This is one of the slides copyed from Adam Leventhal''s "Performance Tuning Linux Applications With DTrace". But I don''t know what it tells. Regards! TJ -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://mail.opensolaris.org/pipermail/dtrace-discuss/attachments/20070322/6f954e7e/attachment.html> -------------- next part -------------- A non-text attachment was scrubbed... Name: unknown.jpg Type: image/jpeg Size: 113506 bytes Desc: not available URL: <http://mail.opensolaris.org/pipermail/dtrace-discuss/attachments/20070322/6f954e7e/attachment.jpg>
???? (Euler Tao) wrote:> This is one of the slides copyed from Adam Leventhal''s "Performance > Tuning Linux > Applications With DTrace". > > But I don''t know what it tells.What I think it indicates (assuming that Adam will chime in with the real story ....) is the speed with which you can use DTrace to provide meaningful performance improvements on production environment. The words in orange would be an indication of how long it took to achieve the percentage performance gain. Eg, for a financial db, a 32% performance increase was achieved before lunch ... and I assume that this would be on the first day of an engagement. Or the data routing application - a 300% improvement after 5 hours worth of work. Not bad at all! Adam - do I win a prize? :-) cheers, James C. McPherson -- Solaris kernel software engineer, system admin and troubleshooter http://www.jmcp.homeunix.com/blog Find me on LinkedIn @ http://www.linkedin.com/in/jamescmcpherson
On Thu, Mar 22, 2007 at 11:25:08PM +1100, James C. McPherson wrote:> What I think it indicates (assuming that Adam will chime in with the > real story ....) is the speed with which you can use DTrace to provide > meaningful performance improvements on production environment.Precisely. I should mention that it''s not my slide. Johnny L (now of Adobe) used the slide at the JavaOne keynote in 2005 which also featured DTrace for Java and my enormous afro.> Adam - do I win a prize? :-)Um. No. Adam -- Adam Leventhal, Solaris Kernel Development http://blogs.sun.com/ahl
On 3/22/07, Adam Leventhal <ahl at eng.sun.com> wrote:> > Adam - do I win a prize? :-) > > Um. No.How about getting a T-shirt: http://openyourmindtoday.com/openballot.php Rayson> > Adam > > -- > Adam Leventhal, Solaris Kernel Development http://blogs.sun.com/ahl > _______________________________________________ > dtrace-discuss mailing list > dtrace-discuss at opensolaris.org >
James C. McPherson''s email at 3/22/2007 7:25 AM, said:> ???? (Euler Tao) wrote: >> This is one of the slides copyed from Adam Leventhal''s "Performance >> Tuning Linux >> Applications With DTrace". >> >> But I don''t know what it tells. > > What I think it indicates (assuming that Adam will chime in with the > real story ....) is the speed with which you can use DTrace to provide > meaningful performance improvements on production environment. > > The words in orange would be an indication of how long it took to > achieve the percentage performance gain. Eg, for a financial db, > a 32% performance increase was achieved before lunch ... and I assume > that this would be on the first day of an engagement.You are correct in your description, and all times are measured from the arrival time on-site. For these five cases, the time includes not only getting the problem description and the DTrace analysis, but implementing the corrective action as well.> > Or the data routing application - a 300% improvement after 5 hours > worth of work.In this case, (in the interest of full disclosure) the 300% was identified and corrected, but the improvement was measured after market hours using a test harness. The win in production happened the next day at market open. But then again, the analysis part only took about an hour. The fourth one is kinda funny too since the slide says production and the application says "benchmark". Indeed the analysis was done using a stage environment with a test harness, but the code was the same and moved to production the next day. Okay, I know what some of you are thinking - "Who, in their right mind, would make these changes in production so fast?" Well, if you are losing money, and there is a way to stop the hemorrhaging, you''ll do it. I don''t advocate production deployments this fast, but sometimes it is a necessity. I wish confidentiality wasn''t what it is and I could disclose the details, but alas, I am scared of attorneys. Also, there are some bigger wins as well, but we didn''t think anyone would believe it in slideware. Jarod
Jarod Jenson wrote:> James C. McPherson''s email at 3/22/2007 7:25 AM, said:[snip]>> Or the data routing application - a 300% improvement after 5 hours >> worth of work. > > In this case, (in the interest of full disclosure) the 300% was > identified and corrected, but the improvement was measured after market > hours using a test harness. The win in production happened the next day > at market open. But then again, the analysis part only took about an hour. > > The fourth one is kinda funny too since the slide says production and > the application says "benchmark". Indeed the analysis was done using a > stage environment with a test harness, but the code was the same and > moved to production the next day. > > Okay, I know what some of you are thinking - "Who, in their right mind, > would make these changes in production so fast?" Well, if you are losing > money, and there is a way to stop the hemorrhaging, you''ll do it. I > don''t advocate production deployments this fast, but sometimes it is a > necessity.Which points to the benefit of being able to use DTrace - if there''s a faster way to achieve a performance improvement, then one should definitely make use of it. Being open to change in performance analysis is a major success criterion.> I wish confidentiality wasn''t what it is and I could disclose the > details, but alas, I am scared of attorneys. Also, there are some bigger > wins as well, but we didn''t think anyone would believe it in slideware.Any tantalizing hints you can throw our way though? James C. McPherson -- Solaris kernel software engineer Sun Microsystems