neorants

Check this out. Or don’t, I’ll give you the synopsis: Cringely thinks Sun will die within 5 years. I can only hope as much. Sun’s bread and butter nowadays appears to be in selling overpriced hardware to companies that have been dealing with them too long to realize there are better options out there now, and the companies that are sucked in by a me-too mentality.

That’s to say nothing of the OS. Here is an OS grounded in the very roots of UNIX — it’s particularly evident, because it feels like they never bothered to take it much further, aside from supporting new hardware. Here we have a company that started a campaign to promote the idea that “the network is the computer,” yet their OS doesn’t allow the user to set a default route during the install. They only recently started supporting DNS resolution out of the box, formerly requiring users to specify “other” as their name resolution choice and configure DNS resolution manually if they chose to forego using NIS (which, incidentally, they are now dropping in favor of LDAP).

The OS doesn’t ship with a compiler. Want to compile your own software? Download and install gcc or pay $1000.00 for a license to their development software, Forte, which is the only reliable option if you plan to generate 64-bit executables. You don’t care about 64-bit executables, you say? A free compiler is fine? Awesome! Then, run into a roadblock when your copy of Apache on Solaris won’t support more than 255 open filehandles for logging at a time, because they never bothered to update the 32-bit stdio library to fix the limitation, and you have to go 64-bit to fix it.

I have seen quotes online stating that Solaris doesn’t come into its own until you have around 16 processors in a box. That must be the case, since through my own independent observation, I can get better performance out of a single 1GHz Intel processor box running FreeBSD than I can out of a quad-processor Sparc box. RISC architecture benefits, my butt. Of course, the fact that they refuse to optimize such critical utilities as “grep” on the system (compare GNU grep’s performance to the standard Solaris grep, and be amazed) may be indicative of missed opportunities to optimize elsewhere in their software.

The point is, in an age where scalability and reliability are key, and clustering is becoming increasingly easy to pull off (why, yes, the network IS the computer, thanks for reminding us!) why would I want to attempt to get that scalability and reliability in a single piece of overpriced iron when I could buy many more boxes built with commodity x86 hardware, any single one of which would outperform a box costing 10 times as much from Sun for my purposes, and load balance them with some kind of backend storage device?

Not to mention, I could stop having to edit /etc/defaultrouter on every single box I install.