Foster returns to our Q&A for his perspective on the market changes since 1996. He’s also news because his firm is piloting a new version of its flagship DataExpress, one that links queries from the 3000 with data on NT and Unix boxes — a “multi-platform join.” He’s also chaired the largest group of 3000 vendors, SIGSOFTVEND, for more than a decade. We think this makes him a great touchstone on how software vendors do business, a crucial topic in the 3000 market’s future. Three years ago we asked Foster to tell us how the 3000 could survive. This month we ask him what it will take to make it thrive: handing its relationships with other platforms, especially NT, and how HP might attract application developers to a platform before its “Year 2000 tunnel” of safety disappears. Schooled in economics, he can sum up business strategies like “Economic Darwinism” that help put the platform outside that tunnel and in the richest possible light.

What impact do you expect multi-platform joins to have on the HP 3000 marketplace?

It’s a reality that virtually every site has either an HP 9000, an NT box, or both in the HP 3000 marketplace. As people move NT from its role of file sharing or print sharing into a role where it might have a departmental or small application, people want to join the data on their HP 3000s with the data on their NT box to produce a result. Multi-platform joins are a reality that needs to happen, one way or another. Either people are already shipping a file, or the join is being done manually. People are going to want to do this more auto-magically.

NT’s success opened a door for the HP 3000 renaissance. How has NT’s suitability as a mission-critical platform changed in the past few years?

It not only has not scaled, but it still needs to get the robustness that people are so used to on the HP 3000. The issue comes down to the Microsoft promise that something would be enterprise-ready by May, 1998. That never delivered. In fact, the version that is scheduled for the end of 1999 doesn’t even contain all the features that were talked about in May of 1998.

And the things being left out of NT look pretty mission critical?

Some of the items, yes. If you had a feature that was in there to promote the whole platform, it was there because it made it enterprise complete. If you’re dropping it out, are you less than enterprise-complete? I would say probably. The fact that you had it in the list of features and you’re now removing it didn’t change the fact that it was enterprise-necessary.

The sidebar on this is the 3M story. 3M has changed the NT deployment strategy and has backed its [NT] projects out — and they’re now committed to building them on HP 3000s. This is a three-year [HP 3000] commitment, starting in December of 1998. They did this because NT doesn’t scale. You have sites that run regularly with 2,000-plus users on an HP 3000, and you can’t get an NT box to do that. One day, if Microsoft fulfills its promises, it will do that. But today it won’t.

What conditions do you think might lead some HP 3000 customers to move their mission-critical applications to NT after the Year 2000?

When NT doesn’t have to be rebooted when you install new software. Typically, installing a piece of software on an NT box brings new Dynamic Linked Libraries (DLLs). The way that Microsoft has built the operating system and handles DLLs is such that the majority of the software will require a reboot at the end of that process. That’s unacceptable if you’re running a mission-critical application and you need to be 24x7.
Once they get to that point, it will get more robust and won’t fail as often. People don’t just move systems for the sake of moving systems. There has to be a business reason for doing it, and it has to be a compelling business reason that gives it a priority.

Why do you think 3M wanted to move their 3000 systems?

They were hoping that the promise of a lower-cost computing platform would arrive. You have to admit it’s a lot easier to actually say, “Well, that’s what all the magazines said to do.” Nobody has gotten fired for buying NT yet. Here’s 4,000 pounds of paper that said this was the right thing to do.

Do you think HP 3000 application availability is a condition that would lead customers to move mission-critical applications?

You have to have the equivalent or better application on the other side to move. The reason people have computers is to run applications. Not just the print and file sharing applications that NT was famous for in the days Microsoft was bragging its growth rate was 60 percent. That was wonderful to talk about, but when you’re growing from zero, of course the percentage is high.

What are the essential elements of a data access application in the HP 3000 marketplace?

First of all, you need to be aware of all of the data types you might run into. Whatever your data is stuffed away in, you need to be able to access it on your desktop. That’s the formula we use in DataExpress: all the data to all the desktops.

The other element is that you don’t just have data access. That’s like asking for running water — I’ll run a pipe into your living room and you’ll start complaining that your carpet’s getting wet. What you really wanted was water, but with controlled access, a tap. People asking for open data access through a box have asked the wrong question. Unless they have a methodology in place for control, they’re probably going to get a lot more data than they want.

Do you really want all 6 million records in the history file to arrive at your desktop? That’s not a manageable chunk for most people. You want a working subset, so you can select on that data. You’ve got to have the aspects of simplicity and security covered. Data access is middleware, and will evolve to message-based middleware over the next five years.

So what makes the middleware message-based?

You’ve got to enable things to ask a question, like object request brokers. These are going to be the place to go, because there’s no way people can define every interface in the world. Each piece of software can’t have its own interface. That’s why there’s publish and subscribe methodologies, why all this stuff works with some form of messaging.
It would mean that any client could ask a message of any server. And the request broker will actually broker that request, gather up the data and hand it back.

So far you’ve got all data to all desks, and plumbing instead of just data access. What else is essential for 3000 data access?

The issue with business-critical systems is that often they were written for programmers by programmers. End-user direct access was not contemplated. The names might be downright user-hostile. For a user to be able to access a piece of data and understand what it means from the name might not be easy. You have to have a way to build a view that allows you to do a couple of different things. One, the view should allow you to rename data fields so they can become something that somebody could use. Another would be a date format, so the user could understand what that date was. That’s especially true of the proprietary manufacturing packages on the HP 3000.

Almost every application has its quirks in the way it stores data. Then users actually have to get involved in figuring it out. If a user has a state code like TX they know what it means, only because they use it all the time. If you give them an order status code of H, what does it mean to you? You have to something to translate that in the context of the workflow this data supports.

Another item is that it’s often not just one table you’re looking at. One of the hardest things for end-users to do is to link between databases. The end users could link between the tables, but not have to do it themselves. Somebody from IT sets it up, but then they can get access to their data based on the security and simplicity components put in place on their behalf.

What about support for things like Java and non-3000 databases? Do you think those are essential to 3000 data access?

I would consider those today to be optional. We have a number of customers who are looking heavily at Java. To date, when asked “What’s the JDBC application?” they are unable to articulate any commercially available Java application. That’s an issue.

So JDBC middleware is quite a bit out in front of those commercial applications? You don’t offer JDBC today.

That’s correct. We have a number of customers who are looking at rolling their own Java applications on a Sun box talking to a 3000, and we are considering requests for assisting them with something that would pass a message to our DataExpress product on the 3000 and return an answer.

Why prioritize connecting to non-3000 databases above Java access?

There’s been a lot of small data warehouses built on NT platforms in 3000 shops. They are already deployed, like a data warehouse application or an inventory application that was built under Windows because that’s what the magazines said to buy. This application needs access to data on the 3000.

Do you expect when you get multi-platform joins out in the 3000 marketplace you’ll have a unique position for awhile?

I think we will have advanced the technology sufficiently to raise the bar. When HP was looking for JDBC vendors, we did look at what it would take to do it. We’ve been looking at JDBC for quite some time. Given our lab resources and what our customers were asking for, we decided to deliver [multi-platform joins] first. The timing was right for something that would allow NT boxes to talk to 3000s, where the database on the NT server would do an optimized join back to 3000, to pull all the data together.

Stand-alone ODBC utilities are more common in the 3000 market than the full-featured package you sell. Why not split out the essence of ODBCLink and sell it separately — at least for the smaller HP 3000 sites who have less to spend on the 3000 and need native IMAGE access?

First, ODBCLink uses technology that was developed in DataExpress to do its job. Therefore, that is the way we get leverage on what our lab people do.

Second, from an economics point of view, our software costs less than the Microsoft Developer’s Kit, and people are willing to pay for the Microsoft kit on a small system. We made a conscious decision to bundle it, because we needed a complete solution. We knew customers needed the ability to read 6 million records but didn’t want to do it with 5 hours of hourglass time. We also made a conscious decision that if the customer wasn’t willing to make enough of an investment in this technology, then they probably weren’t going to make enough of an investment in the education. It could be called Economic Darwinism. We know that if a customer’s in that range, we can support them and give them value for money, no matter what size platform they’re on. The issue is not the price, it’s the value.

Does HP need to offer a version of ODBCLInk/SE that doesn’t need Allbase/SQL — and could you supply it? This is a user request that hasn’t died off, even after years of HP saying it wouldn’t happen.

In the long run, people need to look at how they’d be using [the Allbase/SQL] stored procedures and triggers anyway. Typically, that would be a better way to manage the data access in both read and write, to various data source. In the long run, I believe people will realize the value of SQL. In the short run, people don’t know how SQL works, so they want that layer out of it.
Could we supply it? Yes, for the appropriate compensation, anybody can build anything. We do know something about accessing data.

But would you want to do it? Has the existing bundling deal been good for your company?

This deal has been good for our company — absolutely. It’s also been good for the customer base, and it’s been good for HP.

You support Oracle access through Data Express. What needs to happen to make Oracle take off better than it has in the HP 3000 market?

Customers need to say that it’s what they need. I’m dealing with a customer who’s using Ingres on the HP 3000. They didn’t even know Oracle was available on the HP 3000. They’re looking to get out of Ingres. If nobody’s buying [Oracle], you can’t afford to develop it. In the real world, you can no longer assume “If you build it, they will come.”

I’m sure HP didn’t actively invest in having Oracle put on the 3000 without having customers say, “Our main database that we want to live on for our corporate standard is Oracle, and you need to be there.” But the trouble is when there’s no skin in the game, people can change their mind. If the customers had said, “Here’s my purchase order, or my 25 percent down, I want to see Oracle [on the 3000],” it would have been a different show.

You talk to customers who use Oracle in all kinds of environments. Do you have a theory about why they tend to deploy Oracle in environments other than their 3000s?

Because the magazines said so. There was a time when the Oracle story was their platform of choice was the HP 9000. There was a practical reason for doing that. It’s still true today: the HP 9000 represents the number one choice for multiuser business Unix boxes. If you’re a company that’s going to run a business application with multiple users on it, you probably have a 9000 server. It made sense for Oracle to be doing [development] there first, because they’ve got more people to select from when they rolled it out, and they’d get better return on their investment quicker.

Why do EDI on an HP 3000 when there appears to be so more momentum to put the function on a Unix or NT system?

You’d end up being able to integrate it back into your 3000 application, so the job controls would let you put it away and not worry about transferring files all over the place, and worrying about the differences. For something that’s a multi-step process, it’s easier to control it on one machine than on multiple machines.

We’ve also seen people that want the reliability of a 3000. While we actually offer the functionality of a 3000, 9000 or NT for our EDI, people appreciate the functionality that comes with the EDI package sitting on a 3000.

Is your EDI application something that could leverage a new HP 3000 installation?

I know that our EDI application has moved HP 3000s from time to time. People only get those when they already own 3000s in general, because they don’t want to learn how to handle one more box. Just the other day I was talking to a customer that was looking at things, and they were so happy when their CAD manufacturer came out with an NT version of their application. They had the opportunity to reduce the number of platforms supported to two — because NT was there already, and the 3000s were not going away. They eliminated their Unix boxes.

That sounds like customers don’t want to support any more computer environments than absolutely necessary. What do you think of CSY’s philosophy that its okay if everything isn’t available on an HP 3000, so long as HP offers it on another platform?

Customers in general have to look at what the staffing requirements are to perform their business. You can’t put a new platform in for each application. In fact, people don’t even want to handle multiple servers.

Utility and data access suppliers like yourself are the best-known vendors to the 3000 installed base. But even with a lot of knowledge of the potential customer, can companies like yours help drive new HP 3000 sales without applications leading the way?

There are some cases where this does make sense. For instance, even in the case of a database utility, you could have a 24x7 shop where [a new 3000] is where they could go and experiment with things for test or development. But the application has to be there in the first place. Who’s gonna buy a machine just to run Adager or DBGeneral?
You can put add-on boxes around an existing application for test, development, reporting and EDI purposes. But the customer has to be convinced that the HP 3000 is right for them and that it’s a reliable platform. HP has to make sure it gets that message out there a lot.

So HP can visit your company about developing new sales opportunities for HP 3000s — but you don’t think there’s much a middleware supplier can do until there is an application out there they could serve?

That’s correct. One kind of 3000 customer is are going to roll their own application, and they need an infrastructure. HP needs to recognize that it’s in the find-sell-keep game for customers and developers. You have to find the customer, sell the customer and then keep the customer. If you want to do revenue maximization, you have to be handling all three of those things.

Which ones do you think HP isn’t doing as well as it should for the 3000?

All three. The unfortunate truth on this is that we’re at the point now where HP is going after two marketplaces. The end-user marketplace, but we both know they haven’t had an active campaign to attract end users. On the keep side they’ve been doing some of that. But when was the last time you saw a campaign, other than in very specific marketplaces, for HP to bring in customers? This is why I’m glad to see Kriss Rant aiming at beyond the Top Ten 3000 vendors.
Do you think there’s any merit in trying to make a business case for Progress to get its 4GL/database onto MPE, as it was doing a few years back before it called off the project?

That certainly would have given you 2,300 applications to work from, wouldn’t it? For a developer, he’s got to have infrastructure including the databases they might want to use, the networking stuff, compilers and languages. The only people who care about this are developers, typically very big companies. If the 3000 isn’t on a technological par, it’s in trouble. The trouble is that CSY has added those things in the last five years. And they haven’t done a great job of telling the existing vendors — who have or once had packages on the HP 3000 — those things are now there.

One of the things someone at HP might try looking at is visiting with the 3000 vendors who sell on the 3000 to make sure they know what’s there now. If an end-user who owns 90 machines doesn’t know there’s an IA-64 strategy going on, then how do you think the vendor who’s working on more platforms than a 3000 and has less than 90 customers got this communication about this stuff?

There are opportunities. The HP 3000 is a rock-solid box. And if somebody wanted to be able to do a business-critical, worry-free application, the 3000 would be a good choice for a server.

CSY has been saying it’s an economic, business decision that leaves the 3000 out of application developers’ plans. Does this view — that it’s tough to get a company to commit to a smaller market in addition to current platforms — make economic sense to you?

You have to look at what’s the incremental cost versus the potential incremental revenue. The incremental cost to port to the 3000 for somebody who’s supporting multiple platforms is much less than it was. The reason for that is the 3000 used to be missing tons of things that would make it easy to port in. CSY ran a great porting center [when PA-RISC began], and they brought a bunch of things across to the 3000. Steve Capple of the Technology Access Center ran a great crew. He was there to help the vendors and help them make it happen.