MEMDOG: An Online Memory Tester Based On Linux

With the decrease of size and the increase of capacity,the density of memory chips increases,which makes them more and more vulnerable to external factors,such as high temperature,cosmic rays and even dust.The most common type of memory chip errors is one or more bits got flipped in a memory unit.These errors can cause the operating system(OS)and applications crash.Even worse,if these errors are utilized by the transaction processes without knowing them,which are silent errors,the result would be more serious.Therefore,it is necessary to study the reliability of memory chips.MEMDOG can protect applications from memory chip errors proactively.It constructs a safe memory pool,which contains the memory that qualifies the memory test without hardware errors recently.Applications request memory from the safe memory pool to guarantee that all memory they use is errorless.MEMDOG consists of memory test algorithm framework,safe memory pool,memory migration of applications and timer.Memory test algorithm framework uses some flexible interfaces,which allow related algorithms implemented in software level to integrate into the whole system easily.Safe memory pool is used to store the tested memory without errors.Since memory chip errors can happen as time goes by,the memory migration part can migrate the code and data of applications to the new memory allocated from the safe memory pool periodically with the help of timer.The timer is also used to periodically flush the out-of-date memory in the safe memory pool and return them to the Linux memory management system.This thesis also evaluates the effectiveness and efficiency of MEMDOG.The results show that MEMDOG can protect applications free from memory errors effectively and the performance of MEMDOG is acceptable.The main innovation of this thesis is the implementation of memory test algorithm framework in the kernel space which collects memory test algorithms.Since the framework is implemented in the kernel space,it is more efficient than the memory tester implemented in the user space.The efforts of this thesis include the analysis of the mechanisms of increasing the reliability of memory,and the design,implementation and evaluation of MEMDOG in Linux.