Henrik Stoerner wrote:
On Tue, Mar 25, 2008 at 09:40:12AM -0700, Sloan wrote:Henrik Stoerner wrote:On Tue, Mar 25, 2008 at 02:12:42PM +0100, Charles Goyard wrote:On our setup, the problem comes from a counter overflow in uptime handling on some SCO : after 248 days of uptime, the uptime is reported negative, and the rrd handler misses the expression (no "up:").Ahh - the old 31-bit timer tick issue. Linux has the same problem because a timer ticks at 100 Hz using a signed 32-bit integer; so 2^31 / (24*60*60*100) = 248 days and a bit.That's news to me -Sorry, I should have written "Linux HAD the same problem" (past tense). This was fixed somewhere in the early 2.4.x kernels. That's why I wrote it was the "old" problem.
Yes, linux used to wrap around at 497 days uptime, so it must have been unsigned int.
Joe