| As the cluster technology develops, the research of high performance parallel programming on cluster becomes the focus. The parallel programmers have to face many borning works such as load balancing which have no close relations to application yields.So the requirement of free programmers from unnecessary and overelaborate formalities rised.Such is our project's goal.We design a load balancing integrated framework which can be used to easily build efficient parallel programs on cluster. Load balancing is the base for building an efficient parallel program, so it is always the hotspot field of cluster systems. In this paper, after introduce some technologies such as the cluster,distribute computing and dynamic load balancing, we focus on discussing two questions below: 1) how to design a parallel program development framework, and 2) the design and implement of a new dynamic load balancing methord. For the design of the parallel program development framework, we use the ideas of pattern design,software architecture and many successful frameworks to provide the programmers an easily used interface.We build the load balancing decision-making module with plug-in technology,so we can change different decision-making arithmetic in this framework simply.We also design a new dynamic load balancing arithmetic that makes probability decisions primary depends on local informations. This method decreases the data transformations between host,and measure shows it has a good result. Directed by above design ideas,we use mobile agent technology to achieve dynamic transformation,and complete the framework and plug-in with Aglet development kits. In this paper, we focus on the architecture of parallel program development framework and our new dynamic load balancing arithmetic. We also provide two examples to prove the feasibility of the system. The design and implementation of this system can obtain experiences of parallel programming and cluster load balancing.
|
PDF Full Text Request