Microsoft"s Muglia on "integrating the edge"

09.06.2005
Von Carol Sliwa

Bob Muglia, senior vice president of Microsoft Corp."s Windows Server division, this week offered an update on the company"s long-term vision for its server operating system, as well as related software products. Muglia also discussed Microsoft"s high-performance computing plans, as well as the company"s vision for "integrating the edge."

Microsoft said it will ship a Compute Cluster Edition of Windows Server 2003 in the first half of next year. Why did you want to get into that market? It"s about time. It"s growing very rapidly. It"s one of the faster-growing segments of the server business.

In which vertical industries do you see the most interest? Pharmaceutical, oil and gas, financial institutions, engineering organizations, etc. It"s about 7 percent of the [overall] server business right now.

That"s enough? Oh yeah. You"re at [400,000] to 500,000 units, and that"s a pretty big number. It"s an area where we clearly have a lot of catching up to do. It"s an area which is also changing in the sense that, historically, that set of capabilities was typically [centered around] academic [institutions] and very high in government.

We"re going to see a lot of shift toward smaller clusters being built -- four, eight, 16, 32, etc. clustered systems. Remember, a dual-processor, four-cluster system is an amazing amount of computing power. Very literally, scientists will be able to put more computing power under their desks than a supercomputer had even a couple of years ago. And in that kind of a world, you can change the way people work. Instead of building some model and spending time making sure it"s all right and then submitting it to the great computing center in the sky and waiting for two days or whatever for it to come back, you just run it. And it finishes in a couple of minutes, and you run it again, and you run it again, as you make tweaks. It"s a huge opportunity for those industries.

Do you expect much unit volume with the Compute Cluster Edition? Oh yes. We have about 5 percent share right now of compute clusters. And if you think of it as being a 500,000-unit-a-year marketplace, there"s just nothing but upside for us.

How different will this operating system be from other versions of Windows Server 2003? It"s the same operating system with two differences. One, we basically turn off a bunch of workloads, so you can"t do a bunch of other things on it, because it will be priced very competitively. Pricing on this will actually be less than Standard Edition for the operating system piece itself.

Then there"s a whole set of services that need to run on top of that -- batch schedulers, the ability to connect these computers with high-performance connections and those things. Even in the Linux space today, people tend to buy those things from commercial companies, and they pay a fair amount for it.

I think we"ll have a very competitive offering relative to the Linux-based solutions. And we"re building both pieces. We"re doing a restricted-function version of the operating system, and then we"re also building these incremental services you need for compute clusters. What we want to do is provide a very complete solution set for our customers.

Will the services be a separate stock-keeping unit? They will be available as a separate SKU, and we"ll probably put the two together as well to make it convenient.

Will they run on any other versions of Windows, beyond the Compute Cluster Edition? They will run on Standard and Enterprise.

During your presentation at TechEd, you talked about the company"s plans for "integrating the edge" and how that vision will take five to 10 years to achieve. Why it will take so long? The reason why it will take a long time is there"s a lot of components that need to change. Some of them come from Microsoft, and a lot of them are about the infrastructure. [For] example, IPv6 is a key transition. Moving the world to IPv6, some might say 10 years is being very optimistic for that. I think what you"ll see is that parts of organizations will go IPv6. The Internet itself will take a long time to fully move to IPv6 or even mostly move to IPv6.

Now, where should it be, to me, there"s two different dimensions. From an end-user perspective, [a] PC is hooked up to the Internet right now [wirelessly]. I should be able to open it up, turn it on, and I should be able to do anything I could do within my corporate intranet without having to do a single thing. It should just work for me. The only thing I should need to know is, "Is it plugged in or not?" If it"s not plugged in, then I can do things that are cached, if I"m on an airplane or whatever. Even then, I"ll be connected in many cases. But if I"m not connected, I"ll have to work off my cache. If I am connected, which should be most of the time in the future, I should be able to do anything anywhere.

From an IT manager"s perspective, they should be able to define policies as to what I have access to, when I have access to it, within what circumstances. And it should be independent of their network topology. Right now, things are very dependent on topology. So some things are possible with an intranet; some things are not possible. You can VPN [virtual private network] in. That"s kind of an all-or-nothing sort of thing.

To what degree are IT shops thinking that way now? It"s just beginning. Seventy percent of our traffic at Microsoft is authenticated with IPsec. I think we have more IPsec traffic in our environment than anyone else. It may be more than the rest of the world combined. It"s huge. We know that [the U.S. Department of Defense] uses IPsec in a set of ways, and some other big companies do, but in much more limited ways than we use it.

We did that because we really wanted to secure our network. As you can imagine, Microsoft is a focus of hackers coming in. So IPsec has really done a tremendous amount to secure us down. It"s still really hard to do. For example, to build IPsec-based policies, there is a white paper on Microsoft.com that describes it. I"ve looked at it. But I can"t claim I"ve read the whole thing. It"s about 75 pages long. It is dense. It is not easy. This is not making it easy.

Longhorn will take a big step forward in terms of making it a lot easier, and then there will be steps beyond that that we"ll need to take to make it easier. A lot of it is taking these things that are possible today and making them very practical and almost automatic.

What"s it going to take to get greater adoption? We"ll build the enabling technology in the operating system. Applications can take advantage of it. But the most important thing that we can do is have a consistent set of mechanisms for dealing with policies across an enterprise. The use of models to define these are critical. We should get to a state where there are best practices for security policies, best practices for deploying Exchange. There should be 10 best practices. And whether you"re a small business, a medium[-size] business or a very large distributed enterprise, you should be able to choose a best practice and just snap that in and then customize that to meet your individual needs and have the policies emanate from that, versus having to handcraft all of these things every time. But that"s going to take a little while.

Think of it as preconfiguring. It"s like the difference between a recipe book and take-out. We have recipe books today. You buy the ingredients, and you put it all together. We want to get to the take-out sort of style, where you just place an order and it works.

What do you think customers will tackle first? Federation services. I think that technology becomes viable this year for people to deal with when R2 [of Windows Server 2003] ships later this year. Another one that I think we"ll start to see over the next couple of years is more use of two-factor authentication, so smart-card based authentication. ... We"ve now seen that identity theft is one of the biggest problems that"s emerging on the Internet. So that"s a very big thing.

You talked about your vision for distributed storage being years away from being realized. How far along are you now versus where you want to be? I don"t think that we have tapped even a few percentage points of the capabilities of storage systems over the next few years. The storage continues to evolve in amazing ways. ... It"s kind of like Moore"s Law. We just keep getting more and more. It doesn"t get a lot faster, which is interesting. A 250GB disk is not a lot faster than a 100GB disk. But it"s much, much, much denser. So, how do you then access that information?

We do take a big step on that in Longhorn, where we have off-line caching so that applications can work locally, just like Outlook 2003 does. In R2, we have the ability to do it to branch offices between servers when we changed the Distributed File System. And then we"ll continue to make the storage itself richer and richer.