Pacific Connection(英語)

Playing the "Castanets" at Marimba

An Interview with Kim Polese

In 1995, four former members of Sun's Java development team ponied up $15,000 of their own money and self-started a company to do something big with Java technology. The group was certainly credentialed. Arthur van Hoff was a senior staff engineer at JavaSoft and co-author of one of the first best-selling books on Java: "Hooked on Java." His co-author, Sami Shaio, created Java's security model and the abstract window toolkit. Jonathan Payne prototyped the first Java debugger and was the co-architect of the HotJava browser.

Taking the helm as CEO was Kim Polese, the former Java product manager, who picked the name for that product. (Other candidates were "silk," "ruby," and "dynamo.") Polese got to name her new company, as well. A jazz dancer who still performs in her spare time, she christened the startup ">Marimba," its first product "Bongo," and its second, flagship technology "Castanet."

Castanet is hard to pigeonhole into a specific product niche, but many in the industry have classified it as a "push technology." The Internet can be likened to a big ocean in which people go fishing for information--often pulling up old tires and tin cans instead of marlin and tuna. Push products imagine a different sort of fishing expedition, in which you go out on the water and the best fish in the sea--ideally, the ones you are looking for--jump right into your boat. The best known purveyor of push products is PointCast. Download PointCast software, and you can display everything from sports scores to weather reports, to breaking news on your screen, without having to seek it out. Now, many companies, including Netscape and Microsoft, are getting into the act with their own products--each of which in some way brings in information on its own.

Indeed, the term "push technology" has gotten so broad and blurry that many companies, including Marimba, don't want to be saddled with the name. For one thing, there's the down side--bring in too much information, and users are inundated. After all, the original push technology is e-mail and junk e- mail, the hated spam of the Internet, demonstrates just how easily it is to turn a highly practical communications medium into a nuisance.

At the heart of Marimba is a new, patent-pending protocol for sending code over a TCP/IP connection. Potential applications include the ability to download and quickly update applications while keeping network traffic to a minimum. Marimba describes these services in terms of a broadcast metaphor, with Castanet "transmitters" on the server side, "tuners" on the clients, and "channel" applications acting as the user interface.

Marimba's other product, Bongo, began as yet another Java GUI builder--an improvement of the abstract window toolkit . Marimba's development team left Sun on Friday and started writing it on Saturday. But while Marimba still offers Bongo as a way to build Castanet channels, Castanet is now the star.

These days, Marimba is a hot company. After first eschewing venture funding, it received $4 million in financing from Kleiner, Perkins, Caulfield & Byers. It has also signed deals with some big companies, including CNN and Netscape. Unlike many people heading startups, Polese can point to actual working applications for her product. All this attention has kept Polese on the go. Famous for bicycling downhill on winding roads at insane speeds, she agreed to slow down just long enough for an interview with Software Design.

An Interview with Kim Polese

Marimba may be new, but the "push" market is already crowded. Are you concerned about getting crowded out?
It's funny because six months ago nobody was even mentioning the word "push"-it didn't even exist. This is the software business. When there's a new hot technology, people pop up all over the landscape and claim that they are in the space.
By "push," I mean shoving a file or a piece of content across the wire down to the user. It's great if a little flash pops up on your screen when your stock report comes in. But what if you want to do sophisticated "what if" scenarios on that stock based on market conditions? Or what if you want to manipulate an animated chart to see how your portfolio would do if you bumped up one stock and reduced your holdings in another? That's really in the sphere of an application. Deploying an application means deploying both software and data and then maintaining it in an intelligent way. Suddenly you are into intelligent application management, which is really what we are doing.
Where did the idea behind Marimba come from?
We started with the big picture, with a vision of taking Java to the next quantum leap--although, as it turned out, Castanet, is not even Java-specific.
When Java was conceived in Sun labs in 1990, Sun had a very broad view of a networked world, in which everything is connected to a network and every device has a network connection. In this world, applications would have to be capable of zipping around the network. Bits would have to be able to land on a device, install themselves magically, update themselves magically, and all do this in a totally secure, platform-independent environment accommodating everything from small, handheld devices all the way up to big servers. That was the big idea behind Java. That's why programming language was designed specifically for applications that were network- centric.
The first iteration of that language had Java applets running in browsers. That was a very important first step for achieving the full vision, but it was just a first step. Applets and browsers were very small programs, typically only a few hundred KB or even less, and that's why we thus far have seen Java largely used for animations and very simple things. Because it's a browser you have to constantly download the program again and again and it can't be persistent--it can't stick around.
We wanted to unleash Java and accomplish that vision of full- blown Java applications that are deployed automatically and updated automatically across the network. That's what we set out to do when we left Sun. What we ended up building was a product that is not only Java-specific, but is in fact agnostic--you can transfer any kind of bits across the wire.
If "push technology" is a small part of what you do--how would you characterize your product?
We do intelligent application management. For example, a lot of the other companies out there are doing "push" in the sense that they push a file down the network. But if one word or one byte changes in that file, then they push *another* file down, and that file can be several MB in size, even though just one byte changed inside of it. That's not intelligent application management. That's dumb push. That's not what we do.
What we do is just send that one little byte. If it turns out that that byte is shared by other applications that are already on the disk, we won't send it twice. We'll let the application share that byte or that GIF, that image. We do this in an environment that can scale to millions of end users. You've been compared a lot to another company, BackWeb--which also talks in terms of "personalized broadcast channels."

They compare themselves to us. They are very eager to be seen as a competitor of ours.
You don't think it's valid?
They do one small piece of what we do. They do dumb push, and they do it in a way that focuses on very simple information delivery. They have these flashing dolphins that appear on your screen and alert you to the change in a stock, or tell you that something has changed the Web site. They typically pop up a Web browser.
What we do is deliver applications. As an example, Aetna is using Castanet to deliver a full blown application that allows HR [human resource] managers throughout corporations in the United States to determine what benefit enrollments [i.e., insurance, medical plans] are appropriate for their employees, and to manage those benefit enrollments as well as financial planning information. In terms of its scope, the application is on the order of an inventory control application.
And it's considerably bigger than a typical applet?
Yes--large enough to require intelligent management. And that management must be personalized because the HR manager at Nike is going to need to receive a different set of information than the HR manager at Coca Cola. And not only different information, but probably different software as well, which means personalization is really key. So suddenly now we are talking about deploying an application, which is a much more complex problem and requires a sophisticated solution. We're not talking about flashing dolphins appearing on your screen.
I understand you have a deal with Netscape and its Constellation product.
Yes. We have licensed them our tuner software so they can embed it. The tuners are small pieces of client software that act as an invisible service to download and upgrade applications. Netscape is taking the core of our tuner and embedding it in their new next generation desktop, which is called Constellation. So the user interface of our tuner will be essentially theirs. Our user interface will not appear. But Castanet channels, like those that CNN, Yahoo, and Aetna are building, will appear as channels on the Constellation desktop.
Do you expect to do other deals like that where you are not branded?
Absolutely. We've created a very extensible architecture on both the tuner side and the transmitter side--the transmitter being our server software. We are enabling all kinds of vendors throughout the industry to embed pieces of our technology in their products. TriTeal is another company that is embedding our tuner. They are building desktop interfaces for network computers. There are other companies on the server side that are embedding our technology. Oracle, for instance, is embedding the update capability in their universal server so that a database event can trigger an update in a Castanet channel. Cybercash is linking in billing and payment protocols to the transmitter, so you can have business models like pay-per-view or pay-per-download. Those are examples on the server side of how we are extending the platform so other people can plug into it.
Who else is using Castanet?

The range is pretty significant. It's everyone from Aetna to Bellcore to CNNfn [CNN Financial Network] to Yahoo! The reason why the range is so significant is that all these companies share common needs. They all need to deploy applications to a customer base that could scale eventually to millions of users. They need to do it in a heterogeneous environment, in other words, they cannot predict what kind of device or operating system that person will be running on at the opposite end. They need to deploy real applications, not just simple information updates. And they need to provide compelling, media-rich, intuitive applications that can automatically update themselves.
Is there an example you can point to as being the state-of-the- art, in terms where you want to see Castanet headed?
CNNfn is a good one because they are deploying not only financial information, but also the ability to manipulate and personalize it. In other words, again, it's not just simple information delivery. I want to interact with the application and try out scenarios for possible investments I might want to make. And CNN also wants to develop a personal bond with its users, so there's a relationship that compels them to come back again to the service.
Also, they are ahead of the curve technically, as a company. I had the opportunity to meet with Lou Dobbs [CNN executive vice president and host of the popular program, "Moneyline"] and have dinner with him last week. He flew out for a launch event we did last week, and got up on stage and said a few words. What impressed me about him and his small team is that they aggressively adopt new technology. They do a very thorough assessment of new technology and then adopt the best- -like the early use of digital editing systems. CNN, with Dobbs leading the charge, was the first major broadcaster to adopt the DNG [Digital News Gathering] system from Avid Technology. Now it's a de facto standard throughout the broadcasting industry. They are now pursuing the same approach with online technologies.
Do you consider PointCast a competitor or a possible ally?
I consider them a possible ally. They are a content aggregator- -like any network television or cable station--where they take a lot of content from a lot of different sources and then package it up and provide it to consumers. Their business model is advertising revenues-based. We, on the other hand, are a technology company. The difference is our revenues are based on licensing technology, and our customers are not consumers. Our customers are developers.
Does your technology have any bearing on the thin client paradigm?
Actually it does. That's an interesting question. The funny thing about network computers is people have sort of gone to the extreme with the idea--they talk a lot about network computers being diskless and almost dumb. That's not a model that we as a company believe in, in the near term, anyway. Bandwidth is so scarce that if you have to download a word processor, for example, every time you need it, you're going to be sitting there for 15 minutes. We believe that the network computer is absolutely a viable concept, one that makes a lot of sense because of the zero administration focus of it. But you need still some persistent storage so that at least you can have some local files and run some applications locally--so that you don't have to download *everything* from scratch every time you use the device.
The problem has been that when people think disk, they think of a big, heavy operating system to manage that disk--because typically, that's the world we've known. That's where our technology comes in, because instead of needing a big operating system to manage files to manage the disk, the tuner can remotely manage the disk. It can act as a remote disk administration system, a very lightweight one, that communicates with the transmitter at the server, where the content provider or the IT manager is deploying the applications down to the user base. So on an ongoing basis, every time you fire up that network computer, you never have to install software because it's already been installed for you using the tuner, and it's personalized to you and you never have to update it because, again, the updates happen through the tuner and the transmitter pair. You still get all the benefits of having a small, lightweight device because you don't need to have megabytes of disk space and megabytes of memory to hold a big operating system to manage that disk.
What would this machine look like?
Very similar in form factor to all the network computers being produced today We'll just simply have a disk with a flash ROM- -some sort of persistent storage. The funny thing is that disks and RAM and ROM are cheap. CPUs and disks are cheap. What's expensive is bandwidth, and that's why it makes sense to have some form of local storage on these network devices. The first generation of the diskless devices may not succeed. I think that people will start to realize that it doesn't add significant size or weight or cost to have some persistent storage, and that will start happening--they'll start adding that to network computers and we'll start seeing these things being actually viable.
What are your plans for Japan?
We just signed up two of the major distributors in Japan to distribute Castanet and Bongo--Hitachi and Itochu. We are busy signing up additional distributors, some of the biggest companies in Japan, to distribute and license our software. We are working with companies in Japan in two areas. One is distribution, and the other is with device manufacturers who are embedding our tuner in everything from laptop computers to smart phones.
Is that unique to Japan?
Not unique to Japan, but because there is so much development there, so many companies focus on device manufacturing, they are an ideal market.
How that might work?
These companies are building next generation phones that have services that can connect a phone company with a customer and allow them to do things like, for example, follow the stocks on the stock market, or view a restaurant guide for a particular city that they are in. They want to have a phone that can actually morph and change based on the software that it runs. So this week I might use the phone to give me stock market conditions and let me figure out how my portfolio is doing at any given minute. Next week, it might download a restaurant guide. It might even talk to my travel agent, find out that I'm travelling to Washington DC, and give me the Washington DC restaurant guide.
Do you have any sense of Japan as a unique market in its own? These sound like worldwide applications of the technology.
The thing that makes Japan unique is that they are usually far ahead of the curve in providing these kinds of devices and services. Their expertise in microelectronics and in producing devices very cheaply for a mass market is far ahead of any other region in the world. That's why we made three big trips over there thus far, and why we are very busy signing up deals. Our tuner technology is also ideally suited for these very small handheld devices. It's scaled down to a very small footprint.
Your departure from Sun seems to have ended up amicably. Are you working with them in any way?
Yes we are. We haven't announced any deals yet, but yes, we are working closely with Sun.
You say you started out as a Java vendor, but now you are "agnostic" toward it. Did you write Castanet in Java?
Yes, we did. Every line of code was written in Java. We did that for a couple of reasons. One is that it gives us portability: our tuner will go anywhere that Java is, which means everywhere. Another reason is that Java is a superior programming language. It allowed our three programmers--we only had three programmers for the first six months--to create an entire two products and get them to customers and have them fully tested and essentially bug free. That could not have happened with C++. When it comes down to it, Java is a better programming language--it lets you write applications faster with fewer bugs.
I read in the Wall Street Journal that you are not yet a millionaire and you don't own your own house. Is this all going to change?
I never like to presume any outcome. It could change. What I'm really focusing on is execution and the ultimate goal of returning value for my shareholders. As long as I keep my sights focused on those things, the rest will theoretically come.