I know of a Sun V880 installed in January of 2002 with eight 750Mhz processors, 16GB, and four A1000 disk packs that ran Informix 10 with Sybase/IQ with exactly zero downtime for almost three years despite a half dozen or so disk replacements and a Sybase upgrade.
Rebuilt using Solaris 10, four 1.2Ghz USIV dual core processors, and four 3520 external SCSI RAID packs during a half day shutdown over Christmas 2005, that system then ran with zero additional downtime until mid 2006 - and has experienced almost weekly failures since late August 2006 despite having benefited from considerable consulting expertise on both Solaris and Sybase in the months since.
What happened was management: under bosses becoming aware of a problem and taking corrective action - meaning that SPARC was revealed as archaic, Solaris as unreliable, Informix as historical garbage, Sybase as obsolete, IQ as an intolerable barrier to clarity, and everyone involved now agrees that upgrading the database while splitting business intelligence processing from the transaction side will produce enormous business benefits to users.
But what really is the cause of this? These guys aren't outstandingly incompetent - in fact, they do as good a job running rackmounts as anyone else in their positions: it's that obsolete Sun box that's out of place in their data center, not their ideas about how to run it.
Do I want to get into their argument about whether to replace that V890 with a couple of x64 boxes running DB2/Linux feeding a Cognos data warehouse or just to go directly to that nice reliable SQL-Server product under Windows? No - what I want to do is forward them a clean third party document explaining why splitting the IQ and RDBMS transaction functions is as stupidly counter-productive for their business as having Windows programmers "tune" Informix on a perfectly functioning OS they don't understand and most certainly should never have been given the root passwords to.
And that's Sun's biggest failure: they make great products, their labs are cutting edge, their top executives are scarily bright, but they've given up on educating the market and even their own people now often parrot the commodity processor line with its implicit Windows software corolary.
Don't believe me? here's a challenge for you: find something, anything, on Sun's site explaining the value of SMP and in-memory data integration to those who don't already know. Sun makes the biggest, fastest, most integrated SMP machines available from anyone - and instead of telling people why a million dollar 6900 is worth eight times the cost of getting the same memory and processor cycles in twelve small boxes, they tell people how to spend more money dividing that machine's 48 core/256GB memory space into 12 Dells.
So here's the real, bottom line, question for Sun: this is ultimately a recruitment problem - so why not learn from Andersen's boot camps and go after your problem that way?