A Cloud Computing Mode of Dual Parallel Computing Based on MapReduce and GPU
Jiabin Yuan · Computer and Digital Engineering · 2013
The Map and Reduce operations of MapReduce model require frequentcomputing on CPU.While parallel computing tasks are massive,the CPU occupancy rate will be even up to 100%.However,GPU has the better parallel computing capability than CPU.If GPU is used appropriately,it can reduce the CPU occupancy time and balance the computing capability of the system.This paper takes the different advantages of GPU technology and MapReduce technology to build a cloud computing mode of dual parallel computing based on MapReduce and GPU.Through theoretical modeling and experimental validation,the results of this design show that this model can finish parallel processing of multi-GPU MapReduce task to improve the performance of high-performance computing.