Link to Microsoft "Missing the Point," Says IBM Program Director

By Gregory Leake
Group Product Manager, Microsoft Corporation

I am writing about your article titled "IBM Responds to Latest Microsoft Jabs at WebSphere 4.0." In this article, you interview IBM's Stefan Van Overtveldt, program director, WebSphere Technical Marketing.

In the interest of facts and truth, I would like to correct Mr. Van Overtveldt on a couple of points, and present a Microsoft response to his interview and the article.

First, I want to point out that the .NET Pet Shop comparison to the Java Pet Store is separate from the Web Services comparison between .NET and IBM WebSphere 4.0. Both comparisons can be found at http://www.gotdotnet.com/team/compare. The article and the interview discuss both comparisons as if they are the same.

So let's separate them and look at the bulk of what the Mr. Van Overtveldt says about each.

1. With regard to the .NET Pet Shop comparison, he says that:
"PetStore.com is an application that has been written to allow J2EE application vendors to test if all of the J2EE APIs are actually present and functioning well in their J2EE application server. That's the only objective that PetStore.com ever had…. this is an application that is not well written at all. It's not well written for performance. It's not well written for security…. Taking this application and pretending that it's a real live application, and that the customers can draw conclusions with regards to productivity, performance, etc., is missing the point about what this application was intended to be."

This is wrong. The Java Pet Store (Petstore.com) is held up by Sun Microsystems as a primary blueprint application for J2EE that illustrates "best practice architecture" and "best coding practices…for enterprise applications." It has an entire Sun blueprint Web site dedicated to it as a sample application for customers to follow. Sun has even published a Sun Blueprint-series book all about the Pet Store as a design pattern for enterprise applications, and has been telling enterprise developers to follow it for building scalable, reliable applications since Java One in May 2001. Furthermore, IBM appeared on stage with Sun and other J2EE vendors to endorse the Pet Store application at Java One in May 2001, and even demonstrated it running in WebSphere 4.0 in front of thousands of developers as a best practice enterprise application. And furthermore, they ship it in the IBM WebSphere 4.0 product as a sample application for developers to follow. So it is very disingenuous for IBM to come out now and say it is a bad application and should not be used to compare with .NET. They are all of a sudden singing a very different tune based on the release of the .NET Pet Shop. Sun has also now come out and said that the design pattern is valid, but it is not designed to be high performance. In response, we find it a highly questionable customer practice to publish a "best practice enterprise design pattern" and not ensure it will result in high performance applications. In short, if this is the case, shame on Sun for publishing it and promoting it as an enterprise design pattern to begin with. They should either re-write it, pull it from the Web and their blueprint series published books, or stand behind it. As for the benchmark of the Pet Store application, I would like to point out that the benchmark was originally conducted by Oracle as part of their Java Performance Challenge. So in this way a major J2EE vendor invited the comparison. In the end, I think the comparison is quite valid, and the competition is healthy for customers and the various vendors involved. Right now, the .NET Pet Shop remains uncontested as far as published versions of the Pet Store go, with 1/4 the amount of code required to build vs. the Java version, and offering 28 times better performance and over 8 times better scalability. We are very happy to meet IBM, Oracle, or other J2EE vendors in an independent shootout for verification of the performance results.

2. With regard to our .NET vs. IBM Web Services comparison, the bulk of the interview seems to focus on the license cost comparison. Here again, Mr. Van Overtveldt is misleading customers, since he claims that our cost comparison does not take into account client access fees associated with the deployment. I would like to point out that we fully take into account client access licenses in the cost comparison, and that he is simply misinformed on this topic. Mr. Van Overtveldt claims a 4-server WebSphere deployment would cost a fixed amount no matter how many clients access the WebSphere server. He states that the MS cost, however, may be lower out of the gate, but would be higher and grow depending on the number of clients accessing the site. This, he claims, is because the client access licenses required for Windows 2000 Server drive up the cost for .NET as more users connect. This is wrong. In fact, we have included a Windows 2000 Server Internet Connector License in our cost calculation, which includes *unlimited* authenticated client access to the .NET Web Service in question. So we fully stand behind our cost comparison, no matter how many users are connected and using the Web Service. For a 4-server deployment with 8 CPUs per server, IBM WebSphere 4.0 would cost $12,000 per CPU or a total of $384,000. Microsoft .NET would cost $3,999 (W2K Advanced Server) + $1,999 (Internet Connector License) = $5,998 per server for a total of $23,992, no matter how many clients use the W2K site. Mr. Van Overtveldt should make a point of understanding the MS licensing cost before misrepresenting it, and IBM should correct their public document because it is wrong. IBM WebSphere 4.0 costs 15 times more than Microsoft .NET in the example analyzed, which is very typical of a real customer deployment configuration in the enterprise.