In addition to dictating our choice of failure types, our use of crash logs has two additional consequences. First, we observe only failures that cause system crashes, and not a broader class of failures that includes application crashes or non-crash errors. Second, we cannot observe failures due to permanent faults, since a machine with a permanent, crash-inducing fault would be unable to recover to produce, record, and transmit a crash log.
They only include systems that have had a hard crash or a stop error.
They only include systems that have had a hard crash or a stop error.