[PLUG] Facebook & PHP - massive speedup

[Eric Wilhelm]

# from Carlos Konstanski
# on Thursday 04 February 2010 12:37:

>There are a host of commonly-used interpreted languages today that are
>slow because they are interpreted. BASH, perl, python, ruby,

Every one of those interpreters is/was written in C and compiled into 
machine code.  If your compiled program is going to do the same amount 
of work as the interpeter, it becomes the interpreter (definitions 
being what they are and such.)

If you can take a shortcut by dropping a bunch of features from your 
runtime, of course the resultant machine code will be faster.  The 
engineering decision is simply about how long it takes you to code that 
shortcut vs what it's worth at runtime.

Meanwhile, vast quantities of XS extensions are available on the CPAN 
(e.g. dbi, expat, libusb, libgsl, wx, gtk, qt) which let you run the 
heavy lifting through some compiled machine code (at the expense of a 
bit of marshalling overhead) while you happily write more expressive 
code without worrying about memory allocation or buffer overflows.

You can generalize about compiled vs interpreted but it really doesn't 
count for much by the time you pile 100 other trade-offs into the 
decision.

>Let's see the list of counterexamples.

Your unoptimized regular expression engine vs Perl's is an easy one.  
I'll grant that most of the others are hard to win on time benchmarks. 
And that's fine if you get infinite time to write code, debug it, 
maintain it, and have zero cost for downtime or security breaches.  
Spherical horserace in a vacuum and all that.

> FB is making their PHP run faster by compiling it. Why would they do
> that if the slowness of their interpreted language, and the benefit
> gained by compiling it, is a myth?

It's always faster to do less work.  For the machine, there are cases 
where you can accomplish this without giving up the expressiveness and 
dynamic code.  Some JIT work promises to outrun naively compiled C 
because it can examine a running loop to e.g. remove several 
conditionals.  There are more and more shades of gray between compiled 
and interpreted as time goes on.

From what I've read about the facebook php, they're simply caching 
another step in the process.  Interesting tech, but probably vastly 
different in lifetime cost compared to rewriting everything in C++.

"Myth" isn't the right word to describe the typical gross over- 
simplification of comparing dynamic/static interpreted/compiled code, 
but bugs per KSLOC and SLOC/day/developer are still very cruel 
constants.

If you're interested in the shades of gray and some pretty nifty stuff 
happening in the world of VMs, please join us next week at pdx.pm for 
an introduction to parrot.  http://pdx.pm.org/  :-D

--Eric
-- 
Atavism  n:  The recurrence of any peculiarity or disease of an ancestor
in a subsequent generation, usually due to genetic recombination.
---------------------------------------------------
    http://scratchcomputing.com
---------------------------------------------------