No subject

[utzoo!decvax!ucbvax!unix-wizards]

>From decvax!duke!reed!valer at Berkeley Sat Dec 26 18:01:42 1981
To deal with DEC CE's who accuse UNIX of not logging errors well
enough, Reed College invented a logging device to record errors.
It works by stashing records (generated by variable length
argument lists of <bytecount,address> pairs) into clists,
which can then be read out of a character special file.
The device driver is simple (100 lines) with most of its
work dealing with running out of clists at inopportune moments.
Our usage is to record the time, device registers, buffer
struct, and device buffer struct when disk errors occur.  Since most
errors are recoverable in the next retry, we can fairly safely
then record them back on the disk (and if not, it is no real
loss).  Thus, the binary data is saved by running
	cat -u /dev/log >>/usr/adm/diskerror
from /etc/rc.  When DEC arrives, we pretty-print the binary
data providing bit and field definitions of all the structures.
As evidence it is quite overwhelming, permitting analysis by
track and sector, frequency analysis by time of day, or
whatever.  Unfortunately, it does not generally tell them how
to fix the problem but only forstalls excuses.