[vox] [fwd] [svlug] mission critical computing and air safety

[Rob Rogers]

On Wed, Sep 22, 2004 at 11:36:47AM -0700, Tim Riley wrote:
> On Wed, 2004-09-22 at 10:30, Bill Kendrick wrote:
> > Here's a nice long post with some more details about the air traffic issue
> > last week (and his thoughts on the Windows conversions) from Rick Kwan over
> > at SVLUG.  Just thought it was interesting reading...
> 
> Sure is an interesting read. It seems like a memory leak in
> the application caused the problem. However, are memory leaks
> so common that they should be expected? Should the operating
> system return leaked memory to the memory pool? Should a person
> have to reboot a computer every month to reset leaked memory?

Actually, I don't believe it has anything to do with memory leaks per
se. The text of the article says "The servers are timed to shut down
after 49.7 days of use in order to prevent a data overload" I think
they're using "data overload" here in place of something more technical
that might confuse the average reader.  The actual problem (well, I
don't actually know with 100% certainty, but it's not the first time
I've seen the a 49.7 day problem) is that somewhere in the application
there's a 32 bit counter being used. Doing a little math 2^32 =
4.294e09. Divide that by 24 (hours) 60 (minutes) 60 (seconds) and 1000
(miliseconds) gives you 49.7 ... hence the "data overload" being a time
based counter rolling over to zero.

The 49.7 day problem is actually a known problem with Windows 95 and 98
(and possibly others), where they would reboot after that time period
because of an internal Windows counter. Win2K doesn't have this problem
itself, but I have seen applications running on Win2K that had this
problem. In particular a developer friend of mine came to me wondering
why one app in particular started acting differently and needed to be
shut down and restarted every seven weeks. He was the one person I knew
who had more than 6 months uptime on a windows desktop box.