Research And Design Of In-band Fault Detection System For Autonomic Computing

Autonomic computing was proposed in October2001by IBM,which wasdesigned to reduce the complexity and improve the system usability by the means oftechniques manage techniques. Fault detection is one of the most basic and corefunctions of autonomic computing,the in-band fault detection mainly turn towardsindependent compute nodes,the detection system runs above the OS,the detectionprocess consists of monitoring and part of the diagnostic features which was definedin autonomic computing.This paper do some research about the fault coverage and the detection validityof the in-band fault detection,define the fault classification criteria for computingunit.the paper try to solve the following two issues:building the general fault detectionmodel;designing and implementing fault detection tools.Fault detection model mainly try to solve fault detection,dynamic configurationand system queries interface of the object,This model standardize a variety ofdifferent detection obejects,detection processes and detection results so as to make thedetection system more configurable and flexible.Under the guidance of the fault detection model.This paper bring out a series offault detection toolsets which including the hardware layer,the numerical kernel,thenon-numeric kernel and the application layer.The four tools have full coverage of thesystem from top to bottom,from the services to the operating system and to thehardware status,from memory intesive application to compute-intensive applicationand then to IO-intensive application.At last,this paper do some verification of the fault detection toolsets based onLinux operating system and compare the fault detection results with other monitoringtools,at the same time,the paper simulate various fault conditions and take analysis ofthe detection results.The experiments show that these fault detection toolsets coulddetect the corresponding situations when certern objects conk generally and couldsend the corresponding detection information to the subsequent processing modules.