Research On Vertical Scaling Of Resource For OpenMP Application

For the past few decades,OpenMP applications have been mostly executed on high-performance computers.As the scale of the problem increases and users' demand for performance increases,whether to purchase higher-performance computers has become a problem faced by users.High-performance computers do well in the face of increasingly diverse and complex problems caused by the information technology boom,but this requires users to spend a lot of money on machine acquisition and maintenance,and can lead to waste of resources when workloads are low.Cloud computing offers a new way to solve this problem.Elasticity can automatically scale resources to meet different workload demands,which is a basic characteristic of cloud computing.Vertical resilience is considered to be a key driver for effective use of cloud infrastructure through fine-grained resource provisioning.It makes sense to leverage cloud computing to provide an OpenMP application with scalable resources to meet its changing needs.A vertical elastic solution for OpenMP applications is proposed,which is a combination of exponential smoothing and fuzzy logic control.The controller infers the number of required processor cores based on the internal structure of the OpenMP application and the prediction information of the virtual machine resource utilization.To realize the automatic scaling of resources for OpenMP application without modifying the source code.According to the solution,an elasticity controller ECOMP was implemented,and the experimental verification was conducted from two aspects.One is the prediction accuracy of Holt prediction model,and the other is the performance improvement of the elasticity controller for OpenMP application.The results show that the controller can accurately predict resource quantity,complete vertical elasticity scaling of resources,shorten the execution time of the program and improve the resource utilization efficiency.