Email: Password: Remember Me | Create Account (Free)

Back to Subject List

Old thread has been locked -- no new posts accepted in this thread
???
07/14/06 15:22
Modified:
  07/14/06 15:24
Read: times


 		 #120260 - to bite or not to bite
Responding to: ???'s previous message
Erik Malund said:
Jan, what you state is somewhat valid, but what you seem to miss is that 99.9% of the incidents that a WD should catch in fact ARE caught by traditional internal watchdogs.


I never did a statistics. In fact, I personally DO NOT use any WD. Simply, the applications I am working with are better off without a WD. They never run unattended, and they make (almost) no harm if they fail. I prefer them to fail and give the users a chance to report the error - most of the time they fail because of SW error, anyway... However, there are applications, where it is even better to hide a SW error at least ocassionally using the WD. Not nice perhaps, but life is life...

My point is, it really does not matter how the WD is constructed and used. What matters at the end of the day is, what it does. If it costs an extra $1 to have an independent oscillator (which it does not - a ring oscillator in an mcu costs $0.00000001 - in fact it costs more in terms of consumed power than in terms of silicon area) but in 0.1% of cases it potentially saves $100000 or a life, it's worth doing it this way. It's just the convenient thinking of the manufacturers, and some marketing stuff - they know integrated WDs are required to sell mcus and it is SO EASY to add one to the existing clock!

Erik said:
1) is not worth it for the cost vs numbers of catches (I imagine zero).
see above.
Erik said:
2) I know of no decent internal dog that does not
That's exactly what I'm talking about - some of them ARE crap :-).
Erik said:
5) Nice, but why? if you get a WD reset you ARE in doo-doo.
To be able to do the coredump stuff - for diagnostics. Of course, not all applications do that or are able to do that, but, cm'on, it's 10 transistors or so. Here, the (even if potentially small) added value clearly outweighs the related costs.
Erik said:
6) would defeat the purpose of consequtive writes, if the writes do not have to be consequtive, the chance of runaway code keeping the puppy alive is greatly increased.
Again, I never studied the typical mechanisms of runaway etc.etc.; I think we could elaborate on this in great length, but... it's friday today, isn't it? Not quite a time for lentgthy and deep discussions :-)))

Have a nice weekend!

Jan Waclawek




List of 30 messages in thread
TopicAuthorDate
processor fault            01/01/70 00:00      
   Use Watchdog            01/01/70 00:00      
      Not necessarily            01/01/70 00:00      
         Spec of AT89S51            01/01/70 00:00      
            Still too specific            01/01/70 00:00      
   There is no one such thing!            01/01/70 00:00      
   Undefined state !            01/01/70 00:00      
      processor fault            01/01/70 00:00      
         RTFM            01/01/70 00:00      
            reset            01/01/70 00:00      
               Read the data sheet            01/01/70 00:00      
                  WD with indication            01/01/70 00:00      
                     I, too, can visualize some exotic conditions where            01/01/70 00:00      
                        to bite or not to bite            01/01/70 00:00      
                           the REAL issue            01/01/70 00:00      
                              Marketing trick            01/01/70 00:00      
                                 to be dramatic            01/01/70 00:00      
                                    It is helpful, of course.            01/01/70 00:00      
                                 let's put it this way:            01/01/70 00:00      
                           I doubt that it's 10 transistors ...            01/01/70 00:00      
                              the context            01/01/70 00:00      
                     Can you name an example of this sort?            01/01/70 00:00      
                  reset            01/01/70 00:00      
                     AT89S51 WD            01/01/70 00:00      
         How do you propose to make the processor "illumina            01/01/70 00:00      
   Software hang-up, hardware lock-up            01/01/70 00:00      
      How about redundancy?            01/01/70 00:00      
         paranoia            01/01/70 00:00      
            ... but it WILL detect a processor fault!            01/01/70 00:00      
               The ultimate misconception            01/01/70 00:00      

Back to Subject List