What's happening with servers and storage?
There are a couple of things happening. Virtualisation and multi-tenant systems are really going strong. Then there's the leap in performance with the new [Intel] Nehalem architecture — a nine-times improvement in memory bandwidth. We're coming out with servers that support up to 1TB of DRAM. That's incredible performance — you can put your whole database in RAM, or virtualise 20 servers.
We're seeing it accelerating the refresh of server infrastructure, and also accelerating migration from legacy Unix platforms. Virtualisation has the ability to attach many of these servers together to get performance you couldn't get before. It used to be that virtualisation was fine, but you couldn't do this, you couldn't do that. Now there are things you can't do if you're not using virtualisation. There's a lot of headroom in virtualisation.
Storage really goes with the virtualisation trend — we saw this couple of years ago. If a customer had five or 10 servers, they were using virtualisation with no problem. But then they started to virtualise 100, 1,000, 100,000. How do you handle provisioning, snapshotting or volume recovery? We found this company, EqualLogic, that could virtualise pools of storage. Since then we've added over 10,000 new customers, and it's a ginormous business for us, growing very fast.
We're just at the beginning of moving that storage to 10Gb Ethernet. There's a lot of discussion about switching fabrics, and it looked to me that there was a trend where power was getting sucked out of the computer and into these really smart switches. I'm seeing it go the other way now. Switching is getting put into the computer, into the blade chassis. The switch is getting virtualised.
We're putting 10Gb on the motherboards of the next generation of servers, too. We have the greatest solution: it's Ethernet over Ethernet — great competition for Fibre over Ethernet. Just go talk to datacentre folks. They don't like the network within a network, where they have to deal with all this legacy stuff.
What about servers for the cloud?
We went to the largest web companies in the world — Azure, Bing, Yahoo, Baidu — and said, "What do you need?" We used to bring them general-purpose servers that were really nice, but they said they've got all these features that they don't need those. So we went out and built custom servers for these businesses.
When you look at what we built in the Chicago datacentre for Microsoft and Facebook, we sold them hundreds of servers. It's nothing like the most expensive servers in the world, like one of my competitors is advocating. The switching architecture is a lot simpler. It's back to the old Dell play of saving our customers money using open standards.
