Grid- History Based Prediction Architecture (Grid-HPA) is a distributed software tool for predicting resource requirements of a job(s) in Grid Computing. Grid-HPA uses similarity measurement based on centralized Historical data of jobs as a knowledge base for deployed Grid infrastructure. Using statistical analysis and similarity measurements, Grid-HPA can predict and provide minimum resource requirement alternatives as a decision support system for newly submitted jobs by clients. Clustering of historical jobs and dynamic updating of clusters improve the prediction accuracy of the tool in comparison to other relevant projects. This tool can be used as an assistant system for decision support in resource allocation and reservation with less knowledge about resource requirements of jobs in order to manage computing with less costs and in a short time with maximum efficiency which could result higher resource utilization of grid and cloud environments. The concept fits very well to the new issues of distributed systems such as cloud computing. Project members will announce soon the cloud adapted version.
Mahdi Bohlouli is the project leader of Grid-HPA. The idea of this project has been started from his master thesis in 2007. Mahdi published the concept details and implementation results firstly in International Journal of Computer Science  after having successful and more accurate implementation results. Java has been used for implementation part and the project database is based on SQL-Server. Project database consists of to the history of former executed jobs and their execution details as well as used resources for their execution. The proposed algorithm of prediction can be used for every job in every environment. The measured accuracy of prediction in a centralized approach is better than a decentralized method. Both methods have some advantages and disadvantages.
In Grid systems, the Quality of Service (QoS) is not limited to the network bandwidth, but extends to the processing and storage capabilities of the nodes. Thus the focus is on the degree a Grid can provide end-to-end QOS rather than providing only QOS on the network. When a Grid job has QOS requirements, it may be necessary to negotiate a service level agreement (SLA) to enforce the desired level of service. Resource reservation is one of the ways of providing guaranteed QOS in a Grid with dedicated resources. One of the major issues in resource reservation is correct matching of available resources with required resources, this results effectively utilize various resources in the system, such as CPU cycle, memory, communication network, and data storage. Correct matching operation depends on percept job resource requirements. This can be obtained by two methods. In the first method, environment can ask resource requirements from user. This is not proper in QOS. Another method is that architecture itself predicts resource requirement. job resource requirements prediction algorithms operate on the principle that jobs with similar characteristics have similar resource requirements.