ODM for the masses? Facebook's OCP still ain't for you, brother

Go to the Computer History Museum in Mountain View, California, and you’ll see a strange contraption cobbled together from commodity motherboards purchased from electronics stores. It’s one of Google’s first production servers, built in 1999 when it didn’t have money to waste on dead-end projects like Wave, NexusQ and Buzz. …

[Anonymous Coward]

"“Then six months later you need to buy some more, and that box isn’t available any more."

We once needed a more powerful desktop than the company's standard "Office" issue. Tesco had a suitable branded one going cheap - but the company insisted we buy a "business" model from the manufacturer at a significant mark-up. The brand and spec was the same - but the company reasoning was that the manufacturer guaranteed to supply specific "business" models for some extended time.

We bought two. Later on we decided to buy another one as a hardware spare - as our IT department were quoting lead times of several weeks for any repairs for a non-standard box. It then transpired that although the case and model didn't change - the manufacturer couldn't guarantee that the motherboard would be the same. As we needed interchangeable identical units this was a problem. Not long after it turned out that the particular "business" model would no longer be available anyway.

So we paid a premium for a non-existent guarantee of availability.

[Chris Miller]

That's absolutely a problem for anyone buying computer boxes one or two (or even a hundred) at a time. But for large enterprises buying thousands of standard PCs a year, it's common practice for the supplier to commit to building identical machines from identical components for a minimum period. This can never be entirely guaranteed - events such as the 2011 Thai floods may disrupt component manufacturers - but as a major customer you should get preferential treatment, at least to the point where the software build (drivers etc) doesn't have to change.

[joemama]

Where is OCP Quality?

About 1 year ago, Chris Mellor exposed the fact that the OCP also did not have legitimate testing and he brought up several points that sent shock waves through OCP. One of the OCP labs, UTSA, appears to be shut down. And since then, certification is supposedly being handled by an organization called ITRI and the University of New Hampshire. However, since his article, there has been very little information on OCP testing, with the exception of open networking products.

So, in addition to today's article, it's probably safe to say that any customer interested in OCP also needs to build up a test/evaluation team to simply vet whether or not the hardware is good enough to deploy. It's shocking to read that companies such as Goldman Sachs and other are even considering this. OEM quality is not great, but OCP is, IMHO, rolling the dice on something that might have severe problems. For firms responsible for the most sensitive of data, this seems to be penny wise and pound foolish.

[Dadmin]

Re: Where is OCP Quality?

Don't expect Mercedes quality in your Yugo. Don't expect your Yugo to handle, or be comfy, like a Merc.

RTA; the point it that small-time data centers are not yet ready for mass commodity hardware. Stick with the expensive systems, when you can't cope with a massive amounts of cheaply made boxen that will die. Oh, they will die. Google is big enough to play this way. That is the point.

[Cynic_999]

Often exact replacement is not a problem

The cost saving in a DIY unbranded unit will probably enable you to buy enough spares to cope with equipment failure, and by the time you need to expand you'll probably want to also upgrade to better, more modern units so finding an exact replacement is not an issue.

[Rainer]

The worst is OEMed white box stuff

We bought a couple of HP JBOD-HBAs - they are rebranded LSI 9207i models.

I would need a newer revision of the firmware - which is available from LSI, but not for OEMed cards.

HP doesn't care (obviously).

I should have bought the original 9207 cards right from the beginning...

At a previous (very previous) job, we had a ADIC tape library OEMed by Siemens.

At some point, while on the phone and confirming with an ADIC engineer, a firmware-update was installed, which instantly bricked the unit.

ADIC replaced it under warranty, luckily. I just wonder who had to eat the costs ;-)

With OEMed stuff, you generally get less and pay more. The OEMs - to paraphrase Scott McNealy - add about the same amount of value to a product as someone reselling bananas: just additional bruises.

As for those cards - I've now bought an original card (as a spare) and will try to flash the HP OEM card with the original LSI firmware, turning it into an LSI-card. A procedure for this is available somewhere on the FreeNAS forums.

If it fails, I'll use the original card. If it works, we can still decide if we buy 1500 bucks worth of LSI cards to replace the HP cards or just re-flash the HP cards.