Question :
Synchronization barriers are a common paradigm in many parallel applications. A barrier is supposed to block a calling thread until all N threads have reached the barrier.
(Parallel applications often divide up the work, e.g. matrix operations or graphical rendering, among N processes, each of which compute independently, reach an barrier, exchange results, then work on the next phase of computation) Many times the number of threads is not known in advance.
One pseudocode monitor solution is as follows:
Monitor Barrier ( ){
int numThreads ;
int numThreadsAtBarrier ;
newThreadCreated( ){ numThreads++; }
barrierReached( ){
numThreadsAtBarrier ++;
if(numThreadsAtBarrier == numThreads)
numThreadsAtBarrier = 0;
signalAll( );
else
wait( );
} initilaizationCode( ){
numThreads =0;
numThreadsAtBarrier = 0;
}
}
a. Provide comments in the psuedocode, describing the monitor code.
b. Briefly describe a difficulty in a semaphore solution to this barrier problem.