Download now Free registration required
Optimum allocation of resources can make automated systems of an efficient data centre more effective. Data centers that engage in horizontally-scalable Internet services are reliant on numerous automated activities so that system management is more efficient. While earlier systems and controllers relied on a performance model of the system to plan allocation of resources, the models work offline or are deployed at a small scale. Due to these reasons, it is unable to measure the performance of a controlled application. The better solution in such a case would be if the models are trained directly on the production system. This works as a dynamic solution where the model is able to adapt to changes in workload and performance because of which it provides greater control over the web application. This paper suggests deployment of an exploration policy to train the performance model so that one can collect data from different performance regimes of the application. Such a policy will enable controllers of the data centre to balance violation of performance Service Level Agreements (SLAs) with the requirement to collect data and train the right performance model. The paper shows that with this exploration policy one can train a performance model of a Web 2.0 application in less than an hour. It also claims that the performance model can immediately be used in a resource allocation controller.
- Format: PDF
- Size: 640.4 KB