Re: Question on concurrency wait time
Date: Mon, 24 Oct 2022 20:12:55 +0100
Message-ID: <CAGtsp8ksOFQd8Eo0VYXfU14nLuGP5HyfvM=iGnt6CgUkN8TWXw_at_mail.gmail.com>
I am a little curious about the "unix_time_id". It looks like 4 bytes for
the number of seconds since 1st Jan 1970 followed by 16 bytes (maybe a
digest of the row) stored as a character string.
Two things (for strategic planning - this is nothing to do with the
performance glitch).
Regards
P.S.
a) Why not store it as a RAW and save 20 bytes per row and 20 bytes per
index entry.
b) Is the unix time component the same as the part_date_time (with a
deterministic adjustment, perhaps, for daylight savings time). If so could
you do something to make the partition key a virtual column and eliminate
on column from the index.
Jonathan Lewis
While I agree with Stefan's analysis of the wait chains, there's one
further layer to consider.
You said: "but we see the concurrency during the time when there is heavy
activity happening on the database from another app."
The overrun on the change tracking buffer MIGHT be the consequence of the
CTWR process having to wait to get on the run queue because of CPU overload
-- http://www.freelists.org/webpage/oracle-lReceived on Mon Oct 24 2022 - 21:12:55 CEST