As we have seen in class, monte carlo methods of estimating things converge rather slowly. So, to get a relatively accruate result, we must use many many data points. Such a way of calculating desired results often requires supercomputers.

however, I found a paper on how to use Grid-computing to help solve some of this problem. Grid-computing is basically linking many computers or calculation devices (called nodes), so that we can create a virtual supercomputer that can do intense calculation while not having to invest in a lot of money for an actual super computer.

another good thing about using grid-computing, according to the paper is that: “Parallelism is a way to accelerate the convergence of a Monte Carlo computation. If N processors execute N independent copies of a Monte Carlo computation, the accumulated result will have a variance N time smaller than that of a single copy.”-(Yaohang Li and Michael Mascagni, page 2/12) this helps guarantee the result to of the calculations to be more accurate.

the paper then doves into different strategies /ways to prevent bad computations/delays etc. one of the things they talked about was the N-out of M strategy, which pretty much entails : assigning more subtasks than what we need, so that if a node(computer) is being too slow at its calculations, or if it accidentally shuts down, we won’t have to wait for it, because other nodes can keep going and finish the required N number out of the M assigned tasks and we are done.

in addition, monte carlo calculations are often heavily influenced by outliers. so the paper continues to mention error checking methods, such as majority vote, as well as redundant task calculation to make sure the partial results are accurate. all these strategies improve the accuracy and speed of monte carlo methods.

http://www.cs.fsu.edu/~mascagni/papers/RICP2002_3.pdf