In <1994Dec1.215051.24403 AT draper DOT com>, spikes AT vnet.ibm DOT com (Jim 
Smith) writes:
>Following-up on Greg's note:
>
>There have been several users who have reported an unprecedented amount of
>recovery log usage.  It seems that there are certain times when the
>usage 'spikes' when nothing out of the ordinary is going on.
>
My log started small (50 mb) but I had to expand it quickly to 232 mb
(58,880 pages).  The reason was that one node, a Mac on Appletalk, had
to backup 300 MB files.  This session would run for 24 hours.  While
the big file was backing up, the log grew and grew.  I don't know if
all the growth was due to just this session or if all other sessions
backed up behind (this is, if the commits and log flushes were
delayed - other client backups did start and end).  This was 'out of
the ordinary', but couldn't be addressed any other way than to give it
the required log space.
>
>We are currently trying to determine what is causing these 'spikes'.  We have
>a couple of users who are automatically issuing Q LOG commands every 15 minutes
>to determine when the 'spikes' are occuring and then trying to use the activity
>log to determine what could be causing the usage.  We believe that there is
>still some process that is not grouping transactions in a 'nice' manner and
>is using more recovery log space then need be.
>
>If you think that you are seeing these unexplainable 'spikes', please contact
>us (the support center); we would love to get as much information as possible
>and try to get at the bottom of this ...
>
>Jim Smith / ADSM Technical Support / San Jose
>
It is hard to determine what is normal when the server operation is a
black box.  We really don't know how the database and log operate.
For instance, I have never seen a published statement of what a
'transaction' is.

Bill Colwell
C. S. Draper Lab
Email: BColwell AT draper DOT com
Voice: 617-258-1550