October 6, 2004

SysAdmin to SysAdmin: Sun is setting

Author: Preston St. Pierre

As a systems administrator, I try to keep up with as many OS variants as I can. I started my career on Sun's Solaris platform, since most of my work early on was database-related, and almost all of the installations I ever saw were on Solaris. I bumped into AIX, HP-UX, and others along
the way, but generally these left me wanting to be back on a Sun box. It is because of this longtime attachment that my heart is heavy at the thought of Sun just disappearing one day -- but I can't imagine how the company will climb out of the ditch that it continues to dig, both technically and philosophically.

In the most recent shade of Sun's rose-colored glasses, Linux is the arch
rival, Microsoft is an ally, and Red Hat is evil. Oh yeah, and Red Hat is
the only Linux distribution in existence.

Some background

A recent story on CNET News.com talks about Sun's plans for a comeback from the abyss it has fallen into, partly due to major vendor support for Linux. The article notes that the number one item on Sun's to-do list is to "make the
argument that Linux equals Red Hat." This is not a direct quote. However, in a meeting at the Massachussetts Telecommunications Council, Scott McNealy was quoted as saying "We love Linux..." "...we just don't love Red Hat." This to me sounds like Sun is doing anything but equating Linux with Red Hat, but it's clear that Red Hat is in Sun's crosshairs one way or another. To the extent that Linux may be perceived as "equaling" Red Hat by the PHB community, this may be seen as an effective ploy. It's not, but I'll get to that.

The same article says that, while most Sun computers run Solaris, many organizations have moved to Linux on their Sun hardware to save money. Whether this assertion came from another Sun quote or the media disease that causes journalists to lose any faculties related to independent thought
is unclear. However, later in the article McNealy does clearly seem to be addressing (weakly) the price-point argument, saying that, since Red Hat charges out the wazoo for support and services, you can run Solaris for "20 to 30 percent of the cost of 'free.'" In some sense, he may be right on the money here (pun intended), but it doesn't matter. I'll get to that as well.

In addition to these arguments, Sun's actions with regard to the Intel platform in general speak volumes in defense of a rather widely held opinion that Sun views the platform as, essentially, a second-class, "low end" platform. This has historically made sense, since Sun spends a great deal of money on development of its own SPARC platform, and supporting any OS on a second platform can only cost it money.
Releases of Solaris on Intel have pleased a few, but have been a disappointment in terms of revenue, presumably because, in my own experience, they were technically disappointing. Though some of the problems here may be due to the companies' historical inability to play nice, Sun will lose any battle when its plan is founded in a belief that the SPARC platform will be
a dominant one in datacenters come 2010. There are reasons for this, but this is still background -- so keep reading.

Why nothing Sun execs say means anything

Let's talk about Sun's assertion that Linux equals Red Hat. This is a grave error. There are many, many, many systems folks out there with buying influence who no longer see a clear advantage in running Red Hat over SUSE Linux. Vendor adoption is almost identical between the two distributions these days, and both Linux variants have, or are about to have, roughly the sameofferings and very similaradministrative tools. Add to this mix the fact that pricing between the two is highly competitive, and that Red Hat no longer has the benefit of being the only viable U.S.-based distribution, and I think you can make a strong argument that the provider who is to become the next Microsoft of the Linux world is still up in the air.

The above argument does not even account for possible further consolidation, similar to what we've seen from the recent Novell purchases. When was the last time anyone had a look at all the work Mandrakesoft has been quietly amassing in France? Clustering, "enterprise" everything, even portable appliances running the Mandrake desktop system! I have yet to try any of these, but the little I've seen looks impressive. I have no idea if Mandrake is for sale or if anyone is looking, but can we consider it "out of the question" in a post-Utah-company-takes-over-German-powerhouse era? I think not.

Next, let's address pricing. Solaris, running on Sun hardware, with full support for both, is more expensive in every instance than a Dell running any version of Linux. Even so, it wouldn't matter if the pricing scenario were exactly as Sun claims. Sun has failed to attract buyers to its "low-end" Intel offerings. At the same time, it has wrongly perceived the problem as being mostly an OS problem. The problem really is the movement among IT executives to relieve themselves of a vendor lock-in scenario that runs straight through the application layer all the way to the power cord.

While one could argue that moving to Linux on Intel is just another form of vendor lock-in, at least the Intel platform gives them a bit more flexibility of movement at the OS level than SPARC would -- or PA-RISC or any other proprietary hardware platform, for that matter. Sun seems to be the only one of the large Unix vendors to refuse to accept this fact. IBM and HP both have significant offerings in the area of Linux servers (high and low end) running on Intel hardware. Sun seems to have recently begun to see the light, but will it last? Is it viable? Is it too little too late?

Finally, Linux is open source. Sun has said that Solaris 10 may become open source, but then again we've heard Sun use the term "open" before. While it's probably true that many users don't care to muck about with the source code, larger organizations do hire people who are professional "code muckers" to fix a problem, write a driver, or add a feature. This is yet another form of flexibility offered by Linux.

Where's the accountability?

Quotes from Sun executives that have appeared in several publications paint the company as
viewing this issue as mostly (if not entirely) an OS war, though hardware isn't left out in the cold either. The problem with this plan, and the reason I think Sun will fall on its face, is that Sun completely fails to account or take responsibility for things it has been consistently bad at, which, in part, made people look to Linux in the first place.

Sun is so focused on fighting Linux and IBM that it has forgotten to fix the problems in its own platform, and so this whole campaign looks like nothing more than a marketing scheme which will cost all kinds of money that would be better spent fixing the problems that clearly exist internally and crop up as flaws in the OS.

What problems? Well, to give an illustrative example, I've spoken to more than one Sun engineer about what I view as major shortcomings in the Solaris LDAP client implementation. When I brought these issues up to Sun consulting engineers, the typical answer I get is "we have multiple customers requesting the same things as you, for the same reasons as you, and it never gets past the lab coats." That is an enormous problem.

Today there's a viable alternative to Solaris that has an added benefit of being an OS that can be run pretty much everywhere at some point. It'll start at the network edge, work its way into core services, and someday reach the enterprise desktop. It's already doingthis, so if you're thinking I'm a zealot for thinking this way, it's due to the fact that I'm fanatical about keeping up with news in this area.

Sun also fails to take into account the fact that Linux got where it is partially via an underground movement, and partially because Linux makes simple things simple, and hard things doable. Sun does neither. Ever run an strace on a simple getent passwd username under Solaris? Compare that to the output from a Linux machine of your choosing. 85% of the time, this isn't really likely to matter, but it illustrates the difference in complexity of the two systems, and during the other 15% of the troubleshooting occurrences, things are a bit more simple under Linux. Period. There's just no comparison.

Other Solaris annoyances

Another illustrative example: do you know what the Sun-recommended method is for changing the name of a machine? One word: "reinstall." Now, in Sun's defense, this isn't absolutely ludicrous if the machine is running all kinds of services that use a hostname in their config files. But the company recommends this even for workstations, which really do nothing in the way of offering services to other machines on the network! On a Linux box, this isn't (to my knowledge) made super-easy for you, but you can absolutely do it, and have the machine act in a predictable manner afterward. On the other hand, I once took a Sun system administration course, given through a Sun-certified learning center, with a teacher who was a Solaris field engineering consultant veteran with more than 10 years of experience. He had tried multiple times to change a hostname on a Solaris machine. Every time something went wrong. Not to mention the fact that you have to change somewhere in the neighborhood of eight files, and still there's no guarantee. Some of those files are ones that nobody understands what they do.

Solaris needs to change. Mostly, it needs to be fixed. Did you know that if any user in a Solaris environment belongs to more than 15 groups, the entire system becomes unusable? I haven't attempted to do this in Solaris 9, but in Solaris 8 this was definitely an issue. Did you know that a NIS netgroup has a size limit that forces many installations to have really messy nested netgroups? Further, did you know that if you put one machine in two netgroups, and then give mount privileges to both netgroups in your NFS configuration, your NFS server will die?

But there are changes needed, too, in addition to fixes. CDE, Sun's graphical desktop environment, is disgusting. Attempts to port GNOME to Sun have, so far, failed badly. The whole idea that fonts are terrible in CDE seems to be mandated by Sun. The intrusive registration nonsense that requires a support call to fix, the Netscape browser that just about everyone has grown tired of, and the lack of choices in preinstalled applications are not only all things that Solaris users have been griping about for 10 years, they're things that Linux easily fixed, for free.

Why Sun will eventually go away

Sun will eventually go away because its approach to battling Linux consists mainly of equating "Linux" with "Red Hat" to clarify the target. Unfortunately, Novell has bought SUSE, and by
some estimations, SUSE has every bit the shot Red Hat does of becoming the Microsoft of the Linux world. In addition, I can't imagine that Mandrake won't get bought out eventually. I think that, speaking strictly of the technology, Mandrake is a worthy target for acquisition, and a worthy opponent to both SUSE and Red Hat. If it goes, there goes a bearing wall in Sun's so-called structured approach. On the other hand, if Sun bought Mandrake, it'd be better off.

Sun, like Microsoft, is about getting more proprietary in a world that is getting more open, while calling itself "open". Ironic. Sun will eventually come to understand all of this. It will eventually understand that people wouldn't have moved to Linux if Solaris worked properly. It will eventually understand that just because it's called "Sun" doesn't mean the rest of the universe revolves around it. And yes, Sun will eventually go away.

Click Here!