[Asterisk-Users] Asterisk 1.0.9 long term stability <--thread hijack, why not reboot?

Paul digium-list at 9ux.com
Thu Sep 15 13:06:44 MST 2005


Andrew Kohlsmith wrote:

>On Thursday 15 September 2005 11:38, Paul wrote:
>  
>
>>They designed it to be shut down. I guess that means it doesn't  just
>>roll over like a dead cow.
>>    
>>
>
>Actually dead cows aren't back-heavy.  They typically just keep whatever 
>position they were in when they took their last breath, much like the telco 
>crosspoint switches.  :-)
>
>  
>
In order to argue that point I would have to do some rather inhumane 
research :-D


>>You wouldn't design a crane controller so that it releases the load on
>>reboot and then tries to return the cable to the pre-reboot position.
>>The same principle applies here. If it's mission-critical, you might
>>have to spend more on the hardware. A good example would be crosspoint
>>    
>>
>
>I agree -- You need to spend more time on both hardware and software.  For me, 
>a phone system is like a file server or network switch -- it is *not* meant 
>to be rebooted.  If it needs to be periodically power-cycled then address the 
>problem, don't simply put a cron script in.
>
>  
>
>>I did hardware and software design for some devices that were usually
>>placed in very remote areas(underwater, mountains, arctic are a few). We
>>used a very low power clock device that would boot the cpu up. We did
>>whatever needed to be done, set the clock registers for the next wakeup
>>and shut down again. There were other ways to approach the problem, but
>>this approach made coding easier, used less power and allowed us to get
>>more functionality without increasing rom or ram size. If you only
>>    
>>
>
>How did it save you ROM/RAM?  I can see it saving having to put fancy power 
>controller code and hardware in... Is that what you meant?
>  
>
The short answer is that the coding was more linear. Boot, run a state 
observer, setup the clock chip for next alarm and write an I/O bit that 
kills the power bus feeding the CPU area of the board. When data was 
being unloaded through the serial port, I skipped the final write and 
looped back to the top of the state observer.

>  
>
>>needed to take one simple measurement and store it to the ram, you could
>>do a boot/run/shutoff every second and still achieve some additional
>>battery runtime.
>>    
>>
>
>Agreed but again -- you designed for this specific purpose.  Asterisk isn't 
>meant to boot up, answer the phone, process the call and shut down again 
>until the next ring.  (This would be an interesting approach to power savings 
>though if your system boot time was fast enough and call volumes varied 
>enough to make it worthwhile.)
>  
>
I have seen windows systems that boot up and crash. Would this be a good 
start?




More information about the asterisk-users mailing list