Design And Implementation Of Container Clusters Based Load Balancing Platform

Application providers rely on proxy servers to expose their web services inside container clusters.However there are some problems for the existing container clusters based load balancing framework in load balancing algorithm and proxy server management: First of all,the load balancing algorithm is quite simply.For the request part,it does not combine the business scenario to evaluate request resource consumption accurately.For the container part,it fails to present container load accurately because it does not take other factors like business scenario and node load that can influence container load inside container clusters into consideration.Therefore,it is not capable of performing accurate load balancing based on request resource consumption and container load.Then,the existing framework relies on a single proxy server to perform load balancing,if something goes wrong with the proxy server,it lacks perfect mechanism to guarantee its availability and scalability.To address these problems,a container clusters based dynamic load balancing platform is proposed in this paper.Aiming to improve the performance of load balancing,our method takes business scenario,request resource consumption and container load into consideration to perform dynamic load balancing.As for improving the availability and scalability,our method carries out management mechanism for proxy servers based on their load.The main contents of this paper are listed as follows.(1)An estimation method of request resource consumption for web applications.The estimation method divides web application into two types: traditional service and model service.Based on the service type,the method analyses resource characteristic of requests and proposes resource models to represent resource consumption for requests.Finally the resource consumption will server as one part of input for our dynamic load balancing.(2)A request type based matching strategy for containers and requests.The strategy establishes different load evaluation method for container of different service type.Then it performs dynamic matching based on the request models and container load.(3)A dynamic manage mechanism based on container state.The mechanism analyses the business workflow of proxy servers and redefine the lifecycle and state evaluation method for proxy servers.Finally it performs dynamic failover and scaling for proxy servers based on their states.(4)A container clusters based load balancing platform.According to the proposed framework,a container clusters based load balancing platform is constructed for Kubernetes.Experiments and discussion are performed to illustrate the effectiveness of our platform.