[VeniVidiVici]

There was a Microsoft system in air traffic control that required regular restarts to keep running. They were normally scheduled every 30 days to avoid the bug (which blew every 49.7 days), but someone forgot and the machine restarted itself right in the middle of operations, threatening air traffic. It was a problem with using Microsoft's RPC and GetTickCount(). They didn't have that problem when it was running on UNIX.

Do you have a link for this? I have a link that says this was a fault with Harris' implementation of the software that uses GetTickCount, and that the 49.7 day limitation of it in the API is due to it being 32-bits.

Additionally, it went on to say that the FAA had been meaning to get the program changed, but in the interim implemented a 30day reboot sked so as not to come close to the 49.7 limit.

[antiRepublicrat]

Do you have a link for this? I have a link that says this was a fault with Harris' implementation of the software that uses GetTickCount, and that the 49.7 day limitation of it in the API is due to it being 32-bits.

We have the same information. Microsoft's implementation rolled over at 49.7 days (2^32/(1000*60*60*24)). Microsoft's RPC uses GetTickCount(). Pre-fix implementations (Update Rollup 1 for Windows 2000 SP4 last year fixed it) would grow to use 60% of processor resources after the GetTickCount() rollover, killing system and network performance.

Because of this, the machine would become unusable if the FAA didn't reboot often. Harris put in a reset routine when the OS hit 49.7 days of uptime as a safeguard against the slowdown. So you get shutdown instead of slowdown.

It still amazes me though. Microsoft didn't think anyone would want to have their server up for 50 days at a time?