The Linux Religious Wars

May the customers win

A few months ago, I read an article in Linux Journal ("The Great Linux Revolt of 1998," October 1998) that showed a picture of enthusiastic Linux users huddled around a trash-can fire outside a computer store. They were protesting the launch of Windows 98 and holding up signs proclaiming, "Linux Is Free." They had even pasted a "Linux Is Free" sign over the Microsoft billboard inside the store. Users demonstrating their advocacy for an OS—now that's enthusiasm.

When we launched Windows NT Magazine, readers asked us to advocate Windows NT. At that time, Microsoft spent most of its marketing resources pushing Windows 95 and barely mentioned NT. Readers wanted us to stand up for NT. Six months later, readers told us to stop advocating NT: "NT is doing pretty well on its own; stop telling people how great it is." The market had matured enough that users didn't need to be defensive. The time had come to focus on using NT.

Today, our philosophy is this: NT exists. We tell you what works, what doesn't work, and how to avoid pitfalls. We simply want to help NT professionals get their jobs done.

Understanding the Big Picture
Recently, we started writing about Linux in Windows NT Magazine and our email newsletter UPDATE. In the April magazine, Mark Russinovich looked at Linux 2.2's symmetric multiprocessing (SMP) capabilities. He explained in detail how Linux works internally and how its inner workings can affect performance and scalability.

As I write this editorial, the April issue hasn't reached readers yet, but I predict that Linux users will flame Mark as an anti-Linux NT bigot. What those people don't know is that OS experts in IBM's Thomas Watson Research Center reviewed Mark's article. None of these experts are NT bigots. They found his article to be technically accurate.

More important, Russinovich examines Linux in the same way that he has been scrutinizing NT for the past 4 years. In fact, all Windows NT Magazine writers put NT under the microscope, scrutinizing its scalability, availability, security, performance, compatibility, and ease of use. We examine and test vendor claims. If NT doesn't scale or perform as claimed, we tell you. However, we don't whine about NT's shortcomings. We provide solid workarounds that you can use to get your job done.

When people are part of something new and exciting, they sometimes ignore shortcomings and become overly zealous. Some Linux users pounce on any chance to get back at the evil empire—Microsoft. Other Linux users want to remove potential obstacles to Linux's success.

I understand this enthusiasm: We just launched SQL Server Magazine, which targets DBAs, developers, and data-warehouse architects. The release of SQL Server 7.0 has created new enthusiasm for the product. The SQL Server community is buzzing, and we see the lines forming. Fans of competing databases, such as Oracle, are saying, "My database is better than yours." SQL Server 7.0 has a lot going for it, but SQL Server Magazine will point out areas that need improvement and tell you how to deal with them.

Seeing Both Sides
I recently spoke with the CEO of LinkShare (http://www.linkshare.com), an Internet transaction company that switched from Oracle on NT Server to Oracle on Linux to avoid having to reboot its NT machines daily. Now LinkShare pumps thousands of transactions through its Linux Oracle boxes, 24 X 7, with complete stability. In fact, the company achieved higher performance on single-processor Linux machines than on dual-processor NT Server machines.

This story represents a big win for Linux, but LinkShare has to use multiple single-processor machines because Linux can't scale beyond one processor. (This limitation didn't apply to NT Oracle servers.) Fortunately, you can usually load balance Internet applications so that many servers can handle a single application.

Linux has a lot to be enthusiastic about: market buzz, loyal users, and support from top-name hardware vendors (e.g., Dell, Compaq, and IBM) and key software vendors (e.g., Oracle). Perhaps the market is now mature enough for Linux supporters to move beyond religious-fervor and look at the OS objectively. Can Linux solve real business problems? If so, which ones?

If LinkShare's experience is typical, Linux can make a dent in an area where NT used to shine—reliability. NT used to have a reputation as the most reliable OS for the PC. Perhaps Linux can usurp that position and hit Microsoft where it hurts.

Long live Linux. Long live NT. And may the customers win.

End of Article

I don’t believe what I’m reading in Mark Smith’s May editorial, “The Linux Religious Wars.” Linux doesn’t do SMP? Linux has been able to do SMP for about 3 years! True, the Linux 2.0.x kernels weren’t very good at SMP performance because only the boot processor could handle interrupts, but the Linux 2.2.x kernels are more than capable of handling SMP. I recently ran a program that calculates the value of Pi (p) on my SMP Linux box. When I ran this program with two threads, it ran about 198 percent faster than when I used only one thread: Linux created two threads, executed the code, and completed in 0.0029 seconds. Try to get Windows NT to do that. 
--Chris Wade 

I obviously could have communicated better in my editorial. No question—–Linux runs on multiple CPUs. The company I contacted, LinkShare, couldn’t get Oracle on NT to scale effectively on more than one CPU. In the article, I should have defined effectively.
Years ago, Windows NT Magazine readers were excited to buy dual-CPU machines for NT-based CAD packages. Many of these machines weren’t multithreaded, so adding the second CPU produced only about a 10 percent gain in performance. Because you had to pay twice as much to get only a 10 percent performance increase, the magazine recommended that users forget the idea.
Apparently, the performance gain that LinkShare experienced for adding a second CPU wasn’t worth the cost. The solution wasn’t effective enough. Theoretically, LinkShare could have added 8 CPUs, but what performance gain would they realize for the cost? That’s the $64,000 question with Linux—–no published benchmarks exist to help people make that determination.
Your comparison of calculating p is interesting, but not relevant in this situation. Clearly, benchmarks on databases don’t scale linearly. For example, Microsoft SQL Server on NT will get three times the performance when you go from 1 to 4 CPUs. I’ve seen many compute-intensive benchmarks that will scale linearly on NT. Unfortunately, our readers don’t care about those kinds of applications. For the most part, our readers are trying to run messaging, database, Web-server, file-and-print, and line-of-business applications. You have to consider the performance gain from the OS and the application.
Several years ago, Microsoft claimed that Exchange Server scaled to 4 CPUs. Our Lab found that the application was optimal at 2 CPUs, and we reported that. Microsoft said SQL Server scaled to 8 CPUs. We said the application was optimal at 4 CPUs, but you could squeak out a few more percentage points with 6 processors. Microsoft has made some improvements, and we’re reporting those benchmarks. What are the optimal numbers for Oracle on Linux or Apache on Linux? That’s the kind of data people really want to see. IS managers don’t want to overbuy or underbuy systems; they’re looking for meaningful data. Sure, Linux can run on a 10-CPU or perhaps a 64-CPU system, but at what cost for what scalability and what application? Is running Linux worth it? If the Linux community won’t provide the information that IS managers need to make a business decision, who will? 
--Mark Smith

Chris Wade August 09, 1999

I read Mark Smith’s editorial in the May issue of Windows NT Magazine, and I must say that I heartily agree with him. Spending time building a religious fervor around OSs seems kind of silly. The point of the matter is whether the OS does its job well. Microsoft built its empire by producing a better, more accessible product. When another company comes along that can beat Microsoft, so be it. This industry was built on innovation, and having Linux around ensures that innovation will continue. I’m glad that Linux has come on the scene—–its presence will keep things from stagnating. 
--Randy G. Haas

Randy G. Haas August 09, 1999

The fact that many people in the Linux camp have taken a negative stance in dealing with NT or its advocates (or even its everyday users) is unfortunate. More than SMP, scalability, or applications, Linux most needs to evolve its community. The Linux community must be able to remain diverse, but some of its members must also learn that all diversity—–not just our diversity—–is OK. Articles in respectable, widely read publications such as Windows NT Magazine that offer an olive branch of sorts are a necessary step in this evolution. Thank you for providing evidence that the voice of reason will inevitably prevail over shortsightedness. 
--Bill Garrett

Bill Garrett August 09, 1999

We have a consultancy in Brisbane, Australia and have been installing using NT since 3.51. We have several in-house NT servers but we have started to investigate Linux and now have 2 Linux servers and an Internet gateway. We have found that Linux is every bit as stable as the claims made of it and most importantly from our perspective, you can actually see what is going on when there is a problem. Over the years we have spent many, many hours pulling our collective hair out trying to fault-find NT problems with it's anonymous Windows and meaningless error messages. That, coupled with the need to re-boot the server every time that you do anything has made it both a frustrating and costly platform to support. I don't see how Bill Gates' claims of low TCO can be justified.

That said, NT has served us well but we have now installed 4 Red Hat servers for customers, one of which replaced an existing NT server and we have been very impressed with the experinece of installing them. They are easy to install and configure - about on-par NT I would say - very easy to fault-find as all kernel and system-wide error messages are available and you can perform an almost endless set of tests on any of the Linux modules and see exactly what is happening. You are not limited by Windows or X-Windows as you can use the character interface to analyse the system. It also has 6 virtual screens plus the Windows screen which can contain many windows just like NT. This is very helpful and speeds up installation and configuration and diagnosis by an incredible amount. And when running the character interface, the system performance is startling, leaving NT or X-Windows in the dust.

Another extremely useful built-in facility is that you can Telnet in to the server from any other TCP/IP appliance connected to the network, including the Internet. This again provides very easy maintenance of the system from almost anywhere and we can very easily administer the system remotely from the Internet.

The development cycle for Linux is very short too, with bug-fixes or updates available daily and in the past year or so we have had Red Hat 5.2, 6.0 and now 6.1 and each one has been a significant jump from the previous one, with more features and improvements each time.

While NT will probably still have a valid place in the marketplace for quite some time, I welcome Linux as a worthy competitor and am quite excited about it's future. While there are zealots on both sides of the fence, the best and most likely outcome is that in the long run the end-user will benefit the most.

Regards,
Brad

Brad Boundy October 23, 1999

You must log on before posting a comment.

If you don't have a username & password, please register now.