Copyright © 1996 iX Multiuser Multitasking Magazin
Courtesy of iX Multiuser Multitasking Magazin!
People keep asking me ``When is Linux going to be ready for the commercial market''. I guess the first thing to discuss is what is meant by "commercial" in this context.
Some CD-ROM vendors have put the word "commercial" in their name, only to have the technical people think their product is good only for use by banks and insurance companies. Other people look at their product with disdain and say that ``Linux is not commercial quality'', because it is missing some feature they need, or they feel it is unstable (usually without ever trying it even one time).
To me, the word ``commercial'' has lost as much meaning in the marketplace as some of the other buzzwords:
In the old days of computing the commercial market was banks, insurance companies and business-oriented facilities where the use of COBOL or RPG was the mainstream. The technical market was scientific, engineering and manufacturing where FORTRAN and assembly language was used. Somewhere along the way the term ``commercial'' seems to have gotten twisted around to mean ``ready for the mass market'', versus ``ready for hackers''.
For the purpose of this article I will take the second meaning, and address whether I think Linux is ready for commercial purposes rather than the hobbyist and hacker market, and ready for the mass market rather than limited markets.
For those of you who hate reading long articles, or who are short on time, let me give you my conclusion right now. Then you can go out and drink beer or other fun activities:
``Yes, Linux is ready for the commercial market...in some cases''.
In order for an operating system to be ready for the mass market it must have several attributes:
But you can eliminate all of these considerations in today's mass market if only one thing is true:
You have lots of applications.
after all, there would not be 170,000,000 DOS systems in the world if any of the others had to be true.
I almost added that is has to be economical, but history has actually proven me wrong on that. If people added up the total cost of ownership, then Apple would certainly have won over the PC. But people ignore the human costs of someone else (or even themselves) beating their head against the wall trying to get something to work, or the system crashing repeatedly, or the fact that the one keystroke they can hit the easiest (through practice) is
<CTRL><ALT><DEL>
Now what causes this plethora of applications for an operating system? Ease of programming? Good software development tools? Features inside the operating system? Stability of the interfaces over time?
The answer is ``none of these''. While all these attributes may help convince an application developer to port, the one overriding issue is volume of the operating system platform. Again, if MSDOS were compared to MacOS, or even to UNIX and volume were not taken into account, we know which two operating systems would have the most applications, and they would not be from Microsoft.
While it is true that several Linux vendors are working on getting these applications for the mass market (read this " your mother and father"), the number of applications that run on Microsoft platforms have been estimated as high as 35,000. SunOS has an estimated 10,000 applications, with other `commercial UNIX'' systems (including Solaris 2.x) much lower in number. It will take the Linux vendors a long time to get the number of applications necessary to hit the really large mass market, particularly if they did not depend on iBCS2 and DOS/Windows compatibility (which could supply a fair number of current applications), but depended on ``native'' Linux applications.
So applications are king (and queen) for the mass market, and installed base (volume) or the promise of explosive growth (volume) is the key to these. But is the mass market the only ``commercial'' market? The answer is ``no''. The mass market is a subset (albeit very large one) of the commercial market. So let's look at what the rest of the commercial market needs. We will look at this by segmenting the market into:
When I speak of turnkey systems I typically mean a computer system that has one specific (or not so specific) application that runs on it. Examples of turnkey systems are point-of-sale terminals, reservation systems, CAD systems, etc. But in a larger sense, other applications such as Web servers, nameservers (such as BIND), etc. could also be considered ``turnkey'', since they have only a few necessary programs that have to run on the system.
Usually turnkey systems are ones that an Independent Software Vendor (ISV) or Value Added Reseller (VAR) will chose a hardware system, an operating system, port an application to it, then duplicate that system 500 to 1000 times without change to the basic application. These ISVs and VARs will try to chose the lowest cost solution to fit their customer's needs.
Linux is perfect for these types of applications. The operating system is stable enough for the developer to port their application and test the application fully. Once it is fully tested and stable, the entire package is ``frozen'' and duplicated any number of times for the end customer.
Since the operating system may be freely copied, and it runs on inexpensive hardware, their variable costs are minimal. Even a developer who is not familiar with the Linux system (so they need help getting it running on a platform) will quickly pay back the porting and system programmer costs they accrue by not paying $200-$500. per license for the operating system. Plus they have all the source code for the entire system, in case they run into trouble later on. You can buy a lot of Linux support for $200-$500K.
As I said before, I include Internet Service Providers (ISPs) as part of this ``turnkey'' environment, for both external internet and internal internet. Why overload your expensive, high-powered, highly complex general-purpose server to do Web serving when a smaller, simpler box can offload it? Why not run your NIS slaves on a Linux box? Or perhaps your BIND server?
In the early days of Digital UNIX (known then as DEC OSF/1) we did not have very many applications. In fact we had none. The marketing staff came to me with sad faces asking if it was possible to sell an operating system that had no applications. I invented a term called ``Turbocharging'', which allowed a Digital UNIX system using the speed and power of the Alpha processor (as well as the throughput of our networking devices) to offload NIS, NFS, BIND and other services from people's overloaded, slower SPARC machines. We also showed people how they could use the rsh(1) command to allow the Alpha to do a portion of their very CPU intensive processing while delivering the result back to the SPARCs on their desk. This allowed the SPARCs to work more on applications and less on the other ``system administration'' tasks that they were performing. We sold lots of Digital UNIX systems based solely on executing those tasks. Today, of course, Digital UNIX has a lot more applications, and particularly very large memory databases that are extremely fast. But the same principle applies. The database engine runs on the Alpha system, supplying data to the slower SPARC engines as a ``Turbocharger''. I could see Linux systems headed in the same direction, following the same path.
Very large customers often have their own home-grown applications which they need to deploy across a wide network of people. Or they can have management dictate a certain suite of applications, which then can be ported to Linux. Since these customers are so large, their operating system costs are huge, and utilizing the savings using the Linux operating system they may completely cover the expenses of porting their software.
Or these very large customers may ``influence'' their layered product providers to port to the Linux platform. Finally, they may even change some of their computing habits (to use existing programs) if the cost savings are enough to warrant it.
Companies like Caldera are creating a suite of applications and approaching these very large customers to show them the operating system savings that they can achieve if they switch to Linux. While it is true that every application the customer could ever conceive of running may not run on Linux, by using the native applications, the iBCS2 applications, the DOSEMU applications, and applications that run under WABI, a nice suite of applications could be built to solve their needs.
Finally there are what I call ``specialized markets''. Markets that might buy Linux simply because it is Linux, and not because of the application suites that it provides.
In the education field there are three main markets:
The administrative part is the ``business'' aspect of the market. They are looking for easy-to-use systems that can also handle complex administrative tasks that might cover a community the size of a small city.
The ``campus computing'' is the supply of computing power and service for majors of all types, web services and research into non-computer science (for example, molecular modeling) research.
Finally there is computer science education, both on the undergraduate and graduate level, as well as research into computer science.
While the administrative sub-market typically relies more on shrink-wrapped applications, the other two rely on them to a lesser extent (with the computer science education market relying the least). The other two markets can utilize a lot more of the freeware and shareware applications that are already ported to Linux. This gives them a very low-cost (from a software perspective) platform while allowing them to see and (often) modify the source code for the applications they use.
More importantly, in the computer science research area, the results of the research can be freely distributed to others working in the field, or even published as source code to illustrate the results. This can not be done with ``commercial'' operating systems.
Some universities are utilizing Linux more and more to run their campus. From a ``commercial'' standpoint, their needs are the same as many large businesses. Students graduating from college will know about Linux, and bring the word to their future employers.
Finally, there is the computer hobbyist and software developer market. I relate this market to the amateur radio market. In the amateur radio market the radio is often used to simply talk to other people, but at the same time the users investigate new ways of using radio, and improving it. Many electrical engineers started out as amateur radio users. So it can be with Linux, since for the first time both the prices of the hardware and the prices of the operating system source code are within the reach of mortal people.
In conclusion, I feel that Linux does have the items needed for several types of ``commercial'' uses:
What Linux really needs is for the ``commercial'' community to understand what is going on, and to embrace it where it will be useful. This will increase the volume numbers even more, which will attract more applications.
Along these lines I would like to ``advertise'' a joint effort of USENIX and Linux International to happen in January of 1997 in Anaheim, California of the United States. There will be a joint USENIX/Linux development conference, and while a certain part of the Linux conference will be oriented towards the development of the Linux operating system, the bulk of the conference will be oriented towards application developers and marketing people, to better understand the Linux operating system and how to sell their applications and services into the Linux market. We hope to show ISVs, VARs, resellers and distributors how they can make money by selling their applications and services on top of the Linux operating system.
Jon ``maddog'' Hall is a Senior Leader in the Digital Equipment Corporation UNIX group. He has been in the computer industry for twenty-five years, UNIX for sixteen years and has guided the emergence of six operating systems, including Alpha Linux. He has an MS in Computer Science.