Workload Forecasting Framework For Applications In Cloud

With the development of cloud computing technics, an increasing number of applications prefer to be deployed in cloud. Load balancing becomes the key technic for cloud provider to control the resources and cost. But using load balancing with real time data cannot react in time towards workload peak or valley. Thus, workload forecasting is presented to let the cloud provider to get ready for a possible workload change. The workload forecasting algorithm is also a key technic for cloud.There are already many kinds of predicting methods. They have different performances under different workbenches and have different interfaces for user. In this article, we study the workload of applications in cloud and propose a workload forecasting framework. This framework consists of 5 modules: monitor module, data processor module, database module, database maintain module, workload predictor module. They solved the following problems of this issue: How to monitor the workload of specific application in real time? How to process the data for acceleration under big data? How to store the big data and maintain them? How to choose the best algorithm according to the specific application? How to predict workload in real time? How to display all those data?This article uses some special technics for this issue: This article introduces a knowledge base and a knowledge base engine, to store the accuracy of algorithms under different types of application. This article uses non-relational database instead of relational database to improve the performance. This article uses "request intensity" instead of just counting the requests to represent the request workload. This article studies how to set better arguments of "database maintain module". This article introduces a unified series of interfaces for all kinds of workload forecasting algorithms, making this framework algorithm-independent and can load real algorithm at run time.At last, this article implements a tiny cloud, an application and this workload forecasting framework. This article develops a request simulator to simulate the requests according to 98 World Cup access log. After our analysis of the results, it is proved that this workload forecasting framework can work just as this article planned: it monitors workloads of applications in real time, processes the data, and provides feedback of the predicted workload value according to historical data, guiding the cloud provider to allocate resources. And the framework itself is application-independent and light-weighted.