Design And Implementation Of The X86 CPU Based High Available SMP Server

With the rapid development of Internet and the dramatic rise in smart phones,and the arrival of digital home era,computer technology has penetrated into every aspect of life,high reliability multi-processor servers play a significant role.In the SMP(Symmetrical Multi-Processing)server market,Because of the advantages of reliability,availability,and maintainability,the minicomputer represented by the IBM Power platform have been the front-runner.As Intel x86 platform performance greatly improved in recent years,high-end multi-processor server of x86 platform received high recognition,but compared with minicomputer,these products have a wide gap in HA(High Availability).Based on Intel Brickland platform,this thesis studied the high availability of x86 SMP server.The purpose of high availability design is to reduce system shutting down time,and system availability can be measured by system operation time percentage.In order to improve system availability and reduce the mean time to repair(MTTR),This thesis stduty from three aspects,modular design,hot-plug design and hardware partition,the main work are as follows:1.Study Intel Brickland platform architecture,propose modular design method and ensure the modular maintenance can be performed at the front or back side of chassis without opening the case.The realization of front or back chassis maintenance makes no requirement of off-rack maintenance,improves server serviceability and it is a necessary condition for hot plug and hardware partition.Modular design requires high speed interconnection solution and evaluates signal integrity with the technology of high speed signal simulation.2.For modular hot plug design,hot plug function achieves replacing failure components without stopping or shutting down the system.This thesis studies the hot plug design of CPU modular,IO modular,using CPU on-line and off line technology,memory and PCIe device hot plug technology and so on.3.For hardware partition design,this thesis studies partitioning the Brickland eight-processor server to two four-processor servers by hardware.The realization of the design can be applied to dual-server backup,and the system redundancy technology can reduce service interrupt time caused by system failure.Ensure system high reliability requirements in strict environoment.The implementation of the above design is verified and tested on Sugon I980 eight-processor server,the result indicates that the signal integrity meets the requirements of modular design.The solution of hot-plug and hardware partition can be realized.For solving the weakness of maintainability and improving the system high availability of x86 base SMP server,the design and implementation are of great significance.