The BSPlib installation seemed to have gone ok for almost all of you.
 Issues about the programming.
 1. If you run the same code on p CPUs with the same seed
    (because you did not use srandom, or the value was not processor/process
     dependent) then you repeat the same experiment with the same
     random number generator sequence p times. You can avoid them with
     careful reseeding per process (involve for example bsp_pid() and
     or Unix process id). Time might be dangerous (even process id might)
     as all CPU might have the same date/time.
 2. Ideas wise
    Most of you
    x = random() / 2^31-1
    y = random() / 2^31-1
    if ((x-0.5)**2 + (y-0.5)**2 < 0.25) etc
    with optional subexpression optimization (some of you did so).

    One alternative that can slightly speedup things is
    x = random() / 2^31-1
    y = random() / 2^31-1
    if (x**2 + y**2 < 1.0 ) etc
    (i.e. one quarter of a circle with (0,0) center and radius 1 is
    inscribed in the square).
     
    This eliminates some minor subtractions etc.

    Sqrt finding is not required and can only slow down things!

 3. Do the following
#define R (double) (2^31-1)* (2^31-1)
    X = random() 
    Y = random() 
    if (X**2 + Y**2 < R ) etc, and
    crossing fingers about overflows:-),
    the elimination of divisions can shave off 20-30 seconds for
    example from the runtime with 100,000,000 runs.

 4. Optimization-wise when you use
    gcc explore using -O3 -or -O2
    or
    gcc -O? -funroll-loops
            -mcpu=cputype
    where cputype
    is one of: pentium3, i486, i586, pentium, pentium4

    with BSPlib always do -flibrary-level 2 in production code
    (not during debugging where my suggestion is to use -flibrary-level 0)
    
alex