Design And Implementation Of Fault Injectors For Multiple Processors Target RAS Mechanism

Fault-tolerance computer system using for mission-critical applications needs fornot only a high processing performance, but the capability of fault tolerance. At present,computer manufactures have designed and implemented a lot of RAS mechanisms toimprove system’s fault tolerance capability. Evaluation fault tolerance considering RASmechanisms can provide targeted suggestions to vendors. Furthermore, doing this withfault injection methods may greatly improve the effectiveness without doubt. At thesame time, we can find the strengths and weaknesses of different architectures’ RASmechanism， if we can make a tests of all existing computer systems using the unifiedtest case.This article refers to the FARM model theory, first analyses the causes of hardwareerrors. Then, divide those errors into different ranks, making a fault set targeted RASmechanisms and design the module for fault tolerance testing. Following the think offault injection, which includes emulating errors at drives level and depending on theprogrammable testing ports, in this paper we’ll implement CPU fault injection tools andregister injection tools. These tools are fit for x86、ia64and sparc processors，Linux andSolaris operating systems, having the ability of injecting cache, TLB and register fileerrors for applications or kernel process.In order to validate the effectiveness of designed fault injection tools, this paperselect four different server systems to perform cache and register error injectionexperiments. The fault tolerant capability of target system is analyzed from the faultinjection results. It has proved that these fault injection tools are effective to evaluatefour different systems using the same test case. During the experiments, we haveanalyzed the advantages and disadvantages of MCA mechanism and PFA mechanism.According to the result of register fault injection experiments on different serversystems, we came to the conclusion that improving the system’s fault tolerantcapabilities need for the hardware with appropriate operating system.