Fog Creek Software
Discussion Board

Future of the heavyweights

A long time ago I participated in a discussion on MS vs Linux. Youve all been there I imagine. One poster divided the server market into three parts:
The lightweights, home and small buissness servers.
The midweights, about 100 - 10 000 users, depending on the application ofcourse.
And the heavyweights which refered to nationwide DBs, millitary simulation applications, very large company network servers etc aparently dominated by IBM and Sun.
I have basicly had no insight into this feild at all, at least not professionally. Id like to know what you think.

Does this sector still exist and is *nix still the default OS?

It was claimed that Windows had zero marketshare in this field, but I imagine that Linux isnt very common either. Will the current trend of making x86 clusters threaten the heavyweights at all?

(Btw, Is this a reasonable way to categorize the market?)

Eric DeBois
Tuesday, February 25, 2003

"It was claimed that Windows had zero marketshare in this field, but I imagine that Linux isnt very common either. "

Wrong and getting wrong-er every month.  I'm familiar with a few big money US govt institutions which have rapidly growing Linux deployment - as servers, computing clusters, and engineering workstations all.

And the response from "the trenches" is mostly that they can't believe Sun or SGI can stay in business for long.  The bang/buck relative to those offerings is no contest.

As for Windows?  Forget it.  Never was and never will be.

Tuesday, February 25, 2003

I don't have the expertise to answer your question, but will offer an observation:  Computing problems that were "heavyweight" a few years ago are, in many cases, easy problems to solve now.

For example, a state motor vehicles department.  5 million drivers.  Lets say 20 records per driver (registrations, violations, etc.).  That's 100 million records.  Throw in field offices with lousy WAN connections and this used to be an intractible computing problem.  Now, $100K of servers and a few above average developers can deliver a solution.

I think it comes down to trust and marketing.  The State of Oregon recently spent something like $100M on a new DMV system, which subsequently had to be scrapped.  This was done by "heavyweights".  An "easy" problem was made difficult.

Bottom line:  Capabilities (hardware, software, web services, etc.) are growing faster than the computing problems at hand.  Long term, the "heavyweights" will have to adapt or be forced into smaller and smaller niches.

Bill Carlson
Tuesday, February 25, 2003

Maybe you can merge lightweights and mediumweights.

When you see huge intel plateforms and tiny 1U Sun "farm servers", you can believe the previous distinction had its time.

Concerning heavyweights, I believe there is still room for them, simply because they are involved in big projects, made to last.

I've seen a big project in the banking area where IT maxi gurus tried to match the response time of a twenty years old HP system with a java multi-tier architecture and couldn't. So maybe replacement of very big systems will take some time.

To speak about OS, I'd like to warn once again about Linux. It's an excellent product, but it lacks some important features to become a 'solaris killer'.

When I see USB support instead of RAID support in the top priorities of the kernel roadmap, it makes me even more skeptic.

As for very old issues believed to be adressed, try "man NFS", look at the date and change requests.

I love Joel's article about Leaky Abstraction, to believe Linux is an achieved operating system led me to the kind of BIG leaks we all hate.

Ralph Chaléon
Tuesday, February 25, 2003

This week's cover story on Business Week is Linux.

The article is an interesting read, and touches on the heavy vs light computing.

Wednesday, February 26, 2003

The businessweek interview with Scott McNeily sadly misses a an open goal:

McNeily [on IBM's Linux play]: "But then when the majority of your revenue is now service-based, what would be your incentive to make your product offerings less complicated to use? None! "
Tell us Scott: "When the majority of your revenue is based on a proprietary high margin hardware platform, what would be your incentive to make software easlily portable to equivalent commodity hardware? None! So what is the real deal with Java?"

Just me (Sir to you)
Wednesday, February 26, 2003


I'm not sure what you mean.  Linux has had good raid and USB support for a long time.  Linux even has software raid, and a now has a journaling file system. 

Linux certainly competes against solaris in the 1 to 2u server market. 

linux will win in the server custers for a few reasons.

1) it is way more malleable than Windows for custom installations.  Windows is very difficult to customize for your needs.  Linux can be made to do just about anything.  For instance, ask yourself why a headless system needs a GUI.  Windows forces you to have one.

2) It is far easier to manage large linux custers than NT.

3) Linux is cheap.

Just image Google's licensing costs for its 10,000 server cluster.

web service admin/developer
Wednesday, February 26, 2003

In my opinion form an OS core tech point of view systems such as Solaris 8 or Windows 2K are  "more advanced" than Linux. But for most real-world application demands, the Linux kernel is often "good enough".
The heavyweights are in a red queen race against the commoditization horizon. As a result the R&D effort required to keep up has to be spread over fewer and fewer applications. This increases unit cost, and this has a secondary negative feedback on application potential.
Current heavyweights need to find either new applications or transform into some other kind of play (services, commodity, ...).

Just me (Sir to you)
Thursday, February 27, 2003

*  Recent Topics

*  Fog Creek Home