manager (plus sysadmin, janitor...) of a struggling-yet-thankfully-still-alive dotcom, money is really tight. We have around sixty ~400MHz desktops which are increasingly showing their age. Acceptable P4 systems from the big guys run at least $1000. By recycling the OS (Win2k), case, cdrom, floppy, and K/V/M, I figure I can assemble a good AMD system for about $600. That's a 40% savings. Is it worth it? The cost difference could very well determine whether this project proceeds or gets put on the back-burner again."
"Some negatives about rolling my own:
- Management: I won't get the special business features offered by some manufacturers. Dell's OpenImage, for example, looks awfully nice. But how much does that really buy me in a company of 60 machines? I don't use such stuff now; am I missing out on nirvana?
- Time to build: Even though we'd leverage Ghost wherever possible, handmade systems nevertheless take time to build, load, & configure.
- Supporting different platforms: Because money is so tight, I can at best afford a capital replacement rate of 25%-33% (15-20 units) per year. That means I'm committing to the support of 3 or 4 different platforms. Having just one platform is great, but how many companies, even ones that actively strive for it, truly enjoy that luxury? I inherited two platforms (Micron & Gateway); support isn't that bad. With proper planning, I don't see why we can't support four.
- Hardware quality: How much can I trust a popular Athlon chipset in a business environment? I feel silly bringing this up because I have a few Athlon systems at home, each with a different chipset, and they've been nothing but rock solid. But I know the lack of a really good chipset has been a large contributor to why AMD's aren't more prevalent in the business world. (well, that and long term bullying by Intel).
- I don't get a proven, prepackaged system that works right out of the box.
- Cost savings. Plain & simple.
- Increased horsepower per dollar spent.
- By choosing my own equipment (mobo especially), I suffer fewer OEM shortcuts.
- I have to admit that I'd enjoy the pure geek satisfaction of rolling out 'my' creation to the company.
For those that are curious, Ask Slashdot did an article on the AMD issue, here.
Maintaining all of them would give you plenty of job security.
Will Microsoft even allow you to recycle your Win2k license on a new computer?
...but Microsoft might be. You might want to take a look at the EULA from M$ and see if they allow the transfer of operating system. Not that I'm suggesting you follow that load of malarky, but it may be a consideration.
Personally, if they're just office type machines. Get Star Office and Linux and see what you can do. Experiment with a couple of your users to see how much trouble it might be.
Don't forget having to run your own assembly and tech support shop as well. I can usually coerce somebody to come out from Dell and replace my broken (video card, motherboard, CD-ROM drive) with little effort here at work if the need arises and it's covered under warranty. At your shop, YOU are the warranty guy.
Also, factor in the labor costs (which will be substantial), count the amount of time it will take for you to assemble a machine, the cost of ESD straps and mats (you will be using ESD mats, right?), the time it will take to set up an assembly area, and the space that will take up, etc.
I used to build machines for other people (family members, etc.) Now I just tell them all to buy a Dell because the hassle on me to maintain them is WAAAAAY less. The only machine I build myself anymore is my personal box, because I spec out stuff that is too high-end for a manufacturer like Dell anyway.
What's your salary/the salary of the people that will have to build 60 boxes? How long will it take? Are you sure $600 + labor costs + no manufacturer support will be less than $1,000? If not, there's no business case to do it yourself.
-matt
Don't buy more processor than you need; It's expensive. You can always upgrade the CPU later if you pick a good platform. You can do the whole thing for about $450-$500 for each box.
Incidentally, I picked the GF2MX because it has good drivers and VERY fast 2D. If you are doing cad or something, get something from matrox, they have a much better DAC. The 3D is just icing.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Perhaps you should consider Walmart's Microtel PCs without Windows. Assuming you don't need software or monitors, you can get a 1GHz Celeron for $400. The trick is the legallity of transfering your Windows licenses (Which piece of the original computer does the license go with, the hard drive? Can you swap that piece into the new system). [Of course, if you could convert to Linux, that would be cool, but that's probably a separate battle.]
If you must, go out and get some low-end consumer PCs and buy a bunch of spares: it's less work than building your own and still very cheap.
While I can appreciate the geek factor here, I think you'd be nuts to roll your own systems here. It will eat up loads of your time, overall costing your company more than it would to just pay more for each system. And I'm not just talking build time. When (not if) one of the systems go kerput, you'll end up diagnosing it yourself, RMAing the defective component, replacing it yourself, testing, reloading OS (if needed), etc etc. Compare to getting a Dell or something, where you determine software or hardware. If hardware, it's under warranty, you don't have to so much as crack the case open. Saves a lot of time and therefore cash.
Even if they cost a little more, I think you'll find yourself grateful for a warranty to fall back on. Plus, when machines go boom, you aren't instantly blamed. If you roll your own, any system that crashes will be pinned on YOU, and you alone.
I know that's not a situation that I'd like to be in. Would you?
End of lesson. You may press the button.
...and I can't really recommend it.
I worked in a 50-user shop, and provided services and equipment to a 200-user shop under contract.
In our case, the only way to get decent specs and meet the client's budget was to roll our own. The other options were too few systems, or systems too cheesy to contemplate. Cheesy as in crap, not as in creamy goodness.
If you go down that path, my suggestion would be to make sure you have confidence in your component choices, and that all your component choices interoperate flawlessly. Any system you have to see again will blow the savings - your first callback or return could be fatal. Make sure you source quality components, and if you're trying to minimize the number of discrete configurations, buy all your components at once.
Spend money on decent cases with good power supplies. Don't yield to the urge to "cheap out" on components that "don't matter" - they all matter. Don't buy cadillac parts, but make sure everything you do buy is good quality, sound, and durable. Keep extra original parts on hand, especially a mobo or two.
Come up with a logo and have the stickers printed - it amazed us how many people would readily accept a brand they'd never heard of, but would never accept an unbranded system.
Your initial problem will be evaluating a number of different hardware options, then settling on those you want to standardize on. Once you get to that point, what do you do with the bastard love children of your prototype period? Don't deploy them to users, you'll water down any faith and confidence your production systems should inspire.
http://drteknikal.blogspot.com/
Time to build: Even though we'd leverage Ghost wherever possible, handmade systems nevertheless take time to build, load, & configure.
Yes. But make damned sure that you're building them as an assembly line. The principle is that building a second one will only take 50% more effort than building the first; the third will take only 33% more effort than the other two, etc. Whatever old Henry Ford's theorem was. It works.
Set aside a room where no one else will bother you. *GOOD STATIC CONTROL* is mandatory. Do all stages of assembly at once, that way you're not wasting time fumbling back and forth for screwdrivers. Get going at a good clip with quality cases, and you should be able to assemble 100 systems/day - but that assumes you have *everything* where you need it when you need it, there's good padded shelving, and you've got a grunt taking care of taking cases out of boxes for you. It also excludes software load.
Just make sure you get a warranty on all the parts since you will not have one on the entire machineAbsolutely. But, assuming a competent builder (ie. not blowing processors with bad jumper settings or blowing boards by not having them seated right), the parts themselves should be pretty reliable. If you're buying good stuff, the biggest source of problems will probably be static handling.
Keep in mind that a modern memory or processor chip has literally millions of CMOS transistors. CMOS transistors have an incredibly thin layer of glass between the gate input and the source-drain circuit. A voltage applied to the gate influences the flow of current through the source-drain circuit. Trick is, the layer of glass involved is so thin that you can punch a hole in it with 30V. Next trick is that static electricity generates kilovolts (thousands of volts) with sufficient current to blow holes in the gate layer, but be imperceptible to you.
All it takes is one transistor out of the millions inside a modern chip to be defective and the computer will crash apparently at random... you know, when Windows VMM writes a 0 to a memory address and gets it back as a 1 later on... BSoD. Kernel Panic. Choose your flavor.
Wrist straps, static baggies, conductive floors, grounded workstations are *crucial*. Dell, Compaq, Asus and Abit spend millions of $$ on these things, and for similar reliability, you should demand the same standards every step of the way for your home-rolled machines. Make sure your computer store hasn't "helped" you by opening the static baggies. Write that one into the contract with the computer store. And make sure that the hard disk drives are still in their packing "egg-crate" things. You really don't want a box with a stack of hard disk drives. (Western Digital had a great video on hard drive handling floating around the 'Net, you should view it if you're building en masse.)
Fire and Meat. Yummy.
This guy is either going heavy 3D, or something like that...or he has quite a strange concept of "obsolete". Add some RAM to those machine be happy with them.
*sight* People don't know how to take care of computers anymore :-(
Ahhh...the great dumpster continuum. Many a free computer will be found there. -- sowth (748135)
My 2Ghz P4 Dells are:
- 256MB DDR Ram
- 80GB 7200RPM Maxtor HD
- Builtin Sound, Ethernet
- CD-RW Drive
- Full Tower case (solid)
I've been building clone boxes my whole life, but I couldn't pass this Dell up. This is pretty typical on Dell's site.I usually check gotapex.com for deals.
Today they have a business class Dell P4 1.6GHz GX240 for $357.64 shipped. You can't build a loaded clone machine from scratch for that much, let alone one covered by a 3 year warranty.
I know this is a little late, but here's a solution we've come up with that's in the cooker for some of our clients in a similar situation.
:-)
Basically you should invest $4,000 in a single server, RAID SCSI drives, dual athlon, 2 gigs of RAM. You've already got a 10/100 Mb backbone for your network, so you can slip this in just about anywhere.
Now here's where it gets fun. Load your favorite distro of linux, visit the Linux Terminal Server Project, and make a terminal server out of it. Then, check out MOSIX, or Sun's grid-computing (the later sports better redundancy, a feature I adore when working with end-users). Grab nics and boot-roms for each PC, install 'em, and boom, you've got a complete functioning cluster of what, 40 PII's? You have any idea the power those can muster?
Not only will you see a huge boost in computing power, but you also save money. Need to use quick books? What's a single liscence for Citrix cost? You can publish the app natively on your terminal server. Open Office works great for converting all those old MS documents.
Honestly, KDE 3.0 just came out. Use it.
Slackware forever. Honestly, what else would you trust when it absolutely positively has to be stable, secure, and easy
But seriously I have worked on many many systems over the years and I have never had a problem which could be credibly linked to hardware failure brought on by ESD. I think that lots of less-experienced techs and help desk people blame any problem that they can't figure out on ESD. Got a BSOD that you can't explain? Here's a convenient line that no one can really disprove and makes you look smart. "It must have been mishandled by some other ignorant tech years ago and is just now showing symptoms." Right. How could even tell the difference between an ESD problem and a problem caused by irregular AC line voltages or manufacturing defects?
What the hell is a "latent failure"? As was stated above, ESD is measured in kV while CMOS tolerances are more like 30V. Either a transistor is blown or it isn't. I agree that ESD can damage transistors, and I also know that a computer may very well power up after suffering damage from this. However I think the notion that a system would power up and work normally for two years before going south is ludicrous. You seem to think that the static can somehow "weaken" the hardware without fully blowing it out.
My personal solution to the ESD problem is a compromise between the incredibly annoying wrist strap and "going commando" and risking relatively expensive hardware. I leave the power supply plugged into a grounded outlet while working on the machine. I know somebody's gonna flame me for this, but think about it. When the machine is plugged in the entire chassis is a path to ground which can bleed off excess voltage in the case of a static discharge. If you simply touch the chassis before you start working you will discharge any static electricity which is being carried by your body and you're good to go. Unless you are working on your computer while standing on a shag carpet in your socks while rubbing a balloon on your head then this is probably all you need to be safe. You could then unplug the AC line if you wanted, although I don't see any harm in leaving it plugged in during your entire operation. Outside of the power supply the voltages can be no more than 12V and low current so electric shock is really not an issue.
On a side note I think a much more common issue is the failure of the power supply itself, rather than motherboards and chips. In most machines I build the PS will burn out after a couple of years unless I spend a few extra dollars on a step-up model case like an Enlight.
Alright, flame away.
alex
--- Wherever you go, everyone is always connected...
I'm not sure what the chances are you'll even see this. 450 posts and the ones modded up are pretty negative. Oh well. . .
I'm currently rolling custom built machines for our 200 systems network. Oh, and I'm the only tech here. I do the servers, network, help desk planning, everything.
My place is a non-profit where a very small, chaotic budget. I'm never sure exactly when I'll have money to spend or how much. For strange reasons, when we go to spend money we have to go through a maze to buy complete equipment, but components are no problem. We couldn't buy a new company car, but we could buy all the parts to put a car together ourselves. Same goes for computers.
The savings we've seen building ourselves are huge. Adding the costs of the pieces and my time spent planning, building, and supporting these systems it is still cheaper than OEM systems and a support contract. A+ certified techs are a dime a dozen, so support of these wintel systems isn't really a factor if I were to leave. (They'd have to get a half dozen to do what I'm doing by myself in 40 hours a week, plus an MCSE and a CCNA, so I'm not worried about job security).
Here are some tips:
1) Plan out your configuration and use it for the next year. The most important component is the motherboard. It should be able to accept more RAM and a faster CPU than you are going to use initially. Spend lots of time developing a stable, user-friendly software config (OS and apps). As you need to replace systems throughout the year, use this config. After six months update the config with a faster CPU, more RAM, and maybe a larger HD. Update your software config with patches, fixes, stuff like that now also. At the year mark you can plan your new config.
2) Integrated components are your friend. I like the nVidia nForce boards because they have the (good) sound, video, and network integrated. Also, if one manufacturer stops making your board, you should be able to switch to another manufacturer but still use the same drivers. Very important for ghosting!
3) You really don't need the management software for 60 computers. That stuff is usually designed (and priced) for enterprises with several hundred if not thousands of systems. You should be able to keep most of that stuff in your head and in a small text-file database. Learn a little Python/tk and you can even build your own front end to the text-file. Cool!
4) Develop a relationship with a couple local component vendors, and a couple Internet vendors and have them bid for any purchase more than a couple grand. You'll definetely save money this way, especially if they know they are bidding and not just giving a price quote. I've saved thousands of dollars on a single purchase this way. Also, after a while the local guys will probably be able to send a couple guys your way to help out every once-in-a-while when you get swamped or stuck as a thank you for your business. Very Cool!
Following these tips, you only have four platforms to work with, you've saved money, you know exactly what you are working with, and you get a sense of pride from creating something from your own two hands.
I really can't recommend this approach highly enough.
obviously no deficiencies vs. no obvious deficiencies
That's it. I've only fried a single component ever, an old hard drive, and it was because I drove around with the components rubbing against the cloth seats in my car in the sun in mid-winter, sliding around. I reached out and picked it up and felt the shock. uh-oh. Sure enough, it was toast. But it's still pretty rare.
Any connection between your reality and mine is purely coincidental.
Remember the article the other day when we all laughed at the Wilkes Barre IT guy who stopped the IBM maintenance on the AS/400 ? Well, this is the same kind of thinking: penny-wise and pound foolish.
Remember: support contracts are a form of insurance. They insulate you from the risk associated with the issue at hand. When looking at any form of insurance, you have to take into account what the worst-case senario is, and if you can handle it. In your case, the scenario is that you have multiple desktop failures, including critical failures of important machines (ie, severl of your main developers). Do a cost analysis: if I do a roll-my-own machine, what's the cost of it breaking? How much does it cost for that developer to have no (or a seriously inferior) machine for a week or more, vs. the 1 or 2 days a supported machine would be out?
For small companies, (especially those heavy in software development) I can't imagine a situation where the TCO of a fully-supported system is worse than a roll-your-own box. None The downtime and IT personnel time alone will kill that equation. For huge companies, it may pan out, but for a 60-desktop company with 1 IT person? Not a chance.
You need to put this into perspective with Management. Once again, they are looking at only the up-front costs, and none of the hidden costs, which in this case are the majority. Explain to them what the true cost of a desktop is, and how NOT buying a supported machine results in a WORSE return over the next year.
Now, here's a couple of recommendations for getting SUPPORTED desktops into your organization while not breaking the budget and still meeting increased performance needs:
I don't mean to harp on you personnally, but this kind of thing is why IT has a long, long way to go before being really professional. Folks, this isn't a garage. IT folks need to quite thinking like it's an expanded hobby, and also need to remind the Executives of this, too. It's a Profession, not a Trade.
-Erik
Systems/Network Architect and former SysAdmin
There are always four sides to every story: your side, their side, the truth, and what really happened.