utmp rollover at 496 days?

utmp rollover at 496 days?

[jim roy]

I have two machines which have not been rebooted sind Dec 2000.

At day 497, on both machines, the uptime command began anew at day 0.

Anybody know what's up with that?

Jim Roy

Re: utmp rollover at 496 days?

[Scott Taylor]

At 10:53 AM 09/05/2002, you wrote:

>I have two machines which have not been rebooted sind Dec 2000.

Way to go.

>At day 497, on both machines, the uptime command began anew at day 0.

Don't you think it's time for a good cleaning and maybe replace a hard 
drive or two by now?

>Anybody know what's up with that?

Guessing Linus didn't figure you would run it so long, with out some kind 
of maintenance, so using a very short integer to count days would save 
space in the kernel.  I've only seen this question asked about 3 times, and 
once was for an SCO UNIX (maybe XENIX) that would crash on day 498.

The 2.4.19 kernel has or is patched to fix this.

Re: utmp rollover at 496 days?

[Joseph Bueno]

jim roy wrote :
> 
> I have two machines which have not been rebooted sind Dec 2000.
> 
> At day 497, on both machines, the uptime command began anew at day 0.
> 
> Anybody know what's up with that?
> 
> Jim Roy

Hi,

This is a known problem with older kernels.
uptime is kept in a 32 bits counter with a resolution of 1/100th of
a second so it rollover after 497 days.

It has been discussed in linux-kernel mailing list a few months ago
and I think it is now fixed in 2.4 kernel.

Regards
--
Joseph Bueno
NetClub/Trader.com

Re: utmp rollover at 496 days?

[James]

On Thu, May 09, 2002 at 11:09:05AM -0700, Scott Taylor wrote:
| At 10:53 AM 09/05/2002, you wrote:
| 
| >I have two machines which have not been rebooted sind Dec 2000.
| 
| Way to go.

Indeed! And I thought 200 days was good :)

| >At day 497, on both machines, the uptime command began anew at day 0.
| 
| Don't you think it's time for a good cleaning and maybe replace a hard 
| drive or two by now?

Nah, if it still works, leave it be. Although it has rolled over, so it
doesn't look as "eleet" when you type "uptime" any more...
 
| >Anybody know what's up with that?
| 
| Guessing Linus didn't figure you would run it so long, with out some kind 
| of maintenance, so using a very short integer to count days would save 
| space in the kernel.  I've only seen this question asked about 3 times, and 
| once was for an SCO UNIX (maybe XENIX) that would crash on day 498.

There's another way to track uptime now. It requires a very technical
tool known as a "black marker". Every time the counter rolls, make a
mark on the computer's casing :-)

-- 
This punishment is not boring and pointless  
 
6AD6 865A BF6E 76BB 1FC2 | www.piku.org.uk/public-key.asc
E4C4 DEEA 7D08 D511 E149 | www.piku.org.uk wnzrf@cvxh.bet.hx (rot13'd)