Sun Rewires Enterprise

Thin clients, sensor-based networks on tap, says Mitchell.

After leading Sun Microsystems Inc.s Sun Laboratories for three years, Jim Mitchell, vice president and Sun Fellow, is moving on to run the companys High Productivity Computing Systems Research and Development program. A founding member of Sun Labs in 1988, Mitchell successfully meshed innovative research with product development, including the releases of Java and Jini.

At the recent SunNetwork show in San Francisco, Mitchell sat down with eWEEK Senior Writer Anne Chen and Technical Analyst Francis Chu for a look back at the work Sun Labs has accomplished and a look forward at how Suns research will change the future of enterprise computing.

Which projects is Sun Labs focusing on now, and how will enterprises take advantage of them in the future?

Asynchronous logic [is] being used in some of Suns current products and is also part of the reason that DARPA [Defense Advanced Research Projects Agency] awarded us this HPCS contract. Were getting to the kinds of geometries on chips these days where they have become the limiting problem. To get around that problem, you add more complexity, which is bad.

We [are doing] a fair amount of Java technology, particularly on the next- generation 32-bit smart card, Java virtual machine, J2ME [Java 2 Platform, Micro Edition] and Garbage Collector work. We tend to do a number of things on how you engineer Java virtual machines better.

The Sun Ray product was initially a labs project, and weve gone beyond that in a couple of ways. [One way is by using] the exact same hardware—but changing the software on the client and server side so that by doing compression and by doing encryption, you can run server-based applications over DSL and outside the firewall securely.

A current piece of research [involves creating] a Java application that runs on any Linux, Mac or Windows machine and then turning the machine into a wide-area Sun Ray thin client. Weve found a way to turn wireless PDAs into thin clients for use in the enterprise, where you can move applications from a desktop to a PDA.

What kind of security research are you focused on?

We have a lot of work going on … around sensor-based networks—where there are a gazillion devices on the network. Eventually, everything, like light switches and doorknobs, will end up being a network device.

Were interested in the right architecture to make that work. Its not just about systems administration, but automatic networking, and what security is needed in such a system to protect peoples privacy and security. The work were doing is to make sure that all these small devices can have a very high level of security.

You need automatic security, and thats a challenge. The basis for that automatic security is encryption, both public-key and vault encryption. So one of the projects that weve done is to implement elliptic-curve cryptography and do an implementation that is so small that you could put it into small devices.

Its about the cloud of small devices and making a world we [are not] afraid to live in. … The electronic devices I carry with me should be anonymous, and if I want to do a transaction with a commercial entity, then we should talk encrypted. Other stores should not be able to find out that Im conducting that transaction. I want my privacy.

What are some of the hardware and software innovations you are working on for HPCS?

Not only does the hardware have to run fast, but it has to scale to a petabyte. The [worlds] current fastest machine is a 40-teraflop machine, and DARPA wants to get to a petabyte of main memory—which is a factor of about 200 times faster. The only way to do that is to use 100,000 processors. That means you have operating system challenges: How do you manage applications running on 100,000 processors with 300,000 threads? And how do you program that thing?

The difficult thing is writing the code, getting the machine to do what you want while getting the power its capable of. And they want the program to automatically take advantage of the machine. At the same time, how do you make things very low power? These are issues were working on.

What technologies being developed now will make the largest impact on enterprises during the next five years?

At the hardware level, asynchronous logic will … drive down the time it takes to develop things and also make it easier to integrate things because they wont have a common clock.

On the software side, well be taking what weve learned about Java and applying it with other languages than Java so that it can affect all applications that run on a computer. Dynamic compilation is a good idea. Weve done a number of things in Java, like speech recognition, where weve transliterated something written in C and it runs faster in Java. We continue to make Java as ubiquitous as possible, and that will show.