The cloud computing paradigm promises a cost-effective solution for running business applications through the use of virtualization technologies, highly scalable distributed computing, and data management techniques as well as a pay-as-you-go pricing model. In recent years, it also offers high-performance computing capacity for applications to solve complex problems. Improving resource utilization is essential for achieving cost effectiveness. Low utilization has long been an issue in data centers. Servers in a typical data center are operated at 10 to 50 percent of their maximum utilization level . 10 to 20 percent utilization is common in data centers. For a data center, or a subset of servers in a data center that mainly handles applications with high-performance computing needs and runs parallel jobs most of the time, the problem can be significant. A parallel job often requires a certain number of nodes to run. A set of nodes is likely to be fragmented by parallel jobs with different node number requirement. If the number of available nodes cannot satisfy the requirement of an incoming job, these nodes may remain idle. Next one Typical parallel programming models, such as BSP often involve computing, communication, and synchronization phase. A process in a parallel job may frequently wait for the data from other processes. During waiting, the utilization of the node is low. The most basic but popular batch scheduling algorithm for parallel jobs is first come first serve (FCFS). Each job specifies the number of nodes required and the scheduler processes jobs according to the order of their arrival. When there is a sufficient number of nodes to process the job at the head of the queue, the scheduler dispatches the job to run on these nodes; otherwise, it waits till jobs currently running finish and release enough nodes for the job. FCFS may cause node fragmentation and methods such as backfilling and Gang scheduling were proposed to improve it. They do not target on the utilization degradation caused by parallelization itself.