The Design And Implementation Of Nimbus Task Prediction And Query System

VMware Nimbus is a private cloud based on VMware vSphere Suite,developed by Ruby in order to provide automatic tests.VMware Nimbus accepts tens of thousands of test tasks every day.If users submit the tests while system peak,Nimbus will be in resource competition status because of resource bottleneck,which seriously lower the stability of the system.As the result of resource competition,the efficiency of testing tasks submitted by users will drop a lot,which will significantly extend the duration time of test tasks,even make the test tasks failed by time out limit.To solve this problem,Nimbus need a task prediction and query system.The task prediction and query system collects records of user tasks in Nimbus private cloud and determine the type of user tasks,using basic statistical methods to predict user tasks'duration time for users,gives advice to version control system that it can dynamically set the maximum test time for tasks,thus easing the resource contention problem by lower the number of test missions user submited at the peak of system,and prevent the repeatedly submitted failure task caused by unsuitable test limit time,and ultimately improve the availability and stability of the system.Task prediction and query system mainly use the statistical basic methods like parameter estimation and hypothesis validation on task distribution through assumptions,estimates,and other basic aspects of verification.The final distribution is calculated in real-time and eventually attempt to estimate the distribution model of task completion time.The subsequent part of the paper will talk about this process in detail.Under the premise of complete system tasks estimated,in order to further simplify user's Nimbus private cloud operations,task prediction and query system also added some features in the deployment for very commonly used functions.In the development and design process,the task prediction and query system is designed with the thinking of Rest API,combined with Sinatra framework,Mongodb database and other advanced technology,simply and efficiently complete the functional and non-functional requirements,and ultimately improve the productivity gains of VMware products as a contribution.