tag:blogger.com,1999:blog-4169710303065679169.post4909810862902100767..comments2024-03-18T21:07:31.080-07:00Comments on A Wider View : BLOG HAS MOVED *<a href="http://blog.orapub.com">HERE</a>*: SQL Statement Elapsed TimesCraig Shallahamer, President/Founder, OraPubhttp://www.blogger.com/profile/04109635337570098781noreply@blogger.comBlogger4125tag:blogger.com,1999:blog-4169710303065679169.post-30962552883301536802011-11-16T13:21:50.354-08:002011-11-16T13:21:50.354-08:00Craig,
re: Garr... histogram of SQL elapsed times...Craig,<br /><br />re: Garr... histogram of SQL elapsed times, you are quite correct to ask for some kind of statistical analysis to back up any claims about distribution profiles.<br /><br />In fact, it's way worse than you suspect in this case. <br /><br />Not only is it not "exp" or "log-normal," etc., I claim it involves <a href="http://perfdynamics.blogspot.com/2011/08/q-q-plots-for-multi-modal-performance.html" rel="nofollow">power laws</a>. Worse yet, I suspect it's not even a single distribution, but a <a href="http://perfdynamics.blogspot.com/2011/08/q-q-plots-for-multi-modal-performance.html" rel="nofollow">threefer</a>.Neil Guntherhttps://www.blogger.com/profile/11441377418482735926noreply@blogger.comtag:blogger.com,1999:blog-4169710303065679169.post-75867169759999959292011-06-13T18:14:46.002-07:002011-06-13T18:14:46.002-07:00I've recently learned a very powerful graphic ...I've recently learned a very powerful graphic method to compare distributions: quantile plots, ala Q-Q plots.<br /><br />I have created a little tutorial for this technique, using some of Craig's data above:<br /><br />http://aberdave.blogspot.com/2011/06/q-q-plots-to-examine-sql-execution-time.html<br /><br />Methods to prepare Q-Q plots are available in Mathematica and the R programming language.Dave Abercrombiehttps://www.blogger.com/profile/13480651551795199755noreply@blogger.comtag:blogger.com,1999:blog-4169710303065679169.post-30829007716766583122011-03-01T15:21:24.274-08:002011-03-01T15:21:24.274-08:00Very interesting Craig! I do agree that it's v...Very interesting Craig! I do agree that it's very common to look at average values without even considering what kind of distribution is behind the numbers. Next time I'll definitely have this in mind. I do believe the avarages still have some value, but the value must be weighted carefully.L. Jenssenhttps://www.blogger.com/profile/16945919361267764584noreply@blogger.comtag:blogger.com,1999:blog-4169710303065679169.post-74126178670985318632011-02-08T18:52:54.383-08:002011-02-08T18:52:54.383-08:00Interesting work. You point out different executio...Interesting work. You point out different execution plans and different bind sets as possible causes for non-conformance with a statistical distribution. Another factor could be the nature of the wait events. Waits for disk reads might have one distribution (and possibly one for sequential reads and one for scattered) while waits for locks may have another. Not sure how that could be tested though as they don't tie directly to SQLs.<br /><br />Looking forward to hearing you at the Oracle Meetup in Sydney.SydOraclehttps://www.blogger.com/profile/08828771074492585943noreply@blogger.com