Research the phenomenon of data races. Give an example of how an unprotected data race can give rise to data inconsistency. How do OpenMP and Cilk resolve this problem?
Data race is that situation or condition in which multiple threads, without intervening synchronization, try to access the same memory location, with at least one thread performing a write operation. For eg. many data races occur on numerical algorithms owing to chaotic relaxation and asynchronous iterative methods. Also, there are systems where synchronized threads or processes maintain probabilities with best-estimate-matrix and may race with one other to provide updates.This is usually described in higher-level language specs by the phrase: "undefined behavior." A data race could legitimately reprogram the BIOS delete data from the disk and may also stop the processor's fan leading to a multi-core meltdown.
example code:
int checker = 0;
int counts = 0;
Lock lock;
bool TryEnter() {
if (checker == /* get thread id */) {
counts += 1;
return true;
}
if (lock.TryEnter()) {
checker = /* get thread id */;
return true;
}
return false;
}
void Exit() {
if (counts != 0) {
counts -= 1;
return;
}
checker = 0;
lock.Exit();
}
the compiler or the processor may “optimize” this program as:
void Exit() {
if (counts != 0) {
counts -= 1;
return;
}
checker = 42;
checker = 0;
lock.Exit();
}
The effect of this is perhaps not observable in the current thread, nor is it observable by other threads in the absence of data races. Here, the unfortunate thread whose ID is 42 might observe this value and take the ignite the mistake.
OpenMP and Cilk resolve data racing through a technique known as Data parallelism.It focuses on distributing the data across different parallel computing nodes. This is achieved when each processor performs the same task on different pieces of distributed data. In some situations, a single execution thread controls operations on all pieces of data. In others, individual threads control the operation, but they execute the same code.