Study On Auto Scaling Framework For Container In Cloud Computing

Elasticity and scalability are key features of cloud computing and make it possible that users can acquire cloud computing resources as they need.However,there are still some challenges in cloud computing.For instance,applications deployed on cloud are faced with fluctuating workloads all the time,which make owners of applications plan suitable amount of resources in advance to make sure their application have enough resources to handle burst browsing requests.Cloud providers such as AWS provide customers with service to scale automatically their resources according to their usage rate.This kind of service guarantee customers' application not degrading even at the peak of workload while don't waste much resources and this is so called auto scaling.Problems arose by auto scaling in cloud computing have been widely studied.Methods like rule-based auto scaling have been adopted by many cloud providers.However,this type of models have their intrinsic drawbacks.In the other hand,container technology,as a light-weight virtual machine,has been popular recently while there exists few auto scaling framework catered for containers.Thus,this thesis focuses on solutions to auto scaling methods which are suitable for container environment and their implementations.In this research,we proposed a proactive auto scaling method based on time series analysis models of container resource usage.Then,we utilize Kalman filter algorithm to make our model suitable for online prediction.Besides,this research also shows the pros and cons of both vertical and horizontal scaling.We combine these two methods to implement our scaling framework for containers.At last,experimental results show our methods is practical and better than some existing method from the point of views of application SLA violations and average response time.