Ask HN: Why not disclose hardware info in web scale datacenter?

5 points by ZhuHan ↗ HN
I know a lot of large web companies are generous to contribute software project to open source community. Some of them are just for education purpose, not even taking anything back.

Why are they not willing to disclose the hardware specs, or data center operation experience with the outsiders? Do they position it as their secret weapon of competition? Or it's just so boring that nobody is interested at all.

7 comments

[ 35.5 ms ] story [ 744 ms ] thread
If other companies know your numbers, they know what's possible. They then have a strong incentive to keep throwing engineering effort at it until they achieve what you have.

It's like how Jobs's visit to Xerox PARC sparked the Macintosh. Actually, Bill Atkinson went beyond what Xerox did - he misremembered the visit, thinking that he'd seen overlapping windows on a computer, and continued tinkering until he implemented them for the Mac, on the basis that Xerox's demo constituted an existence proof that it was possible. It later turned out that there were no such technology, and the Alto used tiling windows like everything else.

That makes sense. But what will hurt them if google does not disclose any performance metrics, e.g. request/second, disk seek/second, but only the detail hardware configuration and failure model.

Google has published some data on disk failure in the past. The published another paper on failures in storage system in the coming OSDI 10[1]. Funny!

[1]http://www.usenix.org/events/osdi10/tech/techAbstracts.html#...

It's competitive advantage. Their hardware cost is where their margin is. If they can run a server with half the power/cooling, that's a significant advantage over the competitors. The major cost of a server these days is power, cooling, bandwidth.
If google disclosed their hardware configuration in data center, I suspect a lot of other small companies will hurry to the vendor and subscribe the same components. Looks like a good deal for these hardware vendors.

But I do no see any vendor label their product as "Google is using it".

That's because Google builds their own servers.

Rackable Systems is the company that supplies servers to a lot of the other top sites (disclaimer: no personal connection to Rackable Systems.) Most smaller companies just buy 1U servers and hire a good network admin to string 'em all together. (former managed hosting company CEO)

Many data centers use virtualization, so it makes no sense to talk about machine specs. Most VPS deployments are over-subscribed, meaning that if everyone hit it at the same time, then you'd see only a fraction of the claimed performance.
If you really want to know the hardware specs for a given hosting operation, there are two ways to do it:

1) If large company, find out who their network admin is. Buy this person a beer and/or dinner. 2) If smaller or mid-size company, find out who their hosting company is. Buy the lead ops guy/gal at the hosting company a beer and/or dinner.