0

I'm looking for advice on the classic build vs buy question.

We need a new linux server to run Matlab computation on in our lab (academic). Matlab parallel computing toolbox licence allows up to 12 local workers so we are aiming at a 12 core server with 4GB memory per core (total of 48gb). The system will have an SSD for the OS and a raid-5 (4x2tb) for data.

I looked around and found a (relatively) cheap vendor, Silicon Mechanics, that offers a system to our liking (specs below) for $6732. However, buying the components from newegg cost only $4464!

The difference is $2268 which is 50% of the base cost. If buying from a company can be thought of as a sort of insurance, basically my premiums are of 50% of the base cost which to me sounds like a lot.

Of course any downtime is bad, but the work is not "mission critical", i.e. if it takes a few days to fix it when it breaks its no the end of the world. If it takes weeks to months then its a problem. If it breaks 2-3 times in 3 years, not too bad. If it breaks every month not good.

In term of build experience, I set up a linux cluster in grad school (from existing computers) and I build my home pcs but I never built a server before.

The server components I'm thinking about:

1 x SUPERMICRO SYS-7046T-6F 4U Tower Server Barebone Dual LGA 1366 Intel 5520 DDR3 1333/1066/800 ($1,050)

12 x Kingston 4GB 240-Pin DDR3 SDRAM DDR3 1333 (PC3 10600) ECC Unbuffered Server Memory ($420)

2 x Intel Xeon E5645 Westmere-EP 2.4GHz LGA 1366 80W Six-Core ($1,116)

4 x Seagate Constellation ES 2TB 7200 RPM SATA 6.0Gb/s 3.5" ($1,040)

1 x SAMSUNG Internal DVD Writer Black SATA ($20)

1 x Intel 520 Series 2.5" 180GB SATA III MLC SSD $300

1 x LSI LSI00281 PCI-Express 2.0 x8 MD2 Low profile SATA / SAS MegaRAID SAS 9260CV-4i Controller Card, $695

0

2 Answers 2

3

You're looking at a one-off server, with defined performance parameters, and you know your uptime requirements. When considering the DIY or Buy option, you already know about the difference in recovery time between the two options. One thing to consider is the time it'll take to ship a new part once something breaks. For a warranty service next-day or next-business-day is entirely possible. For a DIY server it could take a week depending on the shipping options used.

In an academic environment, I know that departments are sensitive to sudden unexpected costs. Costs like something major breaking and having the replacement be purchased NOW and use not-cheapest shipping when doing it. This is one reason such departments like to front-load costs in the form of a server + warranty.

If you believe you can get the resources to handle breakage during the lifetime of the server, and 'week' is a good-enough recovery-time-objective for you, then the DIY route makes sense in your case.

1

"I'm looking for advice on the classic build vs buy question. " - I barely need to read the rest of your post.

Unless you've got some kind of specialist requirement that makes buying a server in a problem, then that is what you should do. It's usually more reliable, you can get support in case of weird compatibility issues (and they are less likely for parts that came in the server), you can get it backed with a warranty that meets your needs, including hot spares on site if that's what you want to do.

And if cost is an issue, unless you're working for free its probably cheaper to buy a server and use the time it would have taken up to do other work tasks that can't be solved with an off-the-shelf part, instead.

Not the answer you're looking for? Browse other questions tagged .