Yesterday I wrote a small program that scans the history database and
calculates the "lag time". That is, the difference in time between when
the article was posted (according to the Date: header) and the time it
actually arrived on our server.
75% of the articles we get are received within an hour of being posted.
Less than 5% arrive more than 24 hours late. That's pretty darn good.
All this is meaningless of course unless we can solve one key problem. I
recently had to reformat the article spool due to file system corruption,
presumably caused by the BC Hydro power outage a day or two earlier. Well
this morning I'm seeing more corruption of a similar nature. This tells
me that either the power outage actually physically harmed the disks (they
are brand new) or there's a bug in the OS that has to do with the way the
disks are striped.
Today I'm going to install a second set of SCSI hard disks in the server
and copy existing articles onto it (I'm trying to avoid starting with an
empty spool again if possible). If the new set starts to get corrupted
too, then it's obviously not a disk problem and I'll have to use the same
method of storing articles that we used on the previous server.
Mark