What’s at fault in that case is not the OS, but the GNU-based, assumption-laden, open, source code which is flawed in exactly the same way as programs that were written, more than 20 years ago, for only one kind of proprietary, Unix system.

When you say

” it will take a long time for OpenSolaris to gain comparable inertia and an end-user following”

I think you mean “momentum”? Sorry, but it was bugging me. Otherwise good article.

Second, testing Solaris against RHEL last year, it was quickly apparent that standard apps like Apache, nginx, etc., performed horribly without some serious tweaking (Sun engineers spent a month and a half to get it to match performance). Most of the binaries at the time were 32-bit as well, and nothing gnu-based performed acceptably (or, in many cases, failed to even compile).

Third, and I doubt they’ve fixed it as Sun is stubbornly opposed to local filesystems, in order to get Solaris working we had to removed the RAID card physically. Then, we had to remove the fiber ether card because it wasn’t supported, either. Neither card was an unknown brand — a 3Ware RAID and a Tigon card, both of which several varieties of Linux picked up on without a problem.

Fourth, there’s Sun even being involved. We purchased an X4600 with the promise of a super-performing machine that they would support Linux on. Their engineers didn’t know the first thing about RHEL (and in point of fact I had to help them find crucial logfiles in Solaris when they debugged). The X4600 came with Solaris, crashed completely on first boot and crashed again multiple times. The X4600 was easily outperformed by a 32-bit Slackware box and even got beaten in one filesystem test against an NFS mount. Sun talks the talk about open source and Web 2.0 but doesn’t have the first idea about where to step without breaking eggs.

I did try the FUSE-based zfs port to RHEL and it was an unqualified disaster.

It’s almost an aside but these experiences have given me the impression that the “Enterprise class” solutions are all about turning Linux into Windows for organizations without the technical skill to use the OS. RHEL, for example, relies on the 32-bit-ish ext3 and offers no options where xfs/jfs easily outperform it; a 64-bit compile of Slackware (Slamd64) is a very unforgiving Linux which outperforms RHEL 4/5 in some cases 3:1. IMHO the community could do wonders for itself by honoring standard kernels and supporting compiles like that which offer lots of power with minimal overhead.

-d

Is that like being kind of pregnant?