Evaluation of the Device Driver Availability in Dawning4000A

摘要

Device drivers were claimed to be the most error prone in kernel source. A lot of error tolerance or error prevention approaches have been developed or suggested after this claim. But after analyzing the event log and maintenance record of Dawning4000A for three month, we find that device driver errors are not the most crucial crash causes in this previous TOP10 supercomputer. We believe device driver errors need developing and debugging efforts, rather than tolerance. We also suggest drivers to achieve better tolerance to device errors, especially on storage device.