🚀 go-pugleaf

In article <RJC.92Mar5114747@brodie.cstr.ed.ac.uk>, rjc@cstr.ed.ac.uk (Richard Caley) writes:
|> For an operating system in particular MTBF is very meaningful.

I partially disagree: MTBF means something for some types of OS defects
and is meaningless for others.

One possible reason for an OS crash is that a race condition exists
such that the OS crashes whenever, say, a pair of independently generated
events occurs in a particular time window.  If you plot the statistics
of these failures they will resemble MTBF statistics for hardware
components, so if the OS has only these kinds of defects, MTBF might
mean something: modify the OS to decrease the length of the window,
or eliminate some race conditions but not others, and MTBF goes up.
But a "bad guy" could write a program to exploit the race condition
(in much the same way as you crack a system when someone is dumb enough
to have a setuid root shell script); that program will blow the OS
away imediately.

But OSes also have bugs that cause instant crashes as soon as a particular
program does a particular thing.  In this case, MTBF is meaningless;
if a user is running crashme and the OS has a defective trap handler,
"MTBF" might be a few seconds; otherwise, it might stay up for years,
until a particular program comes along that triggers the bug, in which
case the system dies instantly whenever that program is run.

-- 
Joe Buck	jbuck@ohm.berkeley.edu