Columns
Is Linux ready for the enterprise?
Quick, tell me the name of the hottest OS around. Did you say Windows
2000? Sorry, that's old news. Every IT department knows they will use
it, but no one is all that excited about it. Linux is now the media darling
of the moment. If you doubt it's the hottest OS around, you need only
look at Red Hat's recent IPO to see the effect media buzz has on a stock
price.
There is a meaningful basis for this interest in Linux. International
Data Corp. (IDC) estimates that server usage of Linux grew 212% in 1998.
And 30% of all Web pages on the Internet are served up by Linux systems.
IDC also estimates that 17% of all servers run Linux. These figures alone
tell us we should keep an eye on this OS, but it's not this data creating
the media excitement.
What has the media all a-tingle is speculation that Linux could replace
Windows NT as a server platform for the enterprise. In slightly more hushed
tones, they speculate that Linux could even replace Windows on the desktop.
These considerations are based on the claim that Linux is more stable
and therefore more reliable than NT. Proponents also claim that Linux
is speedier, as well as a cost-effective solution because it is free.
Speculation about Linux on the desktop is further fueled by work on the
KDE and Gnome desktop environments.
But are these claims justified? I decided a few months ago that a good
way to evaluate these claims was to set up my own Linux system and network.
My plan was to tie in my Windows NT and 95 boxes and evaluate how much
work it took to get the network running. This type of approach is fairly
typical for an IT skunk works project. I believed my experience would
duplicate that of an IT department testing the waters with Linux.
One of the intriguing features of Linux is its ability to run quite
well on lower powered machines. Since I was paying for this experiment,
my wallet was thrilled by this possibility. I had an old 486 machine at
home that was working well as a paperweight. I threw away everything except
the case and power supply, and rebuilt the system as a 350MHz Pentium
II system. This cost less than buying the equivalent box and it let me
make sure that all of the hardware was on the Linux compatibility list.
I also chose to make this a Linux-only machine. There was no need to complicate
things with dual booting.
When Linux Expo was in town this summer, I bought the CDs for several
flavors of Linux. However, because this was my first Linux installation,
I decided to purchase the commercial release of Red Hat 6.0 so I could
get support if needed. How prescient.
The initial start-up of my newly refurbished machine went fine. I was
able to set up the BIOS correctly and eagerly began my Linux install.
I was well prepared. I had my list of hardware and network information.
I had the installation guide that came with the software. I also had two
other books I'd purchased on Red Hat Linux.
One nice feature of Red Hat Linux is the choice of install options.
Since I planned for this box to be at the heart of my network, I picked
the server installation. This let me avoid dual-boot issues and partitioned
the whole hard disk for Linux. The installation process was not as easy
as it could be, however. I needed to spend time reading and understanding
the implications of the choices as I worked my way through the install
script.
I first went to the linuxconf tool to begin setting up accounts. This
is a great tool, and I much prefer it to the command line approach. As
I explored my system, I found that it was not talking with all of my hardware.
Linux thought it was working with my Ethernet card, but it wasn't. I also
discovered that since I had selected a server installation, the system
hadn't loaded any modules to talk to my sound card. In addition, there
was more configuration work needed to get Linux to talk to the modem.
Clearly, I wasn't as far along in creating a usable system as I had hoped.
One way to get Linux to talk to all of the hardware was to recompile
the kernel with the appropriate modules included. In trying this, the
recompile kept crashing with a signal 11 error. I used Red Hat's online
support tool to report my problem and then decided to try phone support.
The person I spoke with replied in a very nasty tone that they don't deal
with recompiling the kernel. The written reply to my online submission
was essentially the same, with the added suggestion that I search the
online docs for potential solutions. I did just that.
I decided to solve the recompilation problem by doing a complete custom
reinstall and load all modules. This time the system crashed at different
points partway through the install with a signal 11 error. Five days after
I reported the problem, support told me that signal 11 errors are often
caused by memory problems -- hardware that works fine with Windows may
not work well with Linux because Linux stresses the hardware more. I replaced
the system memory and was able to complete the installation.
You would think that would have been the end of my problems, but it
wasn't. Linux found address conflicts between my network and sound cards.
This time, it took over a month to get a useful answer. Most of that time
was spent in limbo because of problems with Red Hat's ticketing system.
The solution that finally came was to use Windows and Windows tools to
reset the Ethernet card.
But these weren't the only problems I encountered. Other advice left
my system unusable and in need of another install. All in all, it was
an insightful experiment -- it certainly shaped my view of Linux's place
in the enterprise.
So, if the claims of Linux's speed, reliability and stability are true,
does my experiment show the OS as ready for the enterprise? I don't think
so. Linux takes too much work to create a functional system that talks
to all of your hardware. I know NT is no piece of cake either, but I've
been able to get NT up and running with much less work. Right now, this
difference means more IT labor and thus more money to deploy Linux. And
this doesn't even address the additional training costs to get your staff
the right skills.
Is your current hardware Linux compatible? If not, you may have a very
large expense in moving to Linux. If you purchase your equipment, it may
be many years before you can afford to replace it. If you lease your equipment,
you may have to wait three or four years before you can change. And being
Linux-compatible may cost you more when you consider the total cost of
ownership.
Do you need timely support? Almost all vendors offer a support contract.
Unfortunately, that's not usually relevant when you are doing a skunk
works project. IT organizations don't typically spend money on support
for experimental technology. If the standard installation support is inadequate,
you may have to bring in a consultant or ditch the technology.
What about Linux on the desktop? Sorry, it's not ready. The KDE and
Gnome environments are fine to work in, but the dearth of office applications
that run in these environments makes them problematic. Basic replacements
for Microsoft Office are not enough. Most corporations use a wide variety
of tools on the desktop. There's a long way to go in this area.
Does all this mean Linux has no place in the enterprise? No. But I do
believe that there are several things Linux vendors must do in order to
succeed in the enterprise market. First, they need to make the product
much easier to install and configure. Until that is done, they will only
make limited inroads into the enterprise.
Vendors also need to provide better support for getting a basic install
up and running with all of a system's hardware. Good installation support
is fundamental to making Linux useful.
Finally, vendors need to support a wider range of hardware. I know Linux
is available on many platforms, and I applaud the developers who made
this happen. I'm talking about a narrower problem. It is very common to
hear about hardware incompatibility problems when installing on the Intel
platform. Until Linux can easily use all the same hardware used by Windows,
the job is not done.
Is there hope for Linux in the enterprise? There certainly is. And if
vendors address these important issues, Linux may find a home supporting
critical applications. Write me and tell me about your experiences with
Linux in the enterprise.
About the Author
John D. Williams is a contributor to Application Development Trends. He is president of Blue Mountain Commerce, a Cary, N.C.-based consulting firm specializing in enterprise, domain and application architectures. He can be reached via e-mail at [email protected].