Re: [Ganglia-developers] Error with random number function in monitor.c

[matt m. <ma...@cs...>]

albert-

there is nothing to say here except you are right.  painfully right.  i'm 
going to check in your changes to CVS now and it will be a part of the 
next release.  thanks for catching this error.

-- 
matt

Today, Albert Chu wrote forth saying...

> Hello,
>  
> I'm not sure about all architectures, but for most architectures, I
> believe random numbers will not be properly generated with the
> rand_int() definition in monitor.c.  I noticed this issue in ganglia
> 2.4.1 running on Linux, but I believe the problem still exists in
> 2.5.3.  As of 2.5.3, the rand_int() definition is:
> 
> #define rand_int(__min,__max) __min + ( __max * rand()/(RAND_MAX+1.0))
> 
> Problem #1:
> 
> - rand_int() is called from set_check_threshold() and set_mcast_threshold().
> 
> - These two functions pass in metric[key].check_min,
> metric[key].check_max, metric[key].mcast_min, and
> metric[key].mcast_max as values to the rand_int() definition.
> 
> - check_min, check_max, mcast_min, and mcast_max are defined as
> integers in metric_typedefs.h.
> 
> - Thus the following problem comes up ...
> 
> rand() = "0 <= integer <= RAND_MAX"
> 
> __max * rand() = "INT_MIN <= integer <= INT_MAX"
>        - because of possible maximum integer overflow
> 
> (__max * rand())/(RAND_MAX+1.0) = "-1 < float < 1"
>        - because numerator is always smaller than denominator
> 
> (int) (__min + "-1 < float < 1") = __min or (__min - 1) after truncation
> 
> By making __max a float, the integer overflow does not become an
> issue, and a random number is properly generated.  
> 
> Problem #2:
> 
> I also believe that rand_int() is incorrect because it only works when
> __min == 1 (which is what is in the Linux manpage for rand()).    
> 
> I believe the code ((__max * 1.0) * rand()/(RAND_MAX+1.0)) will actually
> produce a random number from 0 to (__max - 1).  Thus, this works
> perfectly for when __min is 1, but not when __min is larger than 1.  I
> believe you want
> 
> (((__max-__min)*1.0) * rand() / (RAND_MAX + 1.0))
> 
> My test code & solution:
> 
> I ran it on Linux.
> 
> #include <stdio.h>
> #include <stdlib.h>
> 
> #define rand_int(__min,__max) __min + ( __max * rand()/(RAND_MAX+1.0))
> 
> #define my_rand_int(__min,__max) __min + ( ((__max-__min)*1.0) *
> rand()/(RAND_MAX+1.0))
> 
> int main() {
>   int i,num;
>   srand(1);
> 
>   for (i = 0; i < 25; i++) {
>     num = rand_int(10,20);
>     printf("%d\n", num);
>   }
>   printf("\n");
> 
>   for (i = 0; i < 25; i++) {
>     num = my_rand_int(10,20);
>     printf("%d\n", num);
>   }
>   printf("\n");
> 
> }
> 
> Thanks
> Al
> 
> 
> 
> -------------------------------------------------------
> This sf.net email is sponsored by:ThinkGeek
> Welcome to geek heaven.
> http://thinkgeek.com/sf
> _______________________________________________
> Ganglia-developers mailing list
> Gan...@li...
> https://lists.sourceforge.net/lists/listinfo/ganglia-developers
>