Skip to Main Content
The cost or complexity of serial algorithms is estimated in terms of the space i.e. memory and time i.e. processor cycle. Parallel algorithms need to optimize one or more resources and also the communication between different processors. Parallel processing involves a technique by which complex data sets are broken into individual threads and processed simultaneously on one or more cores. If a system is doing one task which consists of large computation, it will take more time to finish that task. The proposed approach enables the efficient exploitation of parallelism by well-balanced distribution of workload to host processors and minimizes the overhead on single computer. Data division and migration is used for parallel processing. But if the process is migrated, it needs to take care of process for execution like pause, resume etc. It also needs to set check points for processes that become very complicated and time consuming. The proposed work divides the task into some subset on the basis of global decision and migrate that task among the slave computers. The output from the slave computers are collected on the master node. The proposed work considers the SIMD and MIMD computer applications for comparing the result for analysis. The proposed work contains static approach for process migration using data level parallelism. It creates small chunks of data from the large data set to reduce the execution time. It considers the memory availability of the slave nodes for taking the decision of data transfer. The proposed approach considers equal and unequal data division and distribution for load balancing. In addition to computational cost, the cost of memory accesses is an important factor for determining program performance. Proposed work is based on memory availability of slave nodes, as per their availability they are given an additional task to perform.
Date of Conference: 3-5 Nov. 2012