The TOP - 5 wait event section.
A very important point to keep in mind is, irrespective of whether
the performance is good or bad, we would see some or the other
wait events in this list, when there is a problem we check this
list and see which were the top 5 waits which just gives an idea
what the database was doing for the given snap interval, most often
we will have to drill down further to find what is causing the
sessions to wait for such a long time, or why suddenly there were
too many waits for a wait event, and try to correlate this with the
problem symptoms and operations running on the database during the
snapshot interval.
So to say, just looking at the statspack will not always reveal a
performance problem, and this is the reason we always would see more than
one report to compare and see the trend, once you know the problem we would
make the snapshot interval smaller to see how the pattern is and identify
the cause.
Every column in the top-5 event is important, what matters is how has
the trend changed since the time the performance was good to now when the
performance is bad.
for example if you see lots of waits for "enqueue", first identify if there
was a time when the events were never seen or the number of waits has increased,
When number of waits increase it could indicate that many sessions are requesting
the same enqueue, in general we know the concurrency(no. of users) has increased.
There is really no document available which will help you to analyze
a statspack report because most part of it is self explanatory.
You can still read these articles which provide a great insight into the usage of the tool
http://jonathanlewis.wordpress.com/2006/11/29/analysing-statspack-pt1/
>> With respect to the Top SQLs, the sections are ordered by GETS, CPU
PARSE, DISK READS etc, again this should be used based on understanding
the problem and also the rest of the data in the report, once you know
for sure there is a lot of logical reads / buffer gets happening, you may
want to see the Top SQLs under Gets section and again you should establish
a pattern, like with Top 5 events, any SQL having High reads compared to others
in the given interval will be listed here(please note reading 2 blocks is greater
than reading 1 block however is really not a problem compared to 1000 Vs 10).
Thursday, March 12, 2009
A note on Statspack report
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment