Monthly Archives: March 2014

Yet Another HHVM Benchmark Test

Since my last post comparing Go vs Node vs HHVM, I decided I would do it properly.

First off, the previous test didn’t really sort the array one million times, it merely sorted once then iterated over the sorted array 999,999 times. Also, my previous test focussed on HHVM & PHP with OPCache and trusted the results from @jaxbox for the Go & Node. Given I’ve fixed the sorting of the array loop, I had to go and run the node & go tests myself. Thanks to my colleague at work for providing me a C version for comparison!

The code:

PHP:

C:

JS:

Go:

Java:

results.txt

C
$ time ./a.out

real 0m0.175s
real 0m0.172s
real 0m0.176s
real 0m0.170s
real 0m0.171s

=================

GO
$ time ./bubble

real 0m0.313s
real 0m0.313s
real 0m0.314s
real 0m0.313s
real 0m0.312s

=================

NODE
$ time node bubble.js

real 0m0.484s
real 0m0.489s
real 0m0.491s
real 0m0.489s
real 0m0.496s

=================

HHVM
$ time hhvm -v Eval.Jit=true hhvm.php

real 0m2.622s
real 0m2.657s
real 0m2.615s
real 0m2.662s
real 0m2.599s

=================

PHP 5.5 OPcache

real 0m30.421s
real 0m27.603s
real 0m27.695s
real 0m27.336s
real 0m27.706s

=================

Java

real 0m0.750s
real 0m0.717s
real 0m0.686s
real 0m0.709s
real 0m0.689s

Sorry for the crappy formatting of the results above. Maybe one day I shall put them in a table.

But basically you can see that in terms of bubble sorting; C > Go > Node > HHVM > PHP

However you look at it, I think it’s a no brainer to use HHVM!

Summary of best times:

Lang Best Time
C 00.171s
Go 00.312s
Node 00.484s
Java 00.689s
HHVM 02.590s
PHP 5.5.8 w/OPCache 27.336s
Share

HHVM vs PHP vs OPCache vs Go vs Node

Lets start by saying, I was just curious as to how fast Go was compared to PHP. I knew it was faster, but wanted some benchmarks.

I found this great article by @jaxbot:
http://jaxbot.me/articles/benchmarks_nodejs_vs_go_vs_php_3_14_2013

He then went on to dismiss PHP as it was too slow; so didn’t include it in the Go 1.1 benchmark test he did later:
http://jaxbot.me/articles/benchmarks_nodejs_vs_go_11_5_27_2013

Fair game I thought. Until a colleague pondered how fast HHVM would be as he’d recently installed it. Given how easy it is to install HHVM these days I decided to do my own benchmarking.

For Node & Go benchmarks, please refer to @jaxbox’s links above.

There is a minor difference in the code – and that is I am using a function to wrap the code. PHP is generally faster to execute code when it’s in a function, plus HHVM ignores code that isn’t in a function.

The code:

PHP 5.5 (opcache.enable_cli=0)
$ php bubble.php
Run 1: 24.121762990952
Run 2: 24.156540155411
Run 3: 24.948321819305
Run 4: 26.411414861679
Run 5: 24.790290117264
Average: 24.882

PHP 5.5 w/OPCache (opcache.enable_cli=1)
$ php bubble.php
Run 1: 24.675834178925
Run 2: 25.641896009445
Run 3: 26.468472003937
Run 4: 24.278208017349
Run 5: 24.843347072601

HHVM (PHP 5.5 OPCache)
$ hhvm -v Eval.Jit=true bubble.php
Run 1: 2.6463210582733
Run 2: 2.6204199790955
Run 3: 2.563747882843
Run 4: 2.9089078903198
Run 5: 2.6408560276031
Average: 2.672
Interesting is that OPcache didn't fair well in this bubblesort...

Now let's compare to the tests from @jaxbot - he had different hardware, so a direct comparison is almost meaningless.. BUT HHVM speaks for itself when compared to the Zend Engine.

  Node.js Go 1.1 PHP 5.5 HHVM jit
Avg of 5 trials 430ms 326.26ms 24.88ms 2.67ms
Best 420ms 290.27ms 24.12ms 2.56ms

Hope that helps someone who wonders how good HHVM is?! 🙂

I finish with a quote from @jaxbox: "the same rules apply here as last time; take these with a lattice of Sodium Chloride, as the benchmarks tell us information, not answers."

Share