Linux Vs. NT Public Forum
Linux Vs. NT
Shootout - June 16, 1998
Thereafter, the panel became a lively, interactive discussion between the panelists and audience, most of whom were true Linux users. The hour and fifteen minute session saw much heated debate between attendees on the floor questioning panelist Daniel Small from Microsoft. Often these interchanges were funny as well as contentious. For an audio recording of the panel click here [ http://www.isdmag.com/dac/linuxntaudio.html ] (requires RealPlayer). For a written transcript of the proceedings click here [ http://www.isdmag.com/dac/linuxtext.html ].
Linux vs. NT Shootout Transcript
Jonah McLeod : A few more people will probably come straggling in as we get going, but we’re about five minutes into it. So let’s start off our panel this morning.
So we engaged Seva Technologies to act as our consulting tester, and James Lee performed this test for us. He’s going to talk a little bit about the two that he’s run so far. We’ve had a benchmark on simulation with Verilog XL from Cadence, and just recently we did a benchmark on Synopsis synthesis for NT, which James will talk a little bit about.
As a result of those benchmarks, we started getting some e-mail from folks about the fact that although NT was beginning to show promise as an operating system for the EDA environment, there was quite a bit of concern among the user base that NT wasn’t the appropriate operating system, considering its roots in data processing, and not really being built up around the needs of the EDA environment.
As a result of this e-mail we started engaging more with some of the users -- just asking them what problems or concerns they had -- and they began to voice them. We got so many of those that we decided to write up their responses. In the current [July] issue of the magazine, you’ll see the more significant responses that came through, either critical or in support of NT. As you might imagine, they were critical of NT and proposed an alternative approach on the PC platform. The alternative approach that everybody seemed to be favoring was the Linux operating system. I knew a little bit about Linux, but over the course of the last couple of months I’ve come to learn a lot more about it. What surprises me is the amount of fervor that exists on the Net in support of this operating system. Consider the fact that it’s free and it’s remarkably real to us. Everybody is putting effort into keeping it up-to-date and capable, and doing it on a volunteer basis or for very little money.
Today what we’re going to do is open up the discussion about operating systems for the EDA platform. We’ll start with James, talking a little bit about his experience with NT.
Larry Augustin is president of VA Research, and Larry has begun to promote the idea of EDA platforms on Linux. So he’s going to talk a little bit about the Linux operating system and give you some insight about that.
We also have some representatives here from the other folks that have OSs on the PC platform. Of course, Daniel Small from Microsoft is going to be here to talk about NT. If you look on the floor [at the 1998 Design Automation Conference, in San Francisco], there’s plenty of NT systems out there running EDA applications, so it’s a reality: There are robust EDA applications on NT.
Dan Page from Avanti is going to talk a little bit about the EDA vendors’ view of NT over an alternative operating system, such as Solaris or Linux.
And then Venk Shukla, also an EDA vendor -- vice president of marketing of Ambit -- he’s going to talk a little bit about why his company is backing NT, and he has some interesting thoughts about his own users and his own company running [tools] on Linux. I’ll let him share that with you.
Finally, we have someone from Sun, the Solaris camp, Dan Souder . He’ll talk a little bit about Solaris. One of the questions that I most came to find out is -- you know, given the length of time that Solaris on X86 platforms has been around -- what is the status of the operating system currently.
And then, finally, we have Phil Tomson with FreeHDL, who is also interested in the platform wars and what operating system is appropriate for this EDA space.
So with that prelude, let me give it to James now and have him discuss the benchmark results. We’ll have another short presentation with Larry, and then I’ll start opening up [the session] to some questions from the floor, as well as some that I’ll have to start off with.
James Lee : First of all, I want to thank Jonah and ISD magazine for the opportunity to do these benchmarks. Having been in CAE for about 12 years myself, I think I’ve learned how to be an EDA skeptic. I don’t believe anything any vendor tells me until proven otherwise. When Jonah gave us the opportunity at Seva Technologies to do these benchmarks for them, we started out small, and eventually we said, "Well, if we can get a million-gate simulation to run on NT, that’s pretty impressive." What I’d like to do is start off by showing the results of a real quick summary of the results that came out in the March issue of ISD magazine, although I’ve added the new 400-MHz Pentium II machines running NT to those results.
Now, there are several vendors I’ve averaged together to get these, but it gives you a basic idea. At the top, you see the red line at 100 percent because pretty much the entire time the 400-MHz NT machines were faster. The yellow line that’s at about the 75 percentile means the performance of the 300-MHz Windows NT machines, and these are across a number of different test cases. The green line, which is sometimes faster and sometimes slower than the 300-MHz NT machines, is a Sun Ultra 60 300-MHz machine.
For more detail on what these test cases represent, you could see the March [issue of] ISD magazine. Basically the first one is a very small simulation, but it dumps over 100 Mbytes of waveform data, and it turns out to be an I/O performance test. As for the 400-MHz machines, if they hadn’t had the advantage of hardware RAID controllers, the Sun I/O performance would probably have surpassed them. But the hardware RAID controllers gave them a little bit of an edge.
Same thing with the 1.3 million-gate simulation. That thing was an 800-Mbyte virtual memory image. All of these machines, by the way, had half a gigabyte of RAM on them. So these are all fairly industrial-strength workstations. I guess we have to call them NT workstations and not PCs. So that gives you a rough idea of what we were able to find. And again, the March issue of ISD magazine would show you more about this. I think the big surprise was that we were able to run such large test cases and saw very promising results.
The July issue of ISD magazine, which is out -- what, yesterday, Jonah? It was on the stands yesterday, so you can either find it at the ISD booth here on the floor, read it on line, or it’s probably waiting for you when you get back to your office.
Basically, what we ran next was the Synopsys design compiler across a number of different designs. Again, the red line represents the 400-MHz NT machines, which were always the fastest. As for the 300-MHz NT machines in the yellow line, instead of the performance of the 300-MHz NT’s and the Ultra 60’s being about the same, this time we saw that somehow Synopsys makes better use of the Intel architecture and just gets much more performance. Or they had a better optimizing C compiler. I’m not really sure what it is. That dip the two things take relative to the 400-MHz machine at the end -- this again is a huge circuit. We’d actually didn’t compile it. We just read in a very large design and "unique-afied" it. This instantly sucked up almost a gigabyte of virtual memory. So it became -- how fast can you page onto it? The 400-MHz machines have the advantage of hardware RAID to page on, and there’s some interesting stuff that also came out in that article. When we use EDA tools, if we start to page we know everything slows down, but we never really knew how much. I actually got some test data in the March issue.
Let’s look at the final slide, which is sort of the summary of the overall performance of both of these. My concluding remarks are that, first of all, these were designs anywhere from a few thousand gates to over a million gates that we ran on all the tools. I think that what makes Windows NT an EDA platform today is the fact that Cadence and Synopsys support the NT platform, as do many other vendors, as you can see from a look at the DAC floor today. It’s sort of a chicken or the egg problem. If nobody had products on the NT platform, it wouldn’t be an EDA platform. On the other hand, I think we’re sort of in the same boat with Linux. We don’t really have what we consider the industrial-strength EDA tools, such as the Cadence Verilog XL [and] the Synopsis design compiler. Whether or not you use these tools or something roughly equivalent to them, I don’t think anyone will dispute that those are the real dominant players today, and I think those pretty much define the market.
Just so that you guys don’t get me wrong as this discussion moves forward -- personally, I’ve been using Linux for several years. Also personally, on my desktop at work for several years I have either had a Windows 95 or a Windows NT machine for my office productivity [applications]. So I’m sort of neutral when it comes to these things, but because of my expertise running these benchmarks and in using EDA [tools] in general, hopefully, I can be sort of a central point here and say, "Well, maybe one way, maybe the other." But at least [I hope to] put some sanity into the discussions.
I think the overall summary is that, well, today NT is in fact an EDA platform. As we can see with the results, it actually works. If you read the magazine and see more details on the design you’ll find out that we’re not just using cooked-up test cases; these are real designs. And it [NT] works relatively equivalent to what we’d expect on the traditional Unix platforms. Thank you.
Larry Augustin : Thank you everyone for showing up here today. My job, I think, is to stand up here and say that Linux is a real computing platform. I come from the EDA world. I did a lot of work in VHDL about 10 years ago. Some of you may remember that. So I understand EDA users, and when Jonah gave me the opportunity to come up here and tell all of you how Linux has really made it, I was excited to do that. I would love to see [tools from] the top EDA vendors running on Linux, because I think that’s what you as users want. And that’s the basic theme for most of this -- if that’s what you want, that’s what the EDA vendors should do. And I’m going to encourage you constantly to let them know what platform you want.
Now, a lot of the objections to Linux come from things like, "well, it’s not used by corporate IS"; "it’s not a real platform"; "it’s just used by a bunch of hackers." That’s not true. Datapro’s 1997 international user rating survey of Unix and NT, which just came out in February of this year, is a wonderful source of data on real commercial users running Linux, NT, and other operating systems. What that tells us is that 14 percent of Unix installations worldwide are running Linux. Now these aren’t just people downloading it for free and running at home. This is a survey of real corporate MIS people in real companies running Linux.
Linux is now the No. 1 Unix OS in Germany. Linux is the No. 2 Unix OS in the Asia-Pacific region. These are areas that tend to lead in technology. They’re the early adopters. They’re a little bit more distant from the U.S., so marketing doesn’t make as much difference there. They’re more interested in the technology.
In the survey, Linux was rated No. 1 in overall satisfaction. That means, overall, Linux beats Solaris, Hewlett-Packard UX, Digital Unix, AIX, NT -- all of those systems in overall user satisfaction, across all the categories.
Another great telling number on the X86 platform last year: Linux was the No. 1 new server OS installed on that platform. So Linux is real.
Now, a lot of people say Linux is unsupported. Not true. Linux has great support. In fact, most people would say Linux has much better support than NT. Infoworld magazine gave Linux its best technical support award for 1997; I’m sure many of you saw that. It was just a tremendous boost for Linux that people are realizing that the community supports it and that commercial vendors support it, too.
At the same time in this Datapro survey, NT was rated by users as one of the worst operating systems for vendor support. So if you want support, I guess that means Linux is the way to go.
Linux was rated in the same survey as one of the best operating systems for vendor support. And you can get commercial Linux support from a variety of sources -- [I've] listed a few up there.
A lot of people would have you think that Linux is unstable, unreliable, and poorly tested. I mean, what -- it’s just a group of hackers putting something together, right? Well, not according to the real corporate users. In the same Datapro survey, they ranked NT as the worst operating system for product quality and product functionality. At the same time, Linux was rated excellent in product functionality and total cost of ownership, which is very interesting. Total cost of ownership going to Linux means people were looking at maintenance over time. They found they needed fewer system administrators; they needed fewer people sitting in front of machines or having to go in front of a machine to do work to maintain Linux.
Linux was rated ahead of NT in product quality, reliability, scalability, etc. One telling quote there from this Datapro survey: "NT and AIX were the only operating systems where no attributes showed up with a very good score." The end result here is that the people who are running Linux love it. They think it’s a reliable operating system; they think it’s stable for them; these are real corporate users running Linux in mission-critical applications, not just hackers off the net. It’s a real operating system, and if you EDA users want your tools on it, then we need to get the EDA vendors to listen and provide those tools to you on Linux. Thank you.
Jonah McLeod : OK, after that damning criticism of NT, I think that it’s only fair that we give Microsoft a chance to address some of those comments. So I think I’d prefer to ask Dan to just talk a couple of minutes about some of the things that he just heard a minute ago.
Dan Small : Sure, thank you, Jonah. I have to say that for each of those statistics Larry provided, I could probably find others that refute those points and strengthen Windows NT’s position.
In point of fact, the Windows NT workstation market is growing significantly, and I think that over the past couple of years Unix platforms in the workstation market have been essentially flat. In fact, many of the vendors of Unix platforms find their sales declining in the workstation market. Windows NT workstation sales are up 80 percent year over year and continue to show strength not only in the EDA market but across all of the engineering markets.
For example, last year at this very conference you’d be hard-pressed to find a Windows NT system on the floor -- certainly, not from any of the major vendors. Of course, Microsoft had a booth here last year, and we showed some Windows NT products in our booth, but otherwise across the floor there were very few if any Windows NT applications. This year on the floor, they’re conspicuous in their presence. I think that’s indicative of the general trend of acceptance in the EDA industry.
It’s certainly true that Windows NT is not Unix. It’s not Linux, and there are things about Windows NT that are done differently. That’s not to say that they’re done worse; they’re just done differently. It takes a while to become acclimated to the operating system, acclimated to the environment, both from an administrative perspective as well as an operational perspective. And once that acclimatization occurs, Windows NT delivers the same degree of reliability and support as Linux.
One of the things that’s important to understand, especially with a very complex electronics design -- and certainly with any business operation that takes place -- is that you’d like to have some avenue of recourse. In the case of a free operating system where support is distributed, for sure you can get technical support across the network. But if the project you’re designing, if the chip you’re designing, contains errors that ultimately lead to liabilities, who’s to be held accountable for that? There really is no accountability in the Linux marketplace. And while you may scoff at that, that’s really an important consideration as corporations move to design new computing infrastructures.
But there’s really one key element of Windows NT that many people in this particular Audience haven’t been considering in real depth. If Windows NT delivered everything that Unix delivered in the same way that Unix delivered it, it really wouldn’t offer much value at all. And in fact, what’s happening in most of the markets within which Windows NT is placed -- certainly on the commercial side but also in the engineering marketplaces -- is that there’s a very strong movement toward integration that is essentially a value-chain integration from the designer’s desktop to the distributor’s retail outlets. The key is to develop an infrastructure that allows seamless integration of parts and product design, development, assembly, and distribution, and that’s where real productivity gains will occur. That’s what Windows NT ultimately will offer far and above any of the Unix variants that exist in the marketplace today.
Part of that comes from an existing infrastructure that we have deployed in many commercial situations, but also we see this increasingly occurring in the designer’s desktop. Certainly, in the mechanical engineering marketplace there’s a high degree of integration between tools, and that integration leads to significant productivity gains.
As Windows NT becomes more prominent in the electronic [design automation] marketplace, I would expect to see those same kinds of productivity gains. That’s not going to be possible with Linux, so with Linux you can have a good operating system that performs well, but it’s essentially the old way of doing things. If you want to create innovation and revolution in product design, keeping costs down and improving time to market, Windows NT is really the only operating system that’s going to be able to deliver.
Jonah McLeod : We had a question over here?
Audience 1: I’m curious about your thinking [inaudible] Linux in terms of [inaudible].
Dan Small : That’s right.
Audience 1: [inaudible]
Dan Small : Well, we certainly have the resources to repair problems when they need to be repaired. In terms of warranty, if you don’t like the product you can return it for a full money back, for sure. But we have the resources and wherewithal to fix any problems that occur, for sure.
Jonah McLeod : OK, get this one over here.
Audience 2: I’ve been developing CAD EDA software for eight years, mostly on Unix and a very brief and unpleasant time on NT, and I’d like to address some of the points you’ve brought up.
Dan Small : What was your experience like the first day you were working on Unix -- was it easy?
Audience 2: It was excellent. I came from DOS, so Unix was actually a natural command line -- no problem.
Dan Small : So there should be no problem working with Windows NT then. I mean the same facilities are available.
Audience 2: Well, there is. You’re right, but then on the other hand my Linux box doesn’t lock up randomly once a week for no apparent reason.
Dan Small : Well, I have to tell you I’ve been using Windows NT for five years, and it’s never had that problem, either.
Audience 2: One of the things I’ve noticed in this discussion is [that] there seem to be two worlds. There’s the world of people who have really rotten experiences with NT and the world of people who say, "Well, I don’t have problems." I just can’t reconcile that. Everybody I’ve worked with is on the "hate NT" side; I have to admit that.
But anyway, let me get back to my point, if I can remember it. I came into a completely Windows company. There wasn’t a Unix box in sight, and we had to develop some very specialized CAD applications. I suggested [that] we use Linux. I figure by doing that I saved my company, which is a small company, about $10,000 in OS licenses, compilers, development tools, and all that sort of thing. People weren’t used to Unix, so I put a window manager up that made it look like Windows. And the only way they know they’re not running Windows is that it doesn’t lock up or give them silly error messages.
So there’s been a huge acceptance of Unix by people who are only familiar with Windows. The other thing you said was that Unix is the old way of doing things and NT is the new way of doing things. Unix has been around for about 30 years because it’s a good design, and good designs last. The bicycle design has been around for about 100 years or more because it’s a good design, and you can’t just throw away 30 years of experience, intensive research, and hope to come up with anything comparable in about six or seven years. It’s just not going to happen.
Dan Small : I’ll just answer that one point, which I find to be somewhat specious reasoning. It’s true that a 30-year-old design could be good, but it’s not necessarily so. I mean, for example, if you look at the automotive industry, which offers lots of analogies for the computer industry, you have a company like Lexus, which appeared on the marketplace and took market share for good reasons. They offer a good-quality car that was developed in a very short period of time -- in three years or five years. Mercedes saw their sales drop dramatically, and Mercedes of course invented the car back in the late 1800s. So the length of time in the market really has no bearing on the quality of the design.
Audience 2: Not necessarily
Dan Small : Not necessarily
Audience 2: The final point is about all those figures about NT’s market share increasing and all that. Well, you guys have billions of dollars to spend on promoting NT. Linux probably has a couple of million dollars a year spent promoting it, and some people say its usage is doubling each year. You guys have advertising dollars, and we have a good product, and I think that’s the fundamental difference.
Audience : Applause
Dan Small : For sure, Linux deployments are growing. It remains to be seen whether it will continue to grow at the same rate. It’s easy to go from one copy to two copies and claim 100 percent growth. It remains to be seen how that’s going to fare.
Jonah McLeod : Please, if you want to ask a question, stand behind the microphones so everyone can hear the question.
Larry Augustin : You know, I get the feeling we could sit up here and abuse Dan for quite a long time. That may not be the best use of our time, but it’s OK with me. What I’d like to do is just look out and ask how many of you want to see your EDA tools that you’re using [ported] on Linux? Now, I think that just answers the question. How many of you want to see your EDA tools on NT? Now why are the EDA vendors porting to NT then? Why aren’t they porting to the platform that you want?
Jonah McLeod : Now with that question, let’s try to get one of the panelists to address that. Venk, would you take a crack at that please?
Venk Shukla : Yeah. It’s interesting to see this discussion. Seven or eight years ago, we used to have the same kind of religious wars between Verilog and VHDL. EDA vendors are not necessarily proponents of one religion or the other. EDA vendors are a lot like a person I used to know back in India, Mr. Hussein. He used to supply food to the two warring groups in Afghanistan. After the Communists withdrew and the Russians withdrew from Afghanistan, there was still a war, and [a war] solely based on ideology and which version of religion [should dominate] and to what extent. So it was an ideological war.
Mr. Hussein had his communication lines open to both groups, and he used to barter dried fruit from both those places in return for food. So I would say that EDA vendors are a lot like Mr. Hussein. We don’t care. We don’t care who is right, who is perfect, or who is superior. What we care about is this: Is there a customer demand for that [product]. And it’s interesting that so many hands were raised here for Linux, and yet in the two years that we have been interacting with the customers, not a single company -- not a single company -- has ever asked for our tools on Linux. Now let me point out that there are not many companies who have asked us for tools on NT, but a few have. The few with deep pockets have asked that.
So my suggestion to you guys is that if you really mean this about EDA tools being available on Linux, it’s a very simple thing to do: Just ask your management to ask for it. As far as Ambit is concerned, we were a small start-up, and we had ported all of the software to Linux. And we did it for three reasons. One was a price/performance thing: We couldn’t afford high-end Sun workstations at that time. The second one was that we wanted [to facilitate] off-site development of [our products by] engineers. They wanted to take it home and work from home. So we have extensive experience with Linux, and we are very satisfied with Linux, but it's sort of intriguing to me that there’s never been this kind of a groundswell from any of the companies that we have been dealing with so far.
Maybe a result of this panel discussion will [be to] change that. And that’s good, because we have got the port, and all that you need to do is ask.
Jonah McLeod : OK, we have a question, please.
Audience 3: Yes, I’m sorry to... the gentleman from Microsoft mentioned that there’s sort of a cost of acclimation to NT, and I’m curious to know what Microsoft plans in terms of the long-term cost of keeping an acclimated support staff available. Do you have a specific strategy to address that?
Dan Small : Well, there are a number of studies that have been undertaken by outside organizations that discuss the total cost of ownership of the Windows NT platform. And we can provide those if you’d like. However, we recognize that more work needs to be done in developing elegant centralized management facilities, and a lot of that is under way. Over the next year or two years, you’ll see a tremendous focus from our company to really get a hand on those particular issues to be able to centrally manage Windows NT and also to make it friendlier to existing Unix environments and the investments that have occurred around Unix.
For example, about a month or so ago we announced a new product called Unix services for Windows NT, which among other things will include a Microsoft version of NFS to allow plugging into a Unix-based NFS network easily, as well as password synchronization features, telnet services, and so forth. Over time, you’ll continue to see other utilities and other interoperability technologies included with Windows NT or otherwise provided by Microsoft to make it easier to co-exist in the engineering environments. Does that answer your question somewhat?
Audience 3: [inaudible]
Jonah McLeod : Could you step to the microphone please?
Audience 4: I have a small company, and we are involved in EDA as well, and we are on Linux. I personally sent e-mail asking for Linux tools to basically all of the EDA tool vendors who were players in the FPGA [market] except, obviously, the very top, which we couldn’t afford anyway. And the constant answer was that there is no market need for that tool or that platform. And they had no plans to release [tools running under] it, and the only one released that was anything like a synthesis [tool] was Exemplar. Only they canned it too for not having a market. Now the Audience here seems to be a market. That contradicts what you said that you didn’t have a single request; apparently, there should have been requests and maybe mine has been lost somewhere in the mail.
Venk Shukla : You know, it’s very easy to change. My e-mail address is email@example.com , and anyone who is willing to pay $100,000 for a high-end synthesis tool, send me e-mail and this problem will be solved.
Audience 4: All right, my playing field is one zero less [laughter from Audience ]. But there is another interesting aspect of that marketing business. I realize that quite a few EDA tools sell for Unix and NT on a little bit different price range -- like a ratio of 1:2. I mean that the Unix tools cost X, and the NT tools that do the same thing cost half of that. I don’t really see why that would happen. Also, we are looking for FPGA tool chains, and we came across Quicklogic. For $3,000, you get a Quicklogic place-and-route [tool] for Unix. For the same $3,000 for Windows, you get the place-and-route [tool], you get synthesis, you get Verilog simulation, an editor, and all sorts of gadgets.
Now, either you can tell me why would that be a very good strategy for a tool-making company like Synplicity. Or I can see it as a very hard push on the market to turn to Windows solutions, because a small company like mine obviously would go for the cheaper one. If I have to fork out $20,000 for the synthesis [tool] on top of the $3,000, then, obviously, the Windows solution would be a much more valuable solution. Not for me, because I personally wouldn’t do it; I am a Linux fan. So I don’t really see the market reasoning behind dumping Windows products.
Jonah McLeod : So you want to know why there’s a price difference between the Unix version and the NT version?
Audience 4: Well, on the one hand, that, and why is the price difference so large?
Jonah McLeod : All right, Venk; you have the $100,000 product.
Venk Shukla : Yes, interesting. I’m not aware of any EDA company that’s charging a different price for NT [tools as] compared to Unix. At least not the big ones. I’m not aware of others.
Audience 4: [inaudible] Simplicity has a 1:2 [price difference]. The latest quote I got was $20,000 for Sun and $10,000 for NT. More or less the same was true for every chip vendor place-and-route toolset.
Audience 5: Mentor’s [inaudible] tools [for Unix] are four times the price [for NT]. Viewlogic has a few tools [for Unix] that are four times the price [for NT]. They’re packaging a "lite" version, or a stripped-down version, for NT and a full-function version for Unix. The stripped-down version is pretty much all you need to get your job done. There’s a very large price [inaudible] difference.
Jonah McLeod : OK. Dan [Page], you had something to say.
Dan Page : I’d like to speak from Avanti’s point of view and also the EDA vendors’. I’m going to reiterate one fact, which is that from the EDA vendor’s point of view we really focus all of our value on the content of the software -- not, necessarily, on the delivery. And so the platform turns into a cost for us. It turns into overhead, and it takes away from our ability to put more time and resources into the value of the product. That same argument goes into pricing, and as a company we’ve chosen to keep our pricing very close on NT platforms or Unix platforms. So we don’t really differentiate, because we feel that we’re delivering the value as long as the operating system is capable of running the application.
If you look at what we deliver on, we deliver on those platforms that are demanded by our customers. And so we have a variety of different systems, depending upon the application. We deliver some NT [tools] for certain products, Unix for other products, and so forth. We could probably argue on the merits of both operating systems or three operating systems. All of them can be configured to run EDA applications. I don’t think that’s really an issue, and it will boil down to user preference. But there are a few issues that are kind of interesting from an EDA support point of view.
One of the things that we notice while trying to support EDA applications is that a lot of our problems are environment-related. The application just isn’t performing the way we expect given the environment that it’s operating in. One advantage that NT has is that it's a relatively confined environment. From the Linux point of view, it worries me and troubles me somewhat to think of how many different flavors, versions, and configurations there are out there. This gives us a somewhat unstable -- not from a run-time point of view but from an environment point of view -- target platform to be running on. I haven’t seen that issue addressed, and that’s one that concerns me more than any other. But, certainly, we have applications that have been ported internally to run under Linux, and they do seem to work quite well.
Phil Tomson : Can I address the environment question, real quick? First of all, I mean NT is not that pristine, either. I mean there are several patches out, several patch levels that you can download. DLLs tend to get changed pretty often, and that basically changes your OS. So you’re not really seeing an advantage there.
Audience 6: I have a question for the gentleman from Avanti. Have your customers requested products on Linux?
Dan Page : Very few. We’ve seen, I would say, a small trickle of requests under Linux and then, depending upon the application, a slightly larger trickle under NT. But by far we still see a solid base of traditional Unix platforms, such as Solaris and Hewlett-Packard UX and so forth.
Audience 6: And what has your response been to customers who have asked for products on Linux?
Dan Page : Well, again, our focus is to put most of our effort into the content of the software and try not to deliver [tools for] an excessive number of platforms. If there are enough customers out there to warrant moving our application, we’ll do it.
Audience 6: Do you ever request porting fees for Linux from customers who request it?
Dan Page : Not to my knowledge. Again, we don’t typically port for fees. We typically port when we see enough of a user base out there to warrant moving the application on its own.
Audience 6: I disagree with your assertion about fees.
Jonah McLeod : OK, we’ve got a question all the way in the back. He’s been standing there for a while. And then you, and then the fellow in the middle, OK?
Audience 7: Recently, we did a study [showing] that most of the engineers in our company -- we are in automotive electronics -- do only 20 percent engineering work. About 80 percent of their job is dealing with the customers, with suppliers, with colleagues, and so on. Most of the nonengineering tools that they use are electronic mail, spreadsheets, and word processing and so on. I guess I would ask the panel about this issue: How do we interoperate between those nonengineering tools and engineering tools on a Unix platform?
Dan Souder : I’d probably be a good candidate to take that one. At least in the Solaris environment, some of our third-party vendors offer a number of tools with the same type of functionalities you would see from, for instance, a Microsoft Office suite. We also offer native applications that allow you to view those types of documents so that, at least if things are sent to you in a mail format, you can view them. I think that in general it’s a shame to dictate an engineering platform based on the need for productivity tools. And with the performance of current Unix desktops, you can probably support most of those tools even via emulation, because the performance required to run those tools is just not the same as what’s required to run the EDA tools. There are a number of products that allow you to do that as well on a Unix platform. So you can actually run your productivity tools in a Unix environment very effectively.
Larry Augustin : Linux has available a couple of very nice office suites in Applex and Staroffice. Wordperfect runs under Linux. Corel has announced a port of the Perfectoffice Suite with Quattro Pro and, of course, Wordperfect and all that to Linux. There are various options available, such as Wabi for running Windows tools [in emulation] under Linux. So from what I see in most of the offices, that’s not much of an issue.
Dan Small : I’d just like to add, though, that all the emulation software that’s available today falls short in a number of really important areas, especially in the area of comm [communications] support, which is really the foundation technology that allows applications to interoperate in a pretty seamless way. So while you may be able to support applications on the machine, they really fall short in achieving that seamless level of interoperability.
Jonah McLeod : OK, we have a question over here.
Audience 7: Actually, I had a comment to make. This goes back a little ways, to the gentleman from Ambit who was talking about not getting any requests for Linux tools. I’ve been running Linux at the office on my PC on the sly, because I don’t want our IT department to know about it, for about a year and a half. And I password-protect my computer at boot time, so that they won’t reformat the hard drive while I’m gone.
A big issue is requesting it [Linux]. I’m from Hughes Network Systems, and we spend a lot of money on EDA tools -- all the very expensive tools. We’ll probably be evaluating Ambit in the near future. And the fact is that a year ago, and in the past year, it’s been very difficult to even discuss the fact that I use Linux with anybody in IT, even with my superiors. As I said, I have to deal with the questions: "Well, why are you doing that?" So I’ve been spending the past year taking sort of a grass roots movement approach, and I think a lot of people have. That’s why you’re seeing this get so hot right now, showing that this [Linux] is a viable solution just for everyday use. The next step is doing our engineering applications directly on that platform.
So the fact that you haven’t seen it [the groundswell for Linux] yet is probably indicative of the way that Linux grows, which is grass roots, word of mouth, with very little advertising. Whereas if you compare that with things that come from Microsoft, as another person said, the advertising can precede the acceptance of the product. And that’s exactly what’s happened. NT had been talked about for many years before it actually started to grow significantly. And that’s why it makes sense for the EDA vendors to have ported their tools to that over a year ago, or to have started over a year ago. But even in your own situation at Ambit, it made sense for you to use Linux, but you didn’t talk about it. Again, it’s indicative of the grass roots movement for Linux. I’d be surprised if you don’t start to hear a lot more formal requests in the coming year.
Phil Tomson : It would be interesting to speculate as to what might have happened had Ambit released a Linux port.
Audience 7: I’d like to comment on that. In particular, Exemplar did release it [a Linux tool] and I took note of that. I downloaded it and tried it out. The fact is, we don’t use Exemplar. First of all, they’re just trying to get into the synthesis market and make a substantial impact there and at the same time go into a new operating system market that doesn’t have broad-based acceptance. That’s a tricky thing to do. I would venture to say that if Verilog had done it, if Synopsis had done it [ported tools to Linux], I would have been on my boss immediately to get a license for that [tool] paid for or to have one of our licenses converted to that. I bet he would have gone for it, too. But that’s because we use Verilog and Synopsis. That’s it.
Venk Shukla : Let me comment on that. I think a groundswell of customer demand is certainly one kind of threshold point. Hopefully, all of the underground activity that’s going on will organize itself, and EDA vendors will begin to hear from the management of the companies who make these purchase decisions. That hasn’t happened yet.
But the second important point relates to having a critical mass of software on a particular platform. Let’s say there’s one customer who stands up and says, "I want it." Now that’s the worst possible thing to happen to an EDA vendor -- to have a port for one particular customer. That’s the worst possible punishment you can give to someone.
But even if the EDA vendor decided to be proactive and take that kind of risk, you still need a critical mass of software. For instance, for a synthesis company, it’s an absolute prerequisite that you have simulation software like Verilog XL available on VCS because in our QA suite, we have a huge [inaudible] planning on 70 machines for weeks before we release a product. And all those things get verified against Verilog to check the functionality. So it’s important for two things. One is that there has to be customer demand -- and not just a groundswell of engineers but people who make purchasing decisions for them. And the second is that you must have this critical mass of software. What you see is that in the case of NT both those things exist. There are companies with deep pockets who want the software to exist on NT, and since they’re willing to pay for it there is a critical mass of software now.
Jonah McLeod : OK, we have three questions. Go ahead.
Audience 7: Can I just make two more comments quickly? I’ve been standing up here for a while and other things came up.
The first thing is regarding the difficulty in supporting all the different Linux configurations, particularly with libraries. I would say that’s the only technical issue facing EDA vendors, software vendors in general, and Linux today. And I think that is starting to be addressed. There’s a lot of people that care about that right now. There’s a thing that’s being discussed right now called the Linux standard base, which will take kernel revisions that you can look at and have confidence in the stable kernel revision that you’re going to support. The Linux standard base would take into account all of the basic libraries as well, and then [give them a] version [number] so that you would be able to say, "Linux standard base version such and such or better is what we support." And I think that’s going to help. It’s going to go through a lot of gyrations before it gets resolved, and the commercial Linux guys have got to come out and support that. But I think that’s a significant response to that concern, which, as I said, is the only technical concern. I think anything else is simply a cultural concern.
Audience 8: Can I get a comment in? I think both the two gentlemen from the EDA vendors said that they have in-house Linux ports. I guess the question would be: Why not release them unsupported? Let people, maybe current customers, download them and use them and call it unsupported?
Jonah McLeod : At their own risk?
James Lee : And using their firm licenses that they have via their current licenser running. You’ve got Flex LM already running on a Linux port, so to allow your licensees [to use] Synopsys right now, you get a design compiler license, and whether you use it on Solaris or NT it’s the same thing and it’s Flex LM. So it doesn’t matter what port you’re on.
Audience 8: It would be a good way to gauge the actual demand out there.
Jonah McLeod : OK, we have three more questions, and then we’re going to have to wrap it up. The three that we’ve got are right in a row, so first you, then the gentleman behind you, and then the gentleman at the mike, OK?
Audience 9: I was just going to talk about the Linux standard base also and expand on that. The first products of that [effort] should come out in two or three months. There will be a standard reference platform. If your application runs successfully on that, it will run successfully on every commercial version of Linux. And the second thing [to note], for people who want to get into the market immediately, is that 95 percent of your customers are going to be running Red Hat Linux anyway, so you can just port to that.
The other thing I wanted to say was in regard to what Microsoft said, that I’m glad that they’re willing to fix all these bugs because I’ll be providing you with a list. And I wanted to encourage all the EDA vendors out there to port to Linux because eventually Microsoft no doubt will have its own EDA tools [laughter from Audience ].
Audience 10: That’s a pretty hard one to top. I’m another data point. I’m Phillip Laidlaw at Cadence Design Systems. I’m the manager of platform marketing, and I actually make the decisions on what platforms we’re going to deploy on. I have not heard boo about Linux for 10 years, but within the last two months the phone has been ringing off the hook. Well, that’s good, and it's good that it's a grass roots movement. But unfortunately, they’re primarily university people, and our university software is almost free. The point -- and the point that Venk was making -- is that you have to vote with your dollars, and we have.
There is some pretty stiff platform competition out there for Linux, and there has to be a real groundswell -- and, unfortunately, a groundswell with a lot of money behind it -- before we’re going to vote. I mean, Cadence said, "Let’s seed the marketplace," and in 1990 Cadence was supporting 11 different architectures and trying to deploy tools. And everybody was saying, "Well, when is my tool going to be available?" We’ve pared that down as far as we can. We "end-of-lifed" our Sun OS port, which was 80 percent of our business, to try to get our customers to migrate to newer versions of operating systems. So you guys are totally in control of this, but I know it’s going to be a real war, and you’ve got to get your managers and people that are controlling the POs to say, "We want this." There’s a lot of interest but there’s no demand. We’re not getting beat up about this.
Larry Augustin : Wait. Let me say this about Cadence: I know of one big Cadence user who has told me that they’ve been begging for a Linux port for two years.
Audience 10 [Phillip Laidlaw]: I’ve been in this job for four years, and I’ve never heard one person beg for it. I’ve heard some university students say....
Larry Augustin : Not only that; they said there are reps from Cadence who told them that it’s running. Cadence, you know, Veralog XL, is running on Linux internally, and they just haven’t decided yet [to release it commercially].
Audience 10 [Phillip Laidlaw]: That’s not true. We run our own tools off the CD internally.
James Lee : Phil? I’d just like to comment real quick. I know you and I know each other and go way back, but I also know, as you said, that you’ve dropped Sun OS, you dropped DEC Alpha, and you’re dropping AIX. So you’ve got room for another port.
Audience 10 [Phillip Laidlaw]: We’re not dropping AIX.
James Lee : Well, you will sooner or later. Everyone knows that platform is going away. But the point is, having worked for Cadence for a number of years myself, I do know you have a Linux port. It’s not official, but I know many R&D folks, just like [those at] Ambit and Avanti, were willing to admit it publicly. I know of many engineers who work at home on various portions of various tools under Linux. So you essentially have a Linux port as well. Unfortunately, it’s like the gentleman who stood up before [said]: Linux is the best-kept secret; you don’t dare admit it to your boss. And I really have to thank Venk Shukla for admitting publicly that they do it; he just doesn’t have the belly yet to release it publicly. In terms of the general support of licensees -- and I know the standard base is a good thing dealing with different versions -- every time, right now, Cadence will say, "Well, we support Solaris 2.51; if you’re on Solaris 2.6 good luck."
Audience 10 [Phillip Laidlaw]: We support Solaris 2.6.
James Lee : Well, I’m just saying that as an example. So any time any EDA vendor releases a tool they say, "What is the supported release?" And I think that same thing would hold true with Linux.
Audience 10 [Phillip Laidlaw]: Yeah, but James, you know probably better than I do that, as somebody mentioned, it’s the configuration and it’s the platform situation. We have over 300 tools. We try to limit that configuration exposure as much as possible. So there may be people internally who are running Linux, but if they’re at home and it blows up and they call the CRC, the CRC is going to say, "It’s not a supported configuration."
James Lee : Well, since they're R&D, folks are not going to call the CRC.
Audience 10 [Phillip Laidlaw]: But still, the internal mechanisms to support [Linux] are not there, and that internal mechanism for [inaudible].
James Lee : Well, bottom line, Phil: I agree with you, and I agree with Venk and the gentleman from Avanti, as well. EDA customers have to vote with their dollars, and I think my benchmarks have shown that the performance of 300-400-MHz Pentium machines is comparable to that of the current high-end Unix platforms. So I think the people who are looking for free or quarter-cost-type versions of tools are barking up the wrong tree. But I think if you’re willing to spend the same amount as you spend for Solaris or Hewlett-Packard UX for your tools, that’s the price point that [inaudible] Synopsis and Ambit -- sorry, I don’t know about Avanti, I don’t know about everybody’s tools -- but that’s the price point most people are coming out with. The same price point for NT versus their traditional Unix tools, and that makes sense to me.
Jonah McLeod : OK, any rebuttal?
Audience 10 [Phillip Laidlaw]: That’s it. As I said, I’m definitely seeing the groundswell, and keep it up.
Larry Augustin : And talk to some of us. We know Cadence users who claim they’ve been asking for it [Linux-based tools] for two years now.
Audience 10 [Phillip Laidlaw]: Great. Actually, have them call me.
Jonah McLeod : [inaudible]
__: We’ll do it. Just send me an open letter and we’ll publish it.
__: [inaudible] publishing deadlines [inaudible]
Jonah McLeod : OK, I’ve kind of let the panel slide here. Are there any other comments outside the panel on any of the things that we’ve said so far?
Dan Souder : I’d like to make a comment.
Jonah McLeod : We’ve beat up on Sun pretty badly, so ....
Dan Souder : Actually, of course, we are a Unix vendor, and so Linux is not that foreign to us, first of all. We do actually have some involvement with the Linux committee, and I think that at some point in the future the Unix camps are going to have to come together to stave off any kind of involvement from the other side of the fence. I think in general Sun’s view is that any Unix is better than the alternative [laughter from Audience ].
Dan Small : I’d like to add that we share that view. In other words the more Unix the merrier [laughter from Audience ].
Dan Souder : It just makes the marketing job that much easier. But in general, Sun is involved with the Linux community. I don’t know if you’re going to see a unification of Linux and Solaris. We do also support Intel platforms as well as Linux does. In fact, that’s what we’re running at DAC this year. If you’ve noticed, the DACnet machines are running Unix, not NT, and there was a big reason for that. I was the integrator for it, for one, and I work for Sun.
But the other reason was that we had a massive amount of difficulty in getting the environments to cooperate together. This is a Unix market, and Linux integrates very easily into a current Solaris environment. I would have chosen Linux to run on those machines before I would have run NT, in a heartbeat, because of the simplicity of integrating that environment with a current Unix environment, which is what most of the people in this room are probably running. And most of them are probably running it on Sun. So in general, for Microsoft to be a player here they need to see more integration, more compatibility with the current Unix environment, to be successful.
Jonah McLeod : Any comments?
Larry Augustin : To add to that, for instance, Linux supports NFS and NT doesn’t without buying a $300 add-on, which doesn’t work terribly well.
Jonah McLeod : OK, we’ve got the gentleman in the back, we’ve got one gentleman over here, and then one gentleman in the back over there. Then we’d like to resolve it, because we’re just about finished. OK, go ahead.
Audience 11: I think my question falls under your last comment that for people who are actually working with EDA tools and trying to put flows together, there aren’t [in NT] the little extra bits that you get in Unix to actually cobble the stuff together. Like all the shells, the compilers, and things. If you go to your boss and say, "I need $300 for an NFS server to get something done," the guy says, "Well I’ve got to budget that. You'll get it next month or the week after," or something [like that]. Unix just generally has all the bits that you can cobble together in a couple of hours if you’re a reasonably good programmer in shell or [inaudible] or C or something.
__: And that’s a good point. Most of the Microsoft-based apps are graphical, and many times there’s no way to run them from the command line.
Audience 11: It sort of takes the [inaudible] out of EDA when you use NT, I think.
James Lee : OK, just to iterate his point, I happen to have a slide which lists some of what I believe someone called the bits you need to cobble together. Some are available for NT. I think that when I totaled all of it up, I found you can get all these things. You’re installing 11 different packages on NT, but you can get it all.
Jonah McLeod : OK, gentleman.
Audience 12: I’d just like to ask Microsoft, why don’t you bite the bullet and market a Unix?
Dan Small : We actually have an equity stake in SCO. SCO Unix was originally based upon Xenix, which is Microsoft source code.
Audience 12: They’ve taken that out recently, I think, too.
Dan Small : You might be right.
Audience 12: And there’s a reason for that.
Dan Small : The design points and the design philosophy behind Windows NT are different from those of Unix. It satisfies different needs.
Audience 12: Well VMS was different as well, and look what DEC did eventually.
Dan Small : I don’t understand what your point is.
Audience 12: My point is that, you said that NT is different, and that appears to be the problem. So....
Dan Small : Actually, Windows NT has been very successful. It’s hard to say that its difference is [inaudible].
Audience 12: [inaudible] not a [inaudible] market, though.
Dan Small : Perhaps not in the EDA market, but in the mechanical engineering market -- just permit me [to speak] for a moment -- Windows NT is selling upward of 40 to 50 percent of all new licenses from major companies. All the Unix providers. The bulk of their sales are now Windows NT. It offers real benefits. There are real benefits in those marketplaces. As I mentioned before, when you design products, to be able to share information between the design station and the commercial functions, the people that need to make decisions about how those products ultimately come together and are sold and so forth -- that’s a really important advantage.
Larry Augustin : Comment on the mechanical CAD versus EDA: They’re rather different.
Dan Small : They are different.
Larry Augustin : Mechanical CAD people don’t need to run 20 simulations overnight and ensure that the results are correct and try to script all that. That’s difficult [inaudible].
Dan Small : Yeah, but the scripting is almost a failing of Unix, not a virtue [laughter from Audience ].
James Lee : You can’t run EDA on NT without LSF.
Dan Small : I totally appreciate that. That’s not what I meant to say. But it’s almost a historic issue. If you were to solve that problem today, you would solve it dramatically differently. So I appreciate that it’s necessary.
Phil Tomson : The statement you made was very characteristic Microsoft arrogance. You said, basically, that you don’t need scripting. Scripting says, "OK, we don’t know everything that you’re going to do as a user; you can go off and put these things together."
Dan Small : Scripting means the tools may be not be flexible enough to adapt to every situation.
Jonah McLeod : OK, I think we’re just kind of beating a dead horse here. Let’s throw the next one.
Audience 13: First of all, a noncomment on that last little bit. Everything in life is input, output, and pipes. We are input, we are output, and we are a pipe. So that’s human.
Secondly, I want to talk to the three marketers. I guess, that’s all I really noticed here. I’m confused as to the way you run things. My third [wedding] anniversary is coming up, and I’m going to take this to a relationship analogy with my wife. As a husband, if I sit back and wait for her to say, "You know, Rob, you suck as a husband because of this, this, and this," do I really care about the relationship? No, I need to go to her and say, "Hey, what can I do to be better?" Now, it sounds to me like you guys are back on your heels [saying], "Come to us, tell us what you want better." You need to be saying, "Hey, do you want this, do you want that?" What does a Web-based survey cost you to make?
I hear you guys saying, "Well, our customers aren’t coming to us asking for it." You guys need to love your market share as much as I love my wife. And I think you do; you’re going to say you do, but you’re not going after it. You’re saying, "Well, if there’s market share, come tell us about it, come throw us the money." Are you saying, "Is there money out there"? No. You’re saying. "Give us the money first."
When you were a start-up, you went around telling all your buddies, "I got this great start-up, man. I got the business that’s going to do it for the world." But now you’re all, "Well, you’ve got to come to us. We’re big. We’re slow. We’re not fast." I’m sorry, guys. I used to work for Cisco. We did it the same way: "The customer’s got to come to us." Our SEs didn’t have the effort to ask, "Would you like this? Would you like that?" No, John Chambers did that. John Chambers went to the customers and said, "What would you like?" See, John Chambers has the paranoid [inaudible]. It sounds like you guys don’t have the paranoia yet. I do with my wife; why don’t you guys with your stuff?
Jonah McLeod : OK, I think we’ve pretty much done it here. We’re going to have to cut it off. We’re 20 minutes over the limit, so thank you. We’re not going to solve any more of these problems here.
Venk Shukla : Can I comment on the last one? It’s an interesting story. But you know, the analogy is slightly off. I think a more appropriate analogy would be the kids saying, "I want this, this, and this." But the parents should make the buying decision -- saying, "No, that’s not what you’re going to get. You’re going to get this." So your problem is not with your EDA vendors; your problem is really with your own management. I think that’s a very important thing for this Audience to understand. It’s not the problem with EDA vendors. You’ve got to go to your parents and tell them what you want. It exists.
Jonah McLeod : There’s still a lot of frustration out here. We’ve got one more installment of the Linux versus NT debate coming out in the next issue -- the August issue -- which will air some more of your grievances. We probably collected a lot more e-mail, and we’ll probably do another one that collects the last group.
One of the things that should have come out of this panel here is that there is a disconnection between what the users want and what the vendors are supplying. I want to thank all the participants who came up here. I particularly want to thank Microsoft for biting the bullet and coming up here with this fierce Audience .
OK, with that, enjoy the rest of your day.