That Time I Knew

Digging through the archives again.

I was writing about AI this morning and I’ve been updating my recent CV, so naturally I go back to some stuff I used to do, like run 500 databases on a single server. That was an HP Superdome back in 2005:

So you learn something every day.

This week, I'm playing with the biggest hardware I've ever worked on. I've been on an SP-2 before and I've worked on an E10000. I've even had a fractional Regatta. But the new heavyweight champion in my little world is an HP Superdome with 16 Itaniums. The damned thing is running NT.

But this is no ordinary NT, it's the 64 bit Datacenter Edition, something my many customers have avoided with one notable exception. That was the guys at Waste Management in Houston, who for various reasons were given to continual scoffing.

Today, I'm about to see how monster this baby can go, as I'm heading up a performance benchmark team. Should be interesting, especially if we decide to bump it up to 64 processors. Yow!

Yeah we did. We ran 500 database models on that bad boy. Three years later I was on the cutting edge of something entirely new.

Server Farms
This is pseudo insider information. When I was up in Redmond last fall, Microsoft was seriously hard up for server farms. Internal projects that had been working for months would be ready to roll out into production and suddenly people would find their projects facing no-go decisions because there are simply not enough servers to go around.

How does a company the size of Microsoft run out of server farms? It's complicated, but the short answer is that Microsoft internal security is extraordinarily demanding on end-users. Remember what Schneier said about benevolent worms? This is essentially what MS does to everyone on their intranet - they use benevolent worms to keep everybody on the inside inoculated and safe from the barbarian hordes who obviously want to hack Microsoft.

The other insider bit of info I happen to know is that Microsoft has initiated a new capex system to manage their current server farms better, and this plan is due for a major expansion. So if you were an internal client to the service that doles out servers you get a bill from this internal group. That particular internal group is expanding their capability to manage server resources by a factor of about 20.

The addition and annexation of Yahoo server properties and management would be a huge boon to Microsoft which is rather behind the 8 ball internally and acutely aware of a need to expand. There's a new high tech colocation facility in Seattle, I forget the name of it, but people talk about it in hushed tones - and they are booked to capacity out a year. Nobody has the state of the art server farms that Microsoft needs right now, and I'm sure that is a strategic necessity for them.

I knew I was right but just poking around because I remember a very cool product called Groove designed by Ray Ozzie, I found this gem:

In October 2008 Ozzie announced Microsoft Azure, the first project to emerge from his advanced development labs focused on new and potentially disruptive approaches to Microsoft's business.

So yeah. I couldn’t imagine what the cloud was going to be, but I knew Microsoft was building its infrastructure.