JC - Let me frame this a little differently... A "plain vanilla" X86 machine has availability of less than 90%... the primary hardware failures which take systems down are disk and memory. Disk arrays eliminate the first, ECC memory largely eliminates the second. Next in line would be peripherals, and various schemes for hot swap and hot replacement of NICs and controller cards have taken that out of the equation. That gets us to the current hardware state of the art (excluding clustering), where hardware reliability exceeds 99.5%. At that point, application failures of one kind or another dominate the downtime events. The OS is still way down in the list of failure modalities. That's true for any of the "real" OS variants.
I can take an X86 server and set it up to achieve 99.99% reliability, running Solaris, running NT, running SCO Unix, running Linux... as long as it's not running any applications. So for basic file and print, any of those systems can be easily configured to "never go down" - or at least to have downtime in the minutes per year, scheduled or unscheduled.
None the less, the lower reliability in the real world of NT based systems is not FUD, it's real... but my contention is that this is not because of the quality of the OS. What then is causing the lower reliability?
Part of the answer is cultural - and that is not just SUNW but is also true of the very reliable MVS and other mainframe class systems. People who use those systems have developed design and management methods which reduce or eliminate externally caused disruptions, and which highlight the potential problems with application software. That's not true in the NT world.
In the work I referred to earlier, the MSFT IIS web server was the leading cause of failure in NT / IIS configurations. The Exchange server was a leading cause of failure in those systems... etc. etc... and the community of people who create programs for those systems, who service them, and who create and design them just does not have, for the most part, the same level of interest in reliability as those developing for Solaris. That is the result of many years of evolution in the Unix community. The community developing for MSFT, and the user base, has only just begun to develop the beginnings of that culture.
In that respect, Unix truly is years ahead of NT... you can't create a community of interest, and the inherent understandings of what needs to be done, and how, overnight.
That process was going on in the late '80s for Unix - it had the capability to do big jobs but the community and common understandings weren't there. Partly Unix has "grown up" and partly, the people who had that discipline in the IBM and VAX mainframe world have migrated to Unix.
edit - I mamaged to include SUNW in both of my posts... accidentally on-topic. |
