A Method Of Improving The Reliability Of MINIX3 Block Device Driver

For traditional operating systems, studies have shown that device drivers and extensions contain 3 to 7 times more bugs than other code and thus lots of studies on the reliability of device drivers have been given. Applications that depend on the drivers, however, are not paid enough attention. Tanenbaum moved some kernel modules called servers and drivers into user space to improve the reliability of MINIX3, which's a multi-server, u-kernel operating system, provided a mechanism called RS to detect the failures of the servers and then reboot them if they failed.This article is dedicated on improving the reliability of MINIX3's block device driver by re-execute the command, which leads the driver to fail, after the driver has been rebooted by the RS. For re-executing the command, we must re-construct the context which makes the commands, whether are idempotent or not, be idempotent. Our model also assumes that the errors of the driver are transient errors. I'll describe the problems and try to figure out the solving-model during the implementation.