The computing grid is becoming the platform of choice for large-scale distributed data-intensive applications. Performance measurement, analysis and prediction have become increasingly important in a grid environment, mainly due to resource?s geographic distribution, heterogeneity, dynamic, distributed ownership with different policies and priorities, varying loads, reliability, and availability conditions. Resource management and scheduling based on performance prediction can allocate more availably resources and schedule tasks to meet user?s performance requirement. Resource Performance includes many factors. A performance prediction model is put forwards to deal with task running times .The model is based on host load and can on-line predict the running time of tasks on candidate hosts. The system is evaluated using over 3000 randomized test cases. The experimental result shows that the model is practical and effective and its precision is preferable.