> The kernel can schedule an event at arbitrary times in the future, and > those can be represented as timespec, timeval or even plain seconds.
only true as long as "arbitrary" can fit into a 32-bit signed count in ticks. ie, 248 days with HZ=100, etc. with HZ=1000, that's almost 25 days. i hear today about the being something fixed with HZ=4096, which is only 6 days. i recommend avoiding any kernel timeouts longer than a day. .mrg.
