John Paul on Network Services
On 27 August 1997, IC's Editor-in-Chief, Charles Petrie, and Acquisitions Editor, Meredith Wiggins, spoke with John Paul, senior VP and general manager, Server Products Division, in his office at Netscape's corporate headquarters in Mountain View, California.
First, why don't you tell us a bit about how Netscape's product division is structured, and what you do there.
We have two product divisions at Netscape: clients and servers. I'm responsible for all software that ships on servers. There are two major functions within the division: product marketing and engineering. They work closely together to define, ship, and build the products.
It's interesting that you divided it up between servers and clients. It seems that if you are promoting a vision of the "network enterprise," it's a combination of server, client, middleware, and a whole bunch of other stuff.
It is. We went to this structure last November. There were two potential ways to divide: one was along client/server; the other could have been along messaging and Web and application servers. We debated the pluses and minuses of both. I think we're organized properly at the moment because the battle to maintain a footprint on the desktop is what the client division really addresses. And the battle to provide great application and collaboration services is what we address in the server division. The two require different business strategies, different business models.
Is it your strategy, though, to essentially propagate the virtual machine, the Java Virtual Machine, so that your server applications can then run on a variety of platforms?
So then isn't your client strategy dependent upon what you're doing on your server side?
But the way we're organized actually doesn't keep us from doing that. If there's any liability, it's whether the client takes advantage of every feature we put in a server. That's what we have to constantly work at.
Since we're on this topic, there is the saying "Write once, run anywhere." But it really turns out to be "Write once, test everywhere." How do you respond to that?
The real problem area is the desktop, because there're millions of those. Netscape, in Navigator and Communicator, will ensure the presence of a consistent Java virtual machine. "Write once, test everywhere" is true only if the virtual machines are different, for instance.
They're different on different platforms and operating systems.
They're different for two reasons: inherently because the platforms are different (that's one we need to solve in the Java Virtual Machine environment), and then just because we happen to build them differently. Yesterday we announced an agreement with Sun and IBM to make sure that between the three of us we don't put in inconsistencies just because we're building them separately.
I think we're in the early days of Java virtual machines. And what's nice is that the market will winnow away those that aren't compatible. What's nice about the Net is that if you're not compatible, you very quickly get relegated away. Before the Net, there were tremendous inconsistencies within corporate IT and networking environments. Corporations didn't have the clout to force vendors to interoperate. The Net does. So I think that's a fundamental difference.
Is working with Apple a way to eliminate these inconsistencies?
I think the impact of the Apple-Microsoft announcement has been overblown. It definitely was good for Apple: It helps Apple have credibility. It ensures that high-value applications can run on the platform. But the Apple installed base is what matters, not new shipments. It will take a long time for new shipments to cause any significant change in market share in the Apple installed base.
Wasn't part of the announcement not only that they were going to make Explorer the default browser but also that they were going to start using the Microsoft version of Java?
I think that's how Microsoft would spin it. My opinion is that Microsoft's latest move in not shipping the Java Foundation Classes has really clarified its approach to Java.
And what is their approach?
First off they say Java is only a language, not an environment. Microsoft Windows is the environment. And the rest of us are saying no, it's a language plus an environment. What Microsoft has done with the Apple deal is unify the rest of us.
You make it sound like they've woken a sleeping giant, but isn't the giant Microsoft?
Yes, Microsoft's the giant! There's only one thing bigger than Microsoft, and that's the Internet, right?
Microsoft has not led technology development on the Internet. They are following us. Everyone just realized that Internet and enterprise middleware are the big game in town. And now here's the giant, who has the giant market share.
Not in NT. They don't have the giant network-based market share at the server, do they?
Aha! No, they don't. And I personally think operating systems are becoming less and less relevant.
Wait a minute. The reason they have less market share is because Unix has always provided a more robust, reliable operating system for network solutions. But now you're saying operating systems aren't that important.
I think the middleware layer of network services is what's really becoming important. I can't tell what operating system I'm on, so the differences in functionality between NT, Solaris, IBM AIX, and others--to me that's not what customers should be concentrating on. It should be, does this operating system provide the right platform for robust network services on top of it? Is it reliable? Can I manage it? Microsoft is competitive with NT in this regard; but their fundamental business model is throw out what you have and upgrade to our stuff.
And network systems are typically the most difficult to change.
So if they don't have installed base network systems, it's very difficult to make inroads there.
It takes a long time.
Do you see Sun as the largest installed base for network systems?
It's very heterogeneous. In the Unix space Sun definitely has the lion's share. But Novell has the lion's share of actual servers out there.
For actual Internet servers, it's Unix and Sun, no?
Right. Sun leads the Unix pack, and there's a lot of Unix out there. Our whole approach, in contrast to Microsoft's, is to make our network services easy to deploy.
What does that mean? It means, one, let's run it on what you currently have. An IT department has a budget of fixed size every year. At most I think they could rip out and replace 10 percent of their network. Microsoft's approach is to rip it out, put in new Intel hardware that people generally don't view as scalable enough yet (although every year it gets closer) put in the new version of NT, and then let all of these services run on NT only.
An unproven version of NT.
Unproven, not yet shipping. For instance, Viper transaction processing only works if your database is running on top of NT 5.0. If I've got an Oracle database running on a Sun server and I buy Microsoft's story, unless I move that Oracle database to NT, I'm out of luck.
Let's talk about distributed files. There's a key service for this sort of middleware--the ability to distribute your files, to replicate them, and then to manage consistency. I understand that NT 5.0 is going to have this service. They haven't had it previously. What is Netscape offering in this regard with enterprise Java or your transaction management system?
The key again is that it'll only be a distributed file system on NT systems. They don't embrace other systems. So as long as it's on NT, in a homogeneous NT world, it would work. They have a compelling story. They don't yet have the products--they don't even have a directory shipping yet.
It's very hard to do in a heterogeneous environment.
Exactly. And that's what we spend a lot of our effort doing. We have six platforms that we do, and our partners do another six or seven platforms. Our goal is to put the latest functionality we have on top of every environment and allow them to collaborate.
Managing consistency is a very difficult technical problem that I don't think has really been solved in 20 years of people talking about distributed databases. And yet it has to be solved for this sort of enterprise server concept to work. What are you guys doing on it?
Well one, it's clearly building services on top of an abstracted operating system model.
So this is a super operating system over and above the operating systems of the various platforms?
Or it's making sure your network services are operating-system independent.
But it is the operating system itself that has to be independent of the lower operating systems.
You've got to do transaction management, you're going to have to roll-back stuff on who knows what machines out there. It's a whole operating system.
But if you look at the Net today, does it matter what server is running a Web service?
I'm just saying it's a hard technical problem. What is your approach to solving this technical problem of managing consistency among distributed files and doing transaction management in general?
I think you'll see us partner with people in this area, for example, BDA Systems, IBM. There are companies with these services that run on top of a variety of platforms. Netscape can't develop everything ourselves. In fact, the way we compete against Microsoft is through viewing the world differently. We're very network-centric in our view. They are still desktop-centric and the network is an add-on. So we have a different paradigm. Also, through acquisitions and licensing, we try to build our products faster than they can through brute force. You know, they out-man us in development. But we've got a strategy of partners, licensing, and acquisitions to build our view of the world competitively against Microsoft.
It's interesting. I just asked you about technical solutions for some of these difficult distributed problems and you mentioned your partners, but you didn't mention Novell.
Novell is one of our partners. We've set up a joint venture called Novonyx. The interesting thing with Novell is that it's such a different platform from the rest. It's a unique platform. It has some great advantages. It also has some disadvantages in its architecture.
What we've done is create a team jointly owned by Novell and Netscape that will take our services and bring them to the network platform. And they've been quite successful. This fall we're going to announce Netscape Enterprise Server built on top of NetWare and Netscape Messaging Server built on top of NetWare. The performance they're giving us on this stuff is remarkable.
Tell us about that.
Think about NetWare. It's a great file server. Think about an HTTP Web server. It's fundamentally a file server with a different protocol. Now we have a lot of third-party code within our Web server. A lot of that doesn't exist from those third parties in the network space because it's a different operating system. The guys at Novell understand how to adapt that software to their platform. The independent software vendors don't. So with Novell doing it, we're seeing some really good performance out of our software running on top of a NetWare platform. It's not a general-purpose operating system; it's more of a specialized operating system. It's actually pretty good for network services. I think up until now the problem has been that Novell fell behind in what network services to build on top of this platform, and how open they should be.
Especially the openness. But now again, your response has been "We're gonna get a really super-fast Web server out of this collaboration with Novell." But I don't hear anything about directory services or transaction management or consistency management for replicated files. Where is that stuff coming from? From Novell?
From a technology point of view with Netscape? Not at the moment. We have our own directory server. On the one hand, we compete with Novell in two areas. One is in offering directory services. That's more of a "coop-etition." We both believe in LDAP. I really don't care whether Novell Directory Services is in the customer's environment or our directory service. The important thing is that every network service we build integrates with the directory.
Do you think Java's an issue here?
No, I think it's more that we've staked out a certain set of network services in SuiteSpot. What we intend to do over the next few years is broaden it to include transaction services, message queuing, file, and print. Those services we don't offer today. Right now that's what we're building with our partners.
But this gets back to Microsoft. If they're going to roll out NT 5.0. . . .
I think we'll beat 'em. Clearly, we're trying to beat them. I think the fullness of their platform as they describe it--not yet as they ship it--is what we need to compete against. So we need to broaden SuiteSpot. Take file and print services, for example, something that NT offers today. Customers require it. We don't have a solution. We'll work to add that solution to SuiteSpot. Look at the breadth of network services NT intends to offer. I think you can expect us to match them one for one in a different paradigm, one that's operating-system independent.
So your key is not feature for feature. You'll have whatever features are necessary. Your key is the ubiquity of the services.
Yeah, the breadth of our platform. There are three things we're trying to do really well. One, collaboration. Let people collaborate with other people on the Net. We do that with mail, we do that with Collabra, we do that with calendaring. That will always be a focus of ours: How do we help people collaborate better?
The other major areas are publishing information and applications. That's our Web. It's the enterprise server, it's content management, and it's applications.
Do you see offering special infrastructures for vertical markets--like banking versus publishing?
I think we'll look to partners to do that a little more. Our play is a horizontal play. But typically once you fill out the horizontal play, then you start to look at how you can optimize it in vertical markets. I don't think we're there yet. I think there are certain verticals that have adopted us to a greater extent than others, but we're looking to be a horizontal supplier of collaboration services and applications.
Let's go back to the three major things.
In publishing, the question is, How can I publish information broadly and still manage access? Our Netshare concept shows how individuals can now use a Web server to publish not only HTML, but any document up to the Web server, put access controls on it, have agents that run against it when it changes, do visual differences between versions. It's a very powerful paradigm. It scales. In other words, with X.509 certificates, as long as I have that Web server outside my firewall, I have access controls on it.
It works with any document. And then we have version control built into the Web server, so I can get an agent to send me a visual difference of two versions of a Word file whenever it changes and someone's published it into the Web server.
And it can notify people whenever something changes.
That's the publishing of information either broadly or to a select group with access controls.
The last piece is how do you manage and secure these solutions? That's where we're putting a lot of effort. Fundamentally, it's the directory. The administration technology we're building is something we now call Lava, which is a Java framework based on LDAP. Management of networks today tends to be focused on SNMP. It's low level--it's a monitoring protocol, right? You don't really manage a network with SNMP, you monitor it. We think with LDAP, the directory is the key management infrastructure. You basically make a change to the directory and the network adapts to your change.
That still does not get you transaction monitoring and consistency management and replicating distributed files.
No, it doesn't.
Those are also essential services. The first thing you have to have is structured services.
The directory service is the place. It's a network-based repository of, as I call it, loosely consistent data. You want to use databases for highly consistent data. They have their own replication schemes.
There is this issue of how do I replicate data around the network? It's a hard problem. It's not the first problem that needs to get solved, though. As more and more of our networks get connected, replication becomes less important. Replication's really important when you have disconnected networks. But we're wiring the world.
Wait a minute. This goes against what I thought was Netscape's other thrust-pushing push. The idea that what you want to do is reach out, grab, cache things you can look at later.
That's different. To me that's off-line use, or it's the model that just says, I like to watch TV.
Do you think that's real?
I think there are two approaches. There are people that want stuff to come to them rather than having to go out and find it. The Web can be pretty confusing. So pushing the information right in front of them where they see it, they click, they read.
So there're push people and pull people?
Right. There are both.
But luckily the push people are in a minority.
And luckily you don't have to subscribe to any of it.
I want to talk about intranets. Is this just IS people moving to use TCP/IP and Web technology as a ubiquitous platform--just like they moved to Unix, or just like they moved to Microsoft because having a homogeneous platform lowers their maintenance expense? There doesn't seem to be real technical innovation in intranets. It just seems to be a wholesale move toward. . .
. . . mass standardization, right. But there are a few things we could never do three years ago we can do today. Three years ago, first off, I don't think I had connectivity between every computer in my corporation. That's a big change, the standardization of TCP/IP.
Why couldn't that happen before with LANs?
The problem is LANs grew up with proprietary protocols that didn't talk to each other so there were these islands, sort of LAN nations.
What isolated them were the proprietary protocols, and then TCP/IP opened everybody up.
All of a sudden blew it wide open. I could build an application that reached every employee in my company, regardless of their desktop. I remember the first time I ever saw a Web application where I realized ahead of time no one had to touch my desktop.
That's the beauty.
And the vision is, again, sort of a virtual operating system that can run on many platforms to support the directory services and all the other services?
Right. Applications are built out of these network-based services and deployed on demand to desktops anywhere in the world.
So if the key to intranets being a phenomenon is the standardization through the ubiquity of the protocol and ubiquity of the apps, then your vision is the ubiquity of the enterprise services as middleware?
Yes. An intranet is interesting only to the people inside the company, so the real vision is, Can you take your IT infrastructure and open it up to your customers? Federal Express is the perfect example. It's the first to do that with their package tracking. It shows how they've taken an information system and made it of value to you, the customer.
But if everybody else has this same virtual network operating system, then it should be easy to export.
Exactly. Which is the other thing that drives "Why do you care about standards?" It used to be that people kind of cared about standards because they hoped there was more than one vendor who would compete and drive the price down. Now I think they care about standards because that's how you connect to the rest of the world, and the world isn't going to buy only your stuff. What's nice about the Net today is, if your process doesn't interoperate with others, then you're out of the game.