tag:blogger.com,1999:blog-4169710303065679169.post8826979983153911614..comments2024-03-28T03:53:18.327-07:00Comments on A Wider View : BLOG HAS MOVED *<a href="http://blog.orapub.com">HERE</a>*: Gotta love latch classesCraig Shallahamer, President/Founder, OraPubhttp://www.blogger.com/profile/04109635337570098781noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-4169710303065679169.post-64255757761227090622012-08-30T05:03:07.233-07:002012-08-30T05:03:07.233-07:00Hello!
Thank you for this great post!
It was the ...Hello!<br /><br />Thank you for this great post!<br />It was the starting point of my further investigation of latch classes. <br />Look at:<br /><br />http://andreynikolaev.wordpress.com/2011/01/18/spin-tales-part-3-non-standard-latch-classes-in-oracle-9-2-11g/<br /><br />I would like to add that:<br />1) Assignment of latch to the non-standard class completely changes the way Oracle uses to wait for the latch.<br /><br />2) My experiments showed that _spin_count tuning may be very effective for some performance problems.<br /><br />With Best Regards<br />Andrey Nikolaev<br />http://andreynikolaev.wordpress.com<br />Andrey Nikolaevhttps://www.blogger.com/profile/12111784821702695851noreply@blogger.comtag:blogger.com,1999:blog-4169710303065679169.post-72630993461368834402010-02-16T12:04:08.618-08:002010-02-16T12:04:08.618-08:00Timur, Thanks for your comments. Great stuff!
1) ...Timur, Thanks for your comments. Great stuff!<br /><br />1) I chose to measure the change in Oracle response time (cpu consumption and wait time) and the workload (logical reads) in very Oracle specific terms because it is pure and can be quantified. If I had added a "transaction" or "end user response time" into the mix, the situation and impact becomes much less clear AND I could have created an experiment to prove whatever point I wanted to make. The trick is to take what I demonstrated and understand it and relate it to your real life situation. For example, if what the user is experiencing is highly correlated with acquiring and holding a CBC latch time, changes in CPU consumption, and logical read activity, then altering spin count is something to consider. Otherwise, it is not relevant.<br /><br />2) I shared your concern about altering spin count. The overall message of this post is that changing spin count can make a difference (I demonstrated this), but ALSO that it is also highly unlikely making a change will ever be justified in a real production environment (demonstrated by the text in point #1 and #2 near the top of the post).<br /><br />Thanks again for your comments.Craig Shallahamer, President/Founder, OraPubhttps://www.blogger.com/profile/04109635337570098781noreply@blogger.comtag:blogger.com,1999:blog-4169710303065679169.post-7580977310634474072010-02-05T10:22:18.298-08:002010-02-05T10:22:18.298-08:00Hello Craig,
couple of thoughts
1) could you ple...Hello Craig,<br /><br />couple of thoughts<br /><br />1) could you please add more info on what your gains were after applying non-standard CBC latch spin count, in more convenient units of end-user response time & transactions per second. I think it would be much more representative than some Oracle numbers.<br /><br />2) I'm slightly concerned on the overall message of this post. I hope it won't be considered as a "way to go" approach - since far too often CBC latch contention is caused by inappropriate design and/or code (i.e. by application itself) rather than Oracle's configuration, I bet changing spin count for CBC latches is like "flogging a dead horse" for 99% sites.Timur Akhmadeevhttps://www.blogger.com/profile/15263848611633500971noreply@blogger.com