[Cialug] uptime

[Dan Hockey]

Daniel A. Ramaley wrote:

>On Friday 07 July 2006 15:23, D. Joe Anderson wrote:
>  
>
>>Heh.  I thought one needed to reboot, too, but kept my mouth
>>shut on this one for a while because for a time there was a
>>project I saw referred to as "two-kernel monte" that would allow
>>you to start a new kernel from a running one.  That was back in
>>the 2.4 days, at least, maybe earlier, and I hadn't heard much
>>about it since.  I thought it was defunct, but then you never
>>know.  LVM hit a rough patch there for a while, and now it seems
>>to be common enough, for example.
>>
>>Anyway, my expectation is that even with something like that
>>going, the uptime counter would reset anyway.
>>    
>>
>
>I remember hearing about that as well. If i remember correctly, it would 
>basically load the new kernel into memory, make the new one take over 
>operations of the old kernel, then remove the old kernel. Both kernels 
>had to have been compiled with the proper extension to make it happen, 
>though. And i don't think it could be loaded as a module.
>
>One thing that i've thought is that perhaps if the system undergoes a 
>scheduled reboot that it should write the uptime to a file and then 
>when the system boots reload the old uptime and continue keeping time 
>from there. What is really important for most machines is not how long 
>it has been since last reboot, but how long it has been since the last 
>*unscheduled* reboot. I'm not sure how all the details would work to 
>make this happen. I would guess that some kernel code would have to be 
>modified to allow setting the uptime value on boot. And some other 
>system changes would be necessary. For example, if you used "at", 
>"cron" or "shutdown -t" to schedule a reboot, then the uptime would 
>have to be written somewhere in /var. On boot, if the uptime file was 
>present the uptime would have to be read and the file deleted. There is 
>probably a better way; i just thought of this. But i've wondered why no 
>one has done such a thing (of if they have why it isn't more widely 
>known since lots of people like high uptimes). The big downside, of 
>course, is that you'd have someone who would mess with the uptime 
>statistic and have a machine that reports a 30 year uptime! Which would 
>be kind of neat, but would make one start to question the value of 
>keeping uptime statistics at all.
>  
>
Why not keep track of the average runtime between reboots and/or crashes?
Say for example the kernel would write to a hidden file every minute or so.
If it was a planed reboot then keep the uptine, if not reset the uptime 
or count it as a crash.
This would be a more realistic since all comuters crash at some point in 
time.
-dh