Re: GOOD Benchmark Results for PHP Master 2016-03-18

Hi Yasuo, ----- Original Message ----- From: "Yasuo Ohgaki" Sent: Tuesday, March 22, 2016

Hi all, On Fri, Mar 18, 2016 at 9:30 PM, <lp_benchmark_robot@intel.com> wrote:

-------------------------------------------------------------------------------------------

                        benchmark   relative   change since   change since  current rev run
                                    std_dev*       last run baseline         with PGO

-------------------------------------------------------------------------------------------

:-|   Wordpress 4.2.2 cgi -T10000      0.18%         -0.19% 0.39%            7.67%
:-|       Drupal 7.36 cgi -T10000 7%         -0.32%         -0.64%            5.21%
:-|   MediaWiki 1.23.9 cgi -T5000      0.14%         -0.44% 0.89%            4.04%
:-)           bench.php cgi -T100      0.00%         17.92% 19.80%           -0.67%
:-)      micro_bench.php cgi -T10      0.01%          2.90% 5.37%            3.31%
:-)      mandelbrot.php cgi -T100      0.03%         37.69% 28.50%            9.80%

I'm just curious. Is this due to huge page change?

No, it's from enabling "DFA pass by default" in the optimizer: http://git.php.net/?p=php-src.git;a=commitdiff;h=681de725542cbd776a1b3579c6778780c727d2ad I was going to make a comment when I saw these results also. It helped the synthetic stuff a LOT, and more than corrected the mandelbrot regression (since 7.0), but not "real apps." Still nice to see for comparing with HHVM. :-) And of course it helps isolated areas of different apps, and more optimizations will be added... The "huge page" change you're referring to is Rasmus disabling them by default now? If Intel doesn't enable them again, it should show *reduced* performance on Wordpress, etc. (1% or more? Can't remember.)

Thanks. -- Yasuo Ohgaki yohgaki@ohgaki.net

- Matt