Cloud is one of the rapidly developing technologies as it allows convergence of many new areas like scalability, rapid elasticity and broad network access in cloud services. This paper focuses on resource allocation for the parallel data processing, which also occupies a major issue in cloud computing. The current paper aims to study the two major cloud problems i.e. QOS constrained resource allocation and parallel data management problems. The customer demands the cloud provider to host their application with desired SLA such as throughput, response time among others. We have proposed a methodology which is expected to achieve the specified SLA requirements in parallel data management and forecast/identifies the load to be processed by the cloud. This may use the divide and conquer strategy to predict the load on VM?s (in dividing phase) and which will be allocated to that requested customer. We take Nephele, a parallel data processing framework in which job graph issued by the customer to be converted into an execution graph on an Iaas cloud system as an example to frame our architecture resulting with minimum response time and efficient load balancing service.
展开▼
