Development of Fault Tolerant Linux Kernel 2.6 with Backward Error Recovery

Abstract

Checkpointing is one of the important techniques in fault-tolerance. It is a simple technique known as a backward error recovery. The state of an executing single process is periodically saved into a stable hard disk as a file, and this file is used to recover the failed process after a failure of the executing single process. CKPT is one of the checkpointing techniques. CKPT is modified from KCKPT [1], which is developed on UnixWare, for using it in Linux machine. In addition, CKPT is the mother of ICKPT [2], PICKPT [3,4], and APICKPT [5]. However, CKPT is available only in Linux kernel 2.4 because it was developed on Linux kernel 2.4 environment, It is not applicable to Linux kernel 2.6. In this paper, we address the differences between Linux kernel 2.4 and 2.6 which cause the compatibility issue when CKPT is used on Linux kernel 2.6. Two different types of modifications are needed to make CKPT compatible with Linux kernel 2.6. A modification can be made to the Linux kernel 2.6 and the other is to modify the CKPT source. Additionally, we develop NCKPT, a CKPT which is compatible with Linux kernel 2.6. NCKPT provides compatibility with Linux kernel 2.6. We evaluate the performance of NCKPT by comparing with the performance of CKPT. The main concept of NCKPT is derived from CKPT.