A couple of months ago we shared with you the CPUs that are going into our new server farm. We've actually started physically installing the machines (hence the brief outage over the weekend) so it's time to share another piece of the server puzzle.

The final configuration we decided on was 12 machines. This is a significant reduction of the number of systems we have installed (currently nearly 30) but the performance per box is much higher, allowing for consolidation through virtualization.

We are building two private clouds: a lighter cloud of 8 machines for our application serving needs (including some redundancy in the cloud), and a 4 machine DB cloud to handle the heavier IO. We'll dive into our infrastructure design in the later, full article but for now let's talk about memory.

The application server cloud is light on memory. Each system in this cloud has 12GB of memory (6 x 2 DDR3-1333 DIMMs). The DB server cloud on the other hand has 48GB of memory per box (12 x 4GB DDR3-1333 DIMMs).

Kingston was nice enough to supply the memory for our project with. The 96 sticks of memory were broken down into 48 x KVR1333D3D4R9S/4GI and 48 x KVR1333D3D8R9S/2GI. If you want to see what 288GB of memory looks like, check out the gallery below.

Note that for all of the components we selected for this project, we decided upon the components first and then petitioned the manufacturers second. The stipulation was that the AnandTech server farm would be a publicly visible test bed. Any failures of the hardware are public failures and would obviously reflect poorly on the manufacturer. For CPUs and memory it's not so big of a deal - physical failures there are fairly rare, but for SSDs this provided an interesting challenge. More on that in our next installment.

POST A COMMENT

69 Comments

View All Comments

  • Friendly0Fire - Friday, October 29, 2010 - link

    Basic economics say that revenue != profit. If you factor in the costs (employees, hardware, maintenance, etc.) you'll see they aren't making so much money as to be able to throw it out of the window.

    If manufacturers are happy with donating hardware, I say let them.
    Reply
  • maxusa - Saturday, October 30, 2010 - link

    Your argument, Friendly0Fire, has switched the focus from a failed assertion about the earnings magnitude to suggesting that paying for a core systems overhaul is a waste of money. You need to try harder next time.

    I also recommend you read Anand's recent interview where he hints to very healthy margins after the costs are deducted. In any case, a multi-million enterprise with a handful of staff can easily afford a small capital expenditure every 5 years (that's when they overhauled last).

    Now about your last comment. How do you know that the vendors are "happy with donating" anything? All you know is that they agreed to do it; the rest is a spin. I will let you in on something... it is a business decision on their part driven by a perceived influence over public opinion which this publication exerts. Scroll down if you care to read my other posts about how it is so.
    Reply
  • iamezza - Thursday, October 28, 2010 - link

    you are a tool Reply
  • softdrinkviking - Thursday, October 28, 2010 - link

    seriously, man.
    it's not that big of a deal, it's just some memory.
    obviously from reading the reviews, anandtech criticizes tech whenever they see a problem with their testing, even with the companies who "donate" items.
    the proof is in the pudding.
    besides that, they are getting review items from companies, not store bought tech that everyone else buys, so you have to view the whole thing with a little bit of objectivity in the first place.
    the important thing is, they are being transparent about all of their dealing, and that is the best you could possibly expect from any review site.
    either you trust them to do what they do, or you don't.
    if you don't, then just walk, because you're wasting your breath complaining.
    Reply
  • teng029 - Sunday, October 31, 2010 - link

    do you want to send money for them to upgrade their servers? No? Then STFU and let them do business as they see fit... Reply
  • edpierce - Thursday, October 28, 2010 - link

    Kingston is the most reliable company I have ever come across in all my years of dealing with computers. Sounds like I'm a paid spokesperson but I'm not. Kingston is simply one of the rare companies out there that emphasizes quality control as much as a respectable company should. Reply
  • Speedye1 - Thursday, October 28, 2010 - link

    Interested to hear what platform your virtualizing on and to see the db config. Reply
  • c4keislie - Thursday, October 28, 2010 - link

    Make a 288GB RAM drive out of that memory, that would kick some serious arse. If only...

    I like these small tidbits or teasers or whatever you want to call them, keeps me interested and checking on an hourly basis for the full article (this is silly of course, but I do it anyway).
    Reply
  • Yaos - Thursday, October 28, 2010 - link

    No software calls a group of computers that perform load balancing and failover a cloud, it's called a cluster. Reply
  • Klinky1984 - Thursday, October 28, 2010 - link

    Load balancing and fail-over isn't the only thing involved in 'cloud computing. I agree that the meaning behind cloud computing is rather vague though. Ultimately in the enterprise space it's boiling down to a definition of modular, portable, virtualized servers that can be grouped together on a single machine or expanded to multiple machines easily. Reply

Log in

Don't have an account? Sign up now