Q&A Session with
Bill Gates, Paul Maritz and Deborah Wellingham
Tuesday, May 20, 1997 
New York, NY

MR. MARITZ:So with that I'd like to invite any of you who might be interested to ask questions and Bill, Deborah and myself will endeavor to do our best to answer them.

Does anybody have a question that they would like to ask at this point? Yes, sir, I saw your hand go up first. There's a mike coming down to you, say who you are and what organization you're with and then ask the question.

MR. EVANS: Jeff Evans, Computer Paper Canada. Is there a place in this new scalable NT for thin clients, for network computers?

MR. MARITZ: Bill, do you want to take that one?

MR. GATES: The real question is what applications do you want to run? And we have a Windows family that runs all existing applications and the new applications being developed today. A month from now we'll have an event where we'll really hone in on the whole Windows terminal, which is the only thin client, it's the only client where you don't have an operating system or a browser that has to change. Like other thin client strategies it's moved work to the center, where you load it on to the server. But, in our case, running all those existing applications. And so the strategy we have is not one that needs everything to be rewritten. And it's one that lets you scale, so that if you have a PC with local computation, we take advantage of that. We still give you the portability, but for people who have very simple demands, we give you a cheaper, more stable client, the only true thin client that there is.

MR. FRANCIS: Bill, Bruce Francis from CNBC, when do you think that we can expect to see significant penetration of NT into the traditional heart of UNIX, financial services, banking? When would you condition us to expect that?

MR. GATES: Well, the big switch in terms of people thinking about NT took place last year. And that's where on the desktops for the first time UNIX sales were actually down, while NT sales skyrocketed there. And there's lot of places where people would have bought UNIX servers in the past that they bought NT servers. So that shift is already taking place. And all we're signifying today is that through the eight processor level there is no tradeoff at all. In getting the benefits of the PC, the similarity to the desktop, you get all of those things. Over the next year you'll see even that top five percent that I talked about, some of those design wins will start to go in the NT direction. But, the shift from UNIX, you already see those symptoms everywhere. And we have lots of cases of customer examples where it was an actually shift.

You know, a few years ago if you said to people how will Boeing or Lockheed or General Electric run their mail systems, they would have said, hey, we'd have to use non-PC technology. Well, in fact all those customers today run their mail systems using Windows-NT.

MR. FRANCIS: But, what about north of eight processors and into that very ironclad UNIX base of financial transactions and banking, when do you think you can make some significant penetration there?

MR. MARITZ: I think that's starting to happen now. As Deborah mentioned, I think correctly, Dayton Hudson is running up to 16 processor systems a day, 10 just demonstrated last week. They're non-stop SQL environment running on a 64 cluster system. So while those that are still making their way into the market, clearly the trend is there.

MR. GATES: Yes, Tandem has been the leader in terms of on-line transaction processing. And as you'll see in the partner pavilion, and we've talked about they are demonstrating how they're taking their very high end capabilities and building that around Windows-NT.

MR. CUTTLER: Thanks. Alex Cuttler, DLJ, a couple of questions. I noticed some veiled product announcements. Can you talk about the BackOffice for small business, any plans to reduce the footprint for SQL Server. The other question has to do with Sphinx. You showed us a live demo of the product, how long do you think the beta cycle will be for Sphinx or SQL Server 7. And finally, Larry Ellison feels that you're vulnerable at the low end. And while you bought Web TV, they acquired Navio, can you respond to that, Bill?

MR. MARITZ: Let me take the first part and then I'll let you pick up the Navio part of the question.

MR. GATES: Okay.

MR. MARITZ: The first question had to do with the product announcements, small business edition of BackOffice. We've worked very hard to reduce simplicity -- we think actually the footprint of NT and SQL Server are well within what's acceptable for a relatively low cost hardware environment with small business. You can -- $3000 or $4000 will buy you more than adequate hardware to run, in terms of memory and disk to run the complete BackOffice small business edition environment. We are working on SQL Server to make sure, that as I said, with all our infrastructure, to make sure that it runs client and server. So we are testing the next version of SQL Server, which we code named Sphinx, which you referred to, to make sure that that will run on its server and on a client environment, in Windows-NT workstation, Windows 95 client environment, so that that engine will be able to run both places. So we're working in terms of making sure that that technology is broadly applicable. We don't know the exact chip data: the Sphinx technology has gone into very early tests and we'll make announcements as we hit a milestone on the beta cycle.

With that I'll just turn it back to Bill, which the question, in case you don't remember, was Larry Ellison thinks that we are vulnerable on the very low end, vulnerable I guess specifically to set top boxes and what are your thoughts on that?

MR. GATES: Well, Microsoft has always believed in a variety of digital devices being connected together, whether it's hand-held type devices or an intelligent TV. Relative to the size of the PC market, those markets have been very, very small, but they have been growing. With our work with Web TV, we expect that intelligent TV piece will become significant. With the work we've done with the hand-held PC and some extensions we do, we expect that will become a good business. And all of that is based on Windows CE, the low end of the Windows family.

I don't know particularly what portion or what was paid between Navio and Oracle, or how they're working together. There is some type of announcement that relates to that. Oracle is a competitor of ours in that space and we think it's an interesting space and you're going to see a lot of people go after it as it emerges as just another way of getting at information.

Amy Wold, Track Putter. This was a really exciting demo to see a billion transactions per day. How soon do you think we'll see a commercial user using a system in that range, and are all the pieces ready or ready soon for doing the rest of the applications beyond the operating system itself?

MR. MARITZ: Do you want to take that up, Bill?

MR. GATES: Yes. No one needs a billion transactions a day. With the Internet, there will be people who want a billion transactions a day, but there's nobody who's got that type of situation right now. The biggest Web servers would be up in the 100 million hits per day, and hits are trivial compared to transactions. What we did there was a lot more than just processing a Web server hit. We were able to do almost 100 million off of the single server.

Everything you saw there is shipping software. That's a distinction. The single node terabytes did depend on the next release of SQL, the Sphinx release. Whereas, what we show with the billion transactions a day, every piece there was absolutely shipping software. And so you could take, if you wanted to run any commercial applications, and run that type of transaction mode, you can do it now. The magic of transaction server is what that's all about. I mean, this was a lot of work to put the disks together and to get the router and the servers all geared up, but the thing that makes it work is that once you have a great transaction server that we've now gotten built into the operating system, you can do distributed applications. So there's nothing that prevents people from having the full benefit of that right away.

I'm E.C. Manners from Gartner Group. My question is, at the time when Windows-NT reaches maturity, do you expect that mainframe systems will still have some place? Is there any category of enterprise systems that you're not targeting with Windows-NT?

MR. GATES: Well, there's a lot of existing applications that have been written to MVS and other mainframe subsystems, and people are going to continue to run those applications for a long, long time. And so that compatibility with old code is a very important thing. And that's why mainframes will continue to be in use.

You're not going to run into customers, at least I haven't run into any, who said, hey, I just bought my first mainframe yesterday. You know, that's not going to happen. See, anybody who is building up a business today with new applications is going to build them up around this architecture. There's absolutely no reason in terms of richness, performance, and all that, that they would do it that way. So we take interoperability with the mainframes very seriously. And because in those applications, some people put more loads on them, people will continue to upgrade their mainframes. But there's nothing new that should go down that path.

MR. MARITZ: Go ahead.

Windows-NT then will be able to carry all of the applications that today might still go on MVS or TPF or high-end processors?

MR. GATES: Yes. As we showed, a billion transactions a day is more than TPF is handling in any place. I mean, it's not even -- a billion is way beyond what people are needing in those applications. And so there's no scale reason you would stick with the other architecture. The existing code would be the sole motivator.

Walter, from Paine Webber. To what degree is Microsoft running Microsoft on its own products today?

MR. MARITZ: I think, with a few exceptions, we're almost completed self-hosted on Windows-NT and SQL Server.

Do you want to elucidate on that?

MR. GATES: Yes. All the business systems are Windows- NT based. Historically, we had a VAX which we've transitioned off. I think we still have one application there. And then we used to have some AS-400s, and the majority of those have been shut down. I think there's still two running that run for another few months. But we run SAP/R-3 on a heck of a lot of servers from Compaq and others.

MR. MARITZ: We have some very major applications. I think we support about 1,500 users on SAPR-3 routine active day-to-day users.

MS. WILLINGHAM: Yes, exactly.

MR. MARITZ: Running off a pair of NT SQL server machines. And then we have the --
MS. WILLINGHAM: And in customer service applications, we run up to 5,000 people on a single SQL server for our customer service aps.

MR. MARITZ: Every time you call Microsoft that call gets logged into a SQL server and that is the track to call through. And we sustain 5,000 concurrent users off that system on a routine basis.

Yes. We'll come down to the front of the room down here, the gentleman down here on my left. And then we'll go back to the next gentleman there to the right.

Eric Wilson from the Morning Herald and the -- (inaudible) -- Financial Review. How difficult is it going to be to write in-house cross platform applications? For instance, if I have something that was once written in JAVA, or is being written in JAVA for a AS-400, how hard is it going to be to migrate that to an NT cluster?

MR. MARITZ: The issue, as Bill was saying earlier, what we are trying to do is to provide -- the essence of really making an application work across the cluster systems is to be able to put it under transaction control and, if necessary, partition the database. And our transaction server makes it easy to write an object which can then be essentially structured in such a way that it can be automatically put under transaction control. And our strategy is to provide complete interoperability between objects written in JAVA and objects written in any other language, whether it be Visual Basic, or Visual C. So our whole environment is open to objects that are written in languages -- in any of the important languages. So, unlike other approaches, we don't require you to use a particular language. We don't discriminate against you if you have used a particular language. So you will be able to structure your objects correctly, and obey a few common sense rules, and you'll be able to get a lot of the advantage of this environment in a pretty straightforward way.

Robert Craig with the Horowitz Group (sp). We've got some really great (inaudible) there, and I'm really looking forward to getting together with Jim Gray and learning more about (inaudible) transactions that we have. But I'm wondering, when are we going to see -- and you said earlier that you've got the price/performance numbers in terms of the transaction processing benchmarks. When are we going to see a Microsoft benchmark as the performance leader? Right now, the latest benchmark I saw you were at the 25th percentile.

MR. MARITZ: Well, we've been pushing steadily up the levels to the point where we're -- one of the things about the TPC benchmark that you can't -- the rules are, you can't claim them until they've been audited, and we have some very interesting benchmarks that are being audited right now that are going to put us into that top tier. So you'll just have to wait until those results come out.

When we get into the clustered environment is where you really start to take off and see the serious advantages of having commodity hardware being able to just be slotted for the environment, and see this really reach the very high levels. So we're in very respectable levels now, and they are going higher all the time.

Paul, Bill, or Deborah, John Dodge, PC Week, and ZDNet News. You made a lot of comparisons between NT and UNIX this morning saying NT is eating into UNIX's business in what I presume to be old applications and new applications. You made no, or maybe I missed it, references to NT actually eating into mainframes. I did see one slide that mentioned VMS and MVS, but I didn't really hear much about NT taking over mainframe applications. Can you talk about that?

MR. GATES: The first thing to understand is, most what people are doing are new applications. There are certainly cases where they're unplugging an old system to put the new system in. But the reason this is exciting is because there are so many of these new applications where people want to take all the sales data in their company and make it easily accessible, or they want to have a Web server that has very high volume, and they can do rich analysis of that data.

We're mostly going after those new markets. In doing that, you get performance levels that allow you to see a substitution effect as well, but the volumes of the old markets is not enough that that's really the big thing.

MR. MARITZ: We can actually give one example. The Bridgestone Company in Japan is the largest tire and rubber company in the world who are on a path to literally unplug all their mainframes. They're going to be running their business, end-to-end, off an NT server environment.

Craig Scott, with Scott Consulting Corporation and ENT Magazine. And the unstated assumptions behind that billion transaction a day demo is that there are multiple servers operating concurrently read/write against a shared database. And I guess I have two questions. One, is that assumption true and, if so, how are you dealing with locking?

MR. MARITZ: It's actually a partitioned database. The database is partitioned across the servers as the requests go in, they get distributed. You can go and find out the exact detail after lunch, but basically the transaction gets routed through a server. There is logic that runs that then decides which database server should handle that particular transaction. The transaction then gets routed from database server to database server until it hits the right spot to actually then go against the database of that server. So the whole trick is that that locking is done at a higher level, not down at the actual file or sector level on the disk. So, if you go upstairs, you can get a more detailed presentation on that.

We're out of question time here. Let's take two more questions. One here on the right here, and then we'll come back and take one on the left.

Gary Beech with Computer World and IDG. I have a question for Deborah. Deborah, you mentioned the Premier Technical Service. Could you tell me about the pricing model for that?

MS. WILLINGHAM: Well, the pricing model for Premier just like all of our services is a cost recovery model. The base price is $45,000 annually, and then we will also put together a very customized and flexible deal for our customers, depending on what their support needs are. So, a $45,000 base entry price.

MR. MARITZ: It looks like we can probably take a couple more questions. We'll take one on the left here. Yes, sir, go ahead. Bring the mike down to somebody on the right here, and we'll take the next one. Go ahead.

Hi. I'm Tony from Dow Jones News. How much money have you invested in this scalability initiative? You said it started eight months ago, and how much do you plan to continue spending?

MR. MARITZ: I'll give you one metric, because it's hard. We haven't actually kept a running total over the years. But this fiscal year that we're in right now, if you look at our investment in system software, all of which in one form or another contributes to this effort, we're investing about a billion directly in system software.

MR. GATES: And then eight months ago is when we made this a huge priority. I mean, we've been doing things to make these systems more and more scalable, but it rose to the top along with manageability about eight months ago.

Hi, thanks, Drew Brosham Accounting Company. The trade press and the market research folks seem to be pointing beyond scalability to things like throughput and availability, reliability, and manageability, as you mentioned, as the real keys to making this really go up to the high end application systems. How do you respond to that?

MR. MARITZ: I think we would respond that we agree with that. I mean, as we said earlier, scalability is not just a measure of one single thing. You've got to look at capacity in terms of transactions, capacity in terms of data, number of users. You've also go to look at how do you enable whatever the particular solution is to be not only assembled and running, but then managed thereafter. So the work that we're doing to build manageability into the platform, as well as the partnerships that we have with other management tool vendors, and the service providers, is a critical aspect of this. Hopefully one of the messages that you got today is that we are very serious about that.

One of the great things about getting to higher volume is that you get a lot of third party support. So one of the interesting phenomena that's happened over the last 18 months -- and Windows-NT server sales really kicked upwards in the second half of calendar year 1995, that's when you really started to get the distribution channel working with us as well. Initially, actually, Windows-NT server succeeded in the high-end corporate environment. IT took some time for that VAR channel, the reseller channel, to really kick in, but then the volumes really started to take off about 18 months ago.

With that, if you start getting a lot of third party ISV and tool support, so basically all of the tool vendors who currently sell tools into the UNIX environment, for instance, to help you manage UNIX servers, are seeing a huge market open up with it. So they're moving as fast as they can to move all of those value added tools over to the NT environment. So I think you're going to see a very complete and robust solution evolve there over time.

That being said, it's something that you work at all the time. It's going to take us some time to get to the same level of lights out operation that you can get out of an MVS server today. That's got 20 years of history behind it in terms of the event handling mechanisms. The failure analysis tools, et cetera. But the trend is there. The feedback cycle is working, and the one thing we know about this industry, that when you've got the technical foundations in place and you put the market conditions in place, then the feedback cycle works. And that's what we really are declaring today.

Ken Decks with PC Week Labs. I'd like to ask a question about your future scalability direction. We test a lot of SMP servers at PC Week Labs, and we've seen overall poor scalability at (inaudible) levels, and it gets even works at the (inaudible) levels. I'm wondering if you will continue to invest in SMP systems in terms of moving more to (inaudible) thread, and continue to try to build that out? UNIX goes very well with 32 CPUs on a single box. Or will go you towards clustering, and move away from SMP and like shared (inaudible)? What's your feeling about where the real future for NT scalability is really like?

MR. MARITZ: Both. Bill, go ahead.

MR. GATES: Well, nobody actually scales well above about 16 or 20 processors. In terms of general SQL capability, that's about where anybody starts to get actually diminishing return. We've moved up over the years. We went so that we got great scaling on dual processor, then we got great scaling on four processor. What we're seeing today is, we're getting very good numbers on eight and ten processor systems. And that's required a lot of work, a lot of focus by us. And we'll keep pushing that forward. So there's still more benefit to be had there.

But when you want to go to very large, guaranteed reliability systems, the clustering technique will absolutely be a key element to that. So you get faster chips, more processors, better software, and then clustering sitting on top of it.

MR. MARITZ: The problem is, that once you get above the 8/16 level, all the processors are contending to access to memory, and to get the memory bandwidth to keep modern processors, Pentium Pro and Alpha class processors busy in an SMP environment, you've got to go to some very, very expensive and exotic bus technologies. So what happens is, you find over the 8 and 16 level, although you could continue to scale, the price/performance drops off dramatically. And in order to continue to get scaling, it becomes extremely expensive. It works with the clustered approach where you really can put together standard, off- the-shelf components and keep scaling that way. So we believe both are complementary, but in the long-run, the very demanding applications will be done with the clustered approach. And you see that in the real world today.
Okay. With that, I'd like to thank you all for coming today. I'd like to invite you across to our partner pavilion, pickup the materials there and see the very interesting exhibits that are available there.

Thank you very much. (Applause.)