This seems very misleading to me:
The other item that caught my eye was Cameron Purdy noting that Caucho claims to have a module that compiles PHP to bytecodes and runs it on the JVM four times faster than mod_php (first benchmark, but on a real app not synthetic). It’s GPL’ed. This is more than a little surprising.
Not really. It's not a direct comparison. First, PHP scripts are typically compiled to bytecode on each page execution. In addition, the Java app server they're running this on surely has a cache involved so that the compiled classes are kept in memory and aren't loaded on each page hit.
So, it's actually much closer to PHP when you're using one of PHP's bytecode caches, such as PHP Accelerator, which, what do you know, claims a 5-10 times speedup over normal PHP. Someone else even asks whether they were comparing it to PHP with a bytecode cache, but no response from Caucho.
However, Scott Ferguson (from Caucho) correctly points out that we should expect Java to be faster at certain things compared to PHP simply because of the semantics of the respective languages. Of course that's true. Plus, Java's obviously had way more effort and money thrown at its optimization than PHP has, and it even JIT's down to native code. So all of this makes it even more surprising that Java is only four times faster than Apache/mod_php if in fact they were comparing it to vanilla PHP with no bytecode cache.
Plus, as this is a comparison of server stacks and not merely language implementations, it's important to point out that Apache certainly isn't known for its speed. PHP would likely have performed much better running on lighttpd with FastCGI, as it's almost certain Caucho wasn't merely comparing code speed but requests-per-second. See these posts at TextDrive's weblog for more on lighttpd.
Wednesday, December 21, 2005
Permalink
The guys building php.NET (http://www.php-compiler.net/) also claim better performance: http://www.php-compiler.net/Benchmarks.htm