Why is including multiple files so much slower than bundling into one

Question

I have been doing some tests with PHP, and I have noticed that performing include on multiple files compare to one containing all the functions is much slower.

My test involved creating 1025 files, 1024 of which contained the text <?php class cls$i {} ?> (where $i was the file number), and 1 file that was a concatenation of all the text in the files before. I then had two functions, one for testing each case. The test that loaded the single file took about 6ms to compile to bytecode and make the contents available to the system, however, the combination of 1024 files took almost 600ms .

In terms of size, the 1024 individual files are exactly the same size as the single file. I am also running APC to cache the bytecode, but in practice, it only shaves off a few milliseconds.

I also created a ramdisk which held all the files, but that was only marginally (10ms on average) faster.

So, having said that, why are individual files SOOOOO much slower than a single file? Is it down to significant inefficiencies in the loading engine within PHP, or have I made a considerable cockup in the configuration (on my local system, a standard AMPPS installation)?

Answer 1

My first guess is that it's all the stat system calls going on.

What happens if you turn apc.stat off in your config?

Even though APC caches the bytecode, if apc.stat=1 , it needs to check the mtime on each file via the filesystem to make sure it doesn't need to be re-read.

EDIT : going deeper, in response to your comment. How are you referencing the included files? If you're using relative paths, then you're getting the include_path involved.

In other words:

include "somefile.php";

is likely to be slower than

include __DIR__ . '/otherfile.php';