special section

From the NT camp
To the Editor:

In the Feedback section of the March issue [p. 12], Dan Pinvidic asks, "If you're sitting in front of a PC, how quickly can you print a hard copy of a directory listing? Is it as easy as ls -l > lpr ? Can you generate a list of files that were written by a specific user: ls -l | grep user_name ?"

There seems to be a persistent confusion between the tools that are provided with an operating system and the operating system itself. All of the programs that Pindivic lists are available for Windows NT, and I use them regularly in my work with EDA tools.

Although it's not well publicized by Microsoft, Windows NT has a command shell that is quite capable and distinctly different from the MS-DOS command shell. It supports unlimited command line length, long file names, true named pipes, and so on. Also, Windows NT was designed from the ground up to support multiple user interfaces, although the standard Windows interface is most commonly used, and there are a number of freeware, shareware, and commercial products that provide C and Bourne shells.

On my Windows NT system, I have all of the tools listed, and a few more: ls, vi, rm, grep, sed, awk, lex, yacc, make, cvs, tcl . So to print a copy of a directory listing, I type ls -l > lpt1 , and to generate a list of files written by a specific user, I type ls -l | grep "user_name" (although, to be fair, files are tagged with user names only on NTFS-formatted disks).

Also, in the "EDA Platform Benchmark" in the same issue [p. 62], James Lee makes some comments about the reliability of PCs in the sidebar "We Still Have a Hang-Up," stating that "everyone considering Windows NT for critical engineering applications should understand that it hasn't made hang-ups a thing of the past." He then includes a Windows NT system that sounds as if it had been physically damaged and other machines that had various problems.

This sounds like the same argument I heard at a client site in comparing Macintoshes and PCs: An engineer compared a new Power Mac with a cheap, no-name PC clone and complained that the PC was a piece of ­­­­­. Well, obviously!

Lee's implication is that Windows NT really can't be trusted to run reliably and is still inherently less robust than Unix. Of course software won't run reliably if the hardware it's installed on won't run reliably! That can include systems that have unreliable device driver and hardware combinations, as well as outright damage.

For better or worse, the PC marketplace is wild and woolly. There are literally thousands of different systems available, and choosing a reliable PC is much more difficult than choosing, for example, a reliable Sun workstation. I think that's particularly true when buying PCs from local vendors, who often don't have the resources to test hardware/software combinations extensively.

I do agree with Lee that system crashes are not unknown under NT, but the real reason has much more to do with the nature of the PC marketplace than with the robustness of the OS.

Rylan Luke
President
Rymar Engineering
Cupertino, Calif.

James Lee replies:
Although Rylan Luke's comments on the availability of the standard Unix commands on NT refer to a letter, I'd note that in our benchmark article we mentioned the need for such things and the fact that they are available from several sources. However, we did not use them in our tests. (But see also my reply to Kenneth Ryan and Ross King, below.)

I won't debate reliability, since the argument goes both ways, depending on who's doing the arguing, as evidenced by the letter here.

However, reliability is an important issue. A machine must have high reliability for use with EDA tools. I do agree, though, that you can't compare a cheap clone with a name brand like the Compaq, IBM, and HP machines, which gave us no problems. Many of the clone systems are still built using Intel motherboards, but somehow less than perfect results were achieved.

I think our inclusion of the hang-up problem should be a wakeup call to people thinking about NT to realize that they still have to think of the hardware. I would also like to point out that the machines in the benchmark that hung up were from known national vendors, and one of them was specifically burned in and tested by the vendor. On the other hand, the locally grown clone that hung up at my customer's site was quickly replaced with a system with the proper video board BIOS, the source of the hang-up.

The end result is that, although when we started the research for the benchmark I was skeptical that the NT platform would even be able to simulate a few tens of thousands of gates, let alone simulating 1.3 million gates in just a few minutes, I was happily surprised by the performance and reliability of the NT systems. Nevertheless, since all NT machines are not made by top-quality vendors, anyone thinking of using NT for EDA should be aware of the hardware.

From the Unix camp
To the Editor:

I read with great interest your article "EDA Platform Benchmark" in the March issue. I thought it was a fair and realistic comparison of Windows and Unix performance issues.

However, the article didn't adequately examine how EDA workstations are used. Although Pentium machines are less expensive and almost as fast as the Ultra for many individual tasks, with a large ASIC design, these individual tasks take only 20 to 60 minutes. Using shell, Perl, and native tool scripts, I can start a set of tasks that I know will take several weeks to complete or regression simulations across multiple machines that will take many days to run. Because Unix and Sun hardware are reliable, it's unlikely that these extremely long processes will freeze, "blue-screen," or otherwise terminate abnormally.

On the other hand, I have yet to personally witness a Windows PC system that hasn't crashed or hung even when completing simple office automation tasks for which it was designed.

The reliability of Unix and Sun means I can have my twice-as-expensive Sun machine and just-as-expensive software running 24 hours a day, seven days a week, without wasting time rerunning tasks that didn't complete because of a crash or freeze.

I firmly believe that after considering the increased usage time of software licenses (24 X 7 versus 8 X 5) and the lost time of crashes and hang-ups under Windows, the overall cost of Unix is actually far lower.

Kenneth Ryan
Senior Engineer
Orbital Sciences Corp.
Fairchild Defense
Germantown, Md.

To the Editor:
Your benchmark article on running EDA applications on PCs versus Unix machines is excellent. However, the decision on which machines to buy should involve criteria beyond that of raw performance. I know that's the case for a significant majority of the hundreds of companies with whom I conducted market research over the past year.

I run focus groups in the fields of Windows NT­ and SPARC-based workstations, servers, and software. I do a lot in the EDA area, but I also work in financial markets and other areas where the debate on PCs versus Unix machines is raging. I conduct these groups with designers, MIS guys, system administrator types, and other corporate and government users in the United States, Europe, and Japan.

Stability . Although these participants tell me that PCs are catching up to Unix systems in performance, they have a long way to go in terms of stability. Your authors saw some of that in the benchmarking, with some unexplained crashes.

I can tell you that such experiences aren't at all unusual, and although some users may find that acceptable, it's a significant irritant for many users. When you move to the server level, especially when you're running mission-critical applications, the issue of PC instability makes them unacceptable to many. At least that was the opinion expressed by significant numbers in my focus groups.

Total cost of ownership. PCs are appealing because they're cheap to buy, and they run lots of applications. The price/performance of PCs is impressive; however, when you take into consideration the cost of support, the picture is not as attractive. There has been a ton of debate on the issue of total cost of ownership, and I'm amazed at the noise on this topic from the PC camp. If you talk to MIS guys who are responsible for keeping servers, networks, and the rest of the systems and infrastructure up and running on a daily basis, they think PCs don't have a lower total cost of ownership, primarily because of the higher support costs involved. The "fact" that PCs have a lower total cost of ownership is a myth according to many of the people I interview.

The large number of MIS guys I meet have had a ton of experience running PCs and Unix systems, and those are their conclusions.

Ross King
President
King Systems, Inc.
Pioneer, Calif.

James Lee replies:
As I said, I won't debate reliability (stability). Many readers have written regarding reliability, and there are people who will take one side (Unix) or the other (NT), as Rylan Luke does above. Unfortunately, both NT and Unix machines need an occasional reboot when things go wrong.

I agree with Kenneth Ryan that generally Unix hardware and software are more stable, although there are people who do experience months of uptime on NT systems. We also pointed out in the article that the Unix environment with shell scripts, batch queueing, and other built-in utilities makes EDA automation easier.

Still, NT is viable for running EDA tools. Certainly running jobs locally on your 300-MHz Pentium PC will be faster than running them on your SPARCstation 2, 10, or 20.

NT is currently on the desktop for office automation tasks. If design automation on the desktop can also be done with NT, it will simplify choosing a desktop computer. Still, I see Unix in the backroom for running the big jobs and regression tests.

If EDA vendors license their software in a way that allows me to float a license for a Unix tool to my PC tool, then my software investment would be safe. By day, I'd use the licenses on the desktop; by night, they'd run on the servers for the big jobs.

Regarding the total cost of ownership, Ross King makes an excellent point. Currently the hardware and OS are less expensive for NT, but the price of the EDA tools is about the same. The total cost of ownership of Windows NT, however, starts to climb dramatically when you start to look at the Unix shell scripts that were part of your EDA environment that need to be redone and you add in many third-party packages to fill in the holes in NT, like batch job queuing and automated nightly running of regression tests. You'll also need to add in Microsoft SMS and retrain your entire network support team.

In other words, the cost of supporting the Windows NT operating system for EDA has to include the cost of bringing the entire EDA environment along, not just the applications.

To voice an opinion on this or any Integrated System Design article, please email your message to miker@isdmag.com.

integrated system design  May 1998