FPGA-based Balanced System Management Design For Multi-node Server Clusters

With the rapid development of cloud computing,it requires much higher density of computing and storage,which brings the demand of customized,multi-node and high density server.In order to implement the load balancing in the high density server,each node server must knows all of status of itself while the system al so need to monitor all of load information and health status by Rack Control Module.Since all of nodes are in the same position,it needs an algorithm and mechanism to make sure the RCM can communicate with each node with the same probability.The paper proposes a method of load balancing management system for multi-node server clusters based on a 3U20N high density server product developed by the author.The topic full studies the current background,researching status of cloud computing and the development and reform of cloud infrastructure worldwide.It deeply studies the theory related to server cluster management,I~2C,IPMI,FPGA/MCU and load balancing.It also discusses the principle and advantages and disadvantages of I~2C arbitration mechanism,polling communication and interrupt communication.Based on this,the paper proposes a“REQ-GNT”communication algorithm for multi-node server cluster.This algorithm not only solves the shortcomings of the I~2C arbitration mechanism and the delay problem of polling communication,but also solves the the interrupt prority and the limited logic resources and pins of MCU problem about the interrupt communication.A completed load balancing management architecture for the server is designed based on the algorithm.The paper proposes a method how to implent this architecture through the hardware design for node server,FPGA and MCU circuit on RCM while how to implement the FPGA program by utilizing the hardware programming and pin advantages of FPGA.It also introduces the roles and workflows of MCU in the algorithm,same for BMC.Finally,the Modelsim is used for simulating and verifying the FPGA algorithm to ensure all of logic and function are accurate and the logical compile,mapping,place and routing is implemented in FPGA while the static timing analyse is carried out to ensure that the program can be fully applied to the production and engineering.The overall design and practice shows that the arlgorithm can achieve the load balancing management for multi-node server,while the product is also successfully imported into mass production.