15

This is the canonical question for "Should I build coomputing hardware myself?" questions.

I have put together countless PCs, but never a large server. The geek in me says build it, but the realist in me says let the manufacturer handle it when there is a problem. Ignoring the time penalty involved with the initial assembly time of a built one, which is a better solution? Have you ever run into a problem with a home build server that would have been solved easier/quicker/cheaper by going with a manufacturer? Are there any features that manufacturers give that aren't easily attainable with a home built server?

voretaq7
  • 79,345
  • 17
  • 128
  • 213
Bob
  • 2,917
  • 5
  • 28
  • 32

28 Answers28

26

Buy them. And buy them from alternative sources if you need to be frugal - Craigslist, Ebay, Dell Outlet, etc.

If you end up building them - go with SuperMicro - great gear.

But Commercial Servers will have better out of bandwidth management, better systems management, better support, etc.

And if you need to pinch pennies - use third party memory (i.e. Crucial) - its cheaper and just as good.

Rob Bergin
  • 842
  • 10
  • 14
  • that's why I asked about home or work - at home if I had a rack (and I do) - I would try out a homebrew server (SuperMicro) but with the recent economic condition and alot of people doing upgrade (which is counterintuitive to the economic condition) - craigslist and ebay have had a lot of decent servers on them and its alot less work to just drive to a parking lot, hand over $100 and come home with a nice 1U or 2U server. – Rob Bergin May 29 '09 at 19:57
  • 1
    http://www.abmx.com for the win! (see http://www.abmx.com/low-cost-1u-server) They have *cheep cheep* low-end servers built with Supermicro innards, for less than a user's workstation. We've bought several for non-critical functions and they just run and run and run... – Avery Payne May 29 '09 at 20:00
  • Exactly what are the management features you've seen in a Dell/HP/IBM server that you cannot get from a DIY server with, say, Supermicro/Tyan mobos, Intel NICs and say Adaptec/LSI controllers? – alphadogg Oct 14 '11 at 14:13
  • 1
    ;I ntegrated one step driver update for everything including upgrading the firmware on the drives. – TomTom Dec 08 '11 at 14:21
16

Buy.

  • Warranty.
  • Engineering Design - heat, airflow
  • Parts all guranteed to work together

We bought a server from a local white box store. worst server we ever had. Built exactly to our specs, but the RAID card didn't like the Motherboard. Ended up costing us more than the price of the server in strange errors, testing, rebuilding, ordering other cards, and more testing.

Brian
  • 621
  • 6
  • 12
  • 1
    Heat is a very big thing. You want to pay extra for the engineering that solves this. – sysadmin1138 May 29 '09 at 20:07
  • If you buy a Supermicro or Tyan chassis, you have the same type of "design/engineering" that Dell/IBM or HP have in terms of heat management. – alphadogg Oct 14 '11 at 14:15
10

Buy buy buy. A server is not a PC and the enthusiast/hobbyist in you is leading you down the path of wrongfulness. When you have real live users accessing them and relying on them for their daily work, solidity and reliability are crucial factors.

Maximus Minimus
  • 8,937
  • 1
  • 22
  • 36
6

Go with Dell. Warranty is no issue when you need it. Get it for the length of time you want and the service level you need. Just had a drive fail in a storage server and we had it in 2.5 hours on a Saturday. With a custom bult unit there is often an issue over what has failed and who is responsible. You really want to avoid the finger pointing.

Also have had great servcie and suoppot on IBM and HP but Dell usually beats on initial price.

Dave M
  • 4,494
  • 21
  • 30
  • 30
4

Buy.

The best locations I've seen so far have have been auctions from companies that have folded. I was recently asked to find a new server to match the cost and specs of a previous server purchase.

Turns out that the previous server was bought for a couple of thousand when originally it cost over 50 thousand. GBP here, but I think the difference is clear. Even the normal second-hand routes couldn't come close to the price of a sell-off after a company folds.

Alex
  • 1,103
  • 6
  • 12
4

If you've had the pleasure of following the StackOverflow server build stories (podcast and blog entries), the lesson sounds like:

you should really only consider BUY. There's a caveat to that though: you must value your time.

Just a simple problem with a ServeRAID controller not liking Western Digital SAS drives, but preferring the Hitachi brand. Never mind issues like: mobo/RAID card compatibility, mobo/NIC, power supplies, etc. I couldn't imagine trying to assemble a server together yourself with various parts, and having no warranty on the entire box as a single unit, onsite vendor service, etc.

Having the internals assembled and tested for compatibility are a critical advantage for the big brands.

p.campbell
  • 4,397
  • 6
  • 40
  • 51
4

Definitely, BUY them. The cost saving might look a bit tempting if you try to build it by yourself, but when you encounter a disaster scenario (and believe me, you ARE going to encounter that), you're going to save yourself LOTS of time.

Pros:

  • Better design (airflow, cabling)
  • Easier to look for a driver, software updates, bios, firmware updates (all from the vendor place)
  • Better remote management support (you can reboot and see the booting screen remotely)
  • WARRANTY

Cons:

  • Price
Bimo Arioseno
  • 341
  • 2
  • 6
4

It is perfectly possible to custom build a server to rival anything on the market but you won't do so by buying cheap components at bargain prices. As the price difference for equivalent machines from a manufacturer or built from parts will be quite small, with the latter normally being more expensive, there needs to be a really good reason to go down that path.

When you buy a brand name server, and I'm not talking about some weird brand nobody has ever heard of before, you're (usually) getting a system that will have been assembled with solid components that are have been tested to work together. It will also be backed up be a decent warranty.

If you build your own you may well run into all sorts of issues caused by some of the components not playing nicely with some of the others. You will have no warranty on the complete unit, only warranty on some of the components. You can certainly forget about calling the manufacturer and having them on-site, with a whole set of components, within a few hours to repair the server with minimal downtime.

John Gardeniers
  • 27,262
  • 12
  • 53
  • 108
  • +1 Last paragraph really speaks to this. Great server can be built but a great wararnty is also rather good. Especially at 2:00AM on a weekend. – Dave M Feb 23 '11 at 21:49
  • 1
    Or 3PM Sunday afternoon when the server in question runs the accounting software and the auditors will be there 8AM the following morning. :( – John Gardeniers Feb 23 '11 at 22:06
3

Do you care about the system or will it be mission critical? Buy.

Is it easily replaceable and/or really simple (web server, compute node, DNS server)? Build from a SuperMicro barebone chassis.

Is your time worthless and so is your data? Build from all different parts.

I work in an academic setting so I often end up building (or spec'ing a beige box) to save money. It works but it's not great. SuperMicro barebones take the pain out of it mostly because the motherboard is actually good and for most servers you don't need other peripherals. They like 3ware RAID cards as well. All my compute nodes are dual Xeon 5300 2U chassis from SuperMicro... But if one failed tomorrow, I wouldn't be in a bind.

My core file servers are Dell and HP machines with 24x7 support contracts. That way I can get parts quickly.

My last experiences with a full custom spec'd beige box was a disaster... Vendor sold us a support contract and his only spare motherboard on hand was out of province (1 day delay) and DOA (5 days delay to get replacement)... The big problem with building is getting spare parts quickly. A sys admin I know had to wait 5 weeks for a replacement SATA backplane from Chembro a few years ago, 6TB of storage was offline until they got that part. Not Good.

3

Don't build a server yourself if you intend to rely on it for anything important -- If you're comparing parts apples-to-apples as it were the price differences are typically not huge, and the ability to call one vendor and get warranty service/support is almost worth the price markup the first time something breaks and you have to run around trying to get it fixed.

If you are put off by the price of Dell's hardware you may want to consider SuperMicro or other "whitebox" vendors - roughly equivalent quality, good warranty coverage and typically slightly lower price points.

voretaq7
  • 79,345
  • 17
  • 128
  • 213
1

As many people before me have answered, I'm going to go with buy. How much is your time worth? Sure you may save $200-300 up front on a server by building it yourself. How much do you think your time is worth to the company on an hourly basis? Even if it's really low, say $30 an hour, $300 comes out to 10 hours. I can pretty much guarantee you that a self-built server is going to suck out way more than 10 hours of your life during the time you maintain it.

Working at a startup where a previous admin skimped on gear, I've had no end of headaches dealing with generic 4U rackmount servers built out of home PC hardware. Luckily we only have a few left now since we came to the conclusion long ago that it's cheaper to scrap them and buy something engineered and supported than waste our time playing hobbyist PC shop.

Kamil Kisiel
  • 11,946
  • 7
  • 46
  • 68
1

smaller than 3U - buy. larger than 2U - build

disserman
  • 1,850
  • 2
  • 17
  • 35
1

Building can be worth the risk at the very low end, 2 - 4 servers colo'd with "this guy I know", and at the really high end where you are running all custom code in all custom datacenters (google, top-tier hpc) or just have cheap nocmonkey labor laying around. Nearly everywhere in between its professional malpractice to build. There's more reasons than I could probably fit in the character limit of this post, but they really all boil down to this: are you in the business of competing on the best price/performance/compatibility tradeoffs of a ~dozen part x86 server system, or are you in some other business? Spend your time on your actual business.

cagenut
  • 4,808
  • 2
  • 23
  • 27
1

Building a good compact 1U or 2U server is not as easy as drawing up a decent tower model. There aren't a dozen good sites testing all the components, cooling more a challange, parts more expensive and harder to come by. I have always bought for work, and built for home.

But if I would install a rack at home (I won't. Too much noise), I would build those servers as well since I am a maker at heart and just love DIY when it comes to computing gear.

Peter Stuer
  • 1,473
  • 9
  • 11
0

the issues with build v buy really come down to additional features and case size.

If you want a 1U server, buy it - really, putting the bits in is not worth the trouble, especially as you can't use standard PC bits, they all have to be low-profile, and then you also have to get the airflow right.

2U its a little better, you have more options but its still probably worthwhile to buy.

4U, you could build one - its basically just a PC case on its side. However, we now get into the additional features part: if you build a PC in a rack case, fine, but every 4U server I've worked with (eg Proliants and the like) come with hot-swap capabilities, pull-out drives, even the bits that aren't hot-swap are very easily accessed (and all screwless, you can replace a network card in seconds). In addition, there will be diagnostic indicators all flashing at you when something goes wrong (eg in the network card case, you can tell what made the server fall over because a little light was on in the slot where the network card sat).

Then there's also features like lights-out management, and diagnostic software (but you can get this yourself with an appropriate card, and stick Linux on for better diagnostics than many manufacturer's software)

Issues: with a bought server you're going to have to use the manufacturer for parts, you can't just buy any old network card to replace the old one - not if you want the same features. This means you either have a spare lying around, or support from the manufacturer - some do 4 hour onsite replacement - or you have the time to spend fiddling with the diagnostic utilities to prove to the support guys that the part really has failed (that can be a real nuisance - first thing they ask you to do is run the diagnostics, and they can take hours, only to get a replacement HDD that you know is bust because it rattles when you shake it). If you've built your own, you can buy a spare and slot it in without fuss. This is most relevant if you buy a 2nd hand server - in which case, my advice is to buy an extra one to be used for spares.

gbjbaanb
  • 3,852
  • 1
  • 22
  • 27
0

I build or very small websites or development ( SuperMicro ) and for clients I buy Dell 1u and 2u servers.

0

I suppose it depends on the volume and reliability you're looking for. Mom & Pop shop with 5u worth of machines? Might as well make them to save money.

If you want reliability, then you want homogeny, and you need to buy your servers. I'd recommend against buying from non-primary vendors, unless it's some place like CDW. I buy my network equipment refurbished, not my servers.

Matt Simmons
  • 20,218
  • 10
  • 67
  • 114
0

Definitely buy. And if you're going to get many servers, buy them from the same manufacturer. You won't have to deal with different diagnostic procedures and different technical support if you need it one day.

In the past (long time ago) I was used to build my own servers (mainly because it seemed cheaper) but I can tell you that you won't be able to buy separate parts and reach the same quality of design than servers from professionals.

This said, you now have a large choice of brands : Dell is one of my favorites because their web site is very useful to compare configurations.

Emmanuel BERNAT
  • 312
  • 1
  • 6
0

Have you considered Amazon's EC2 for this? Might be easier/cheaper than having a server in house. Also, it allows for easy replication and creation of new hardware with scalers in front.

0

Buy but try to stay away from refurb. I have built a few servers in my day and even though everything should work well together there will be something just doesn't. Also as others have mentioned you get the engineering, warrenty, and everything working together. Also if it is important to you the out of band management is well worth the cost.

Our development servers are all Dell or IBM refurbs that are coming upon the end of their life (7+ years old) but purchased three years ago. I can no longer find replacement parts for many of the parts.

steve.lippert
  • 698
  • 6
  • 13
  • why can't you just buy more old servers of the same type - ebay for parts :) – gbjbaanb Jan 24 '10 at 15:00
  • I can't waste money in our budget buying more old parts that are doomed to fail sooner rather than later. Lately I have been forced to build due to budget constraints. Hoping the bite in the ass that I know is coming comes MUCH later. – steve.lippert Jan 29 '10 at 18:57
0

I'd say build can be worthwhile in the following circumstances:

  • minimum 2U case, I'm more comfortable with 3U or bigger
  • server that you want to be able to quickly rebuild, but can live with it being down for a while
  • you either have some spares or are willing to buy them. You need spare power supply, motherboard, disk controller, disks, ram.

The one time I did this for an important production server, it was just a storage server - a very simple windows configuration, just a server stuffed full of disks on a good raid card. It was cheaper than a comparable HP server, but not a huge amount. At the time we like the flexibility of knowing that we could get it back up and running quickly if we had any problem.

Ward - Reinstate Monica
  • 12,788
  • 28
  • 44
  • 59
0

For our production servers we went with Dell, I knew I would have someone onsite in 4 hours if something happened to the hardware. Servers have better fans, airflow and included software for monitoring. Tech support is able to see what you purchased, knows every aspect of the system and how each piece interactes with each other. We'd also never consider buying second hand for production, you never know how they hard drives were abused.

For pre-production & the dev labs, we went with custom built, because we didn't care if they were not quite as good, if a part failed and we had to wait a few days not a big deal.

Even the custom built ones, we never actually built ourselves, we just picked all the parts from an online store and ordered it assembled, usually saved us about $400 to $800 a machine compared to the production ones, but we'd never have considered using one of them for production.

LEAT
  • 217
  • 1
  • 2
0

Go with a dell server. I have many years of experience with this!! Custom built servers never cut it.

IT_Fixr
  • 235
  • 3
  • 12
  • +1 Much beeter choice to go brand name – Dave M Feb 23 '11 at 20:37
  • While there are many reasons to use a solid brand name server the comment `Custom built servers never cut it` is massively overstating the case. Maybe you should have had someone else custom build your servers. – John Gardeniers Feb 23 '11 at 20:50
0

Even if the spec is similar, the parts quality will typically not be. Likewise, support. If you don't need reliability and support (including a repair contract) then build it yourself, by all means.

mfinni
  • 35,711
  • 3
  • 50
  • 86
0

Is the local development server critical for you or just a play thing?

Business you can't afford downtime, cause it costs money. Dell provide service and matched hardware. You could build your server cheaper, but do all the parts work well together?

I would go Dell if price wasnt an issue with a hardware support. It definietly pays when your drives fail and you can turn around in a day. It doesn't pay when your supplier of your home built server who especially ordered you that snappy scsi drive, which just now failed has to order you another. May take a few days or weeks..!! Meantime you just past your deadline on that project...whoops

IanN
  • 313
  • 1
  • 6
0

If you got the budget for the Warranty/Service contract with quick part replacement, AND you don't have the on site staff that knows how to troubleshoot hardware, then go Dell.

For my stuff, the fact that I can get parts I need for a white box server within an hour at the local computer shop, is a large benefit.

Porch
  • 680
  • 5
  • 12
0

I own an IT Company. I have built Servers in the past for customers. I can always beat Dells prices by quite a bit I might ad.

Remember 2 things though 1) When you build it, you warranty it, can you touch Dell's warranty without marking the Server and or Parts up in case you need to buy a spare part down the road and stay under Dells prices 2) Will that part be available when you need it 2 years from now. If you buy a dell, they make sure they have the part, in a warehouse somewhere. Can you guarantee that for 1, 2, 3 years down the road?

After building Servers for clients for a couple of years, I realized quickly that I could not compete with Dells prices, including Warranty and Service Contract.

Truthfully there is no question considering the facts. ;)

Tom
  • 745
  • 3
  • 9
0

If you know what you are doing, build.

I have built many servers, both critical and not, and they always perform (per dollar) much better and more reliably than commercial (Dell, HP, etc) servers. They also are always well fitted to the task. And, they fail less often. Some "appliances" may be better purchased, like SANs, and even then, only in high-throughput scenarios. There are lots of very nice Linux-based DIY SANs out there.

Here's some thoughts:

  • OEMs, like Dell, are optimizing their supply + support costs. If a second-tier part costs them less to buy and support (meaning it may cost the OEM less but carries higher support costs, yet the total is less) than a top-tier part, that's what they may go with. That means you are the one left holding the server that pops while the technician comes in with the board to replace. (If you ever seen a field tech come in and say "Yeah, those things fail all the time", you'll know what I am talking about.)
  • If your server is critical, a four-hour-onsite warranty is absolutely not what you need. What you need is a cold spare, a hot spare, a cluster or other availability solution where downtime won't be felt by the end-users. Again, this will be cheaper if you build yourself.
  • A warranty is really a useless thing. You could use some of that money to buy enough spare parts to keep in inventory to cover hardware failures, and swap out the part yourself much faster than waiting for the phone drone to go through the spiral-bound book to diagnose the issue you already figured out, and then eventually send a field tech to replace it.
  • Also, if the hardware failure is intermittent, you are just as likely to have to fight the OEM for a long time to get a part replacement. With your own DIY server, you make the call.
  • The idea that OEMs keep all parts for umpteen years is an exaggeration. Usually, if you get mainstream parts, those (or like replacements) are also available on the market. Let's face it, most OEMs nowadays are JIT assemblers. If an OEM can source the part, or a similar part, likely so can you. Too add, many times when an old serv fails, you sometimes take the opportunity to move up.
  • Yeah, sometimes, some parts don't work well together. But, who deploys a critical server, OEM or not, without testing? If you stick with name brands, you will very rarely have issues. I've built hundreds of servers and have had few incompatibilities between parts.
  • You can't go from never having put parts together to building a critical, high-availability cluster without a) starting small and b) doing your research. The poeple who try "build", have a bad experience and then go to "buy" are usually people who went about it the wrong way.
  • One last thing: Either go with a branded OEM (like HP, IBM, etc) or build yourself. It's the middle ground that can be problematic. Fly-by-night assemblers with poor support abound.

I favor build.

alphadogg
  • 285
  • 4
  • 17
  • To add, see Facebook's Open Compute project. They built their data center with "vanity-free", meaning non-branded, servers. – alphadogg Nov 09 '11 at 14:34