Re: [PHP4BETA] ZTS mode sooo much slower?
| From: | Zeev Suraski | Date: | Mon, 06 Sep 1999 21:09:58 +0000 |
| Subject: | Re: [PHP4BETA] ZTS mode sooo much slower? | ||
| References: | 1 | Groups: | php.version4 |
| Request: | Send a blank email to php-version4+get-4170@lists.php.net to get a copy of this message | ||
At 23:03 06/09/1999 , Rasmus Lerdorf wrote:
We don't really see the performance hit in the stats Thies sent (unless I missed something). Thies - can you send the stats for the thread unsafe version? Zeev -- Zeev Suraski <zeev@zend.com> http://www.zend.com/ For a PGP public key, finger bourbon@netvision.net.iljust did soem "benchmarking" of zts mode agains "plain" mode on unix (solaris gcc 2.8.1) - this test-case just has no real code running - just included > 20 phpfiles.It could be that the C++ compiler is much less efficient and doesn't know how to optimize very well. On a busy site, this can become negligible with the persistent registry. That's a pretty nasty hit though. Perhaps we are not getting the full 8K buffer filled up in the input stream for some reason? Did something kick us into unbuffered mode? Would be silly, but could be one way to get such a huge performance hit.ZTS non-ZTS real 4.20 0.57 user 3.98 0.45 sys 0.05 0.10quantify tells me that the c++ parser eats most of the time. (% of .root.) 63% zendFlexLexer::yy_get_previous_state() 12% ZendFlexLexer::lex_scan(...) 7% zendFlexLexer::yy_get_next_buffer(...) 2% zendparse() in "not-ZTS-mode" zendparse is the "busiest" function... are flex-generated c++ lexers known to be *SLOW* or am i doing something wrong?