Benchmarks

Warning

The following benchmarks does not intend to prove anything but might help understand the possible overheads when using the bridge. They were designed to illustrate the cost of creating objects and calling methods (roundtrips).

Be sure to have read the how it works before.

Simple benchmark

Machine: Laptop i7-6700HQ 2.60GHz, Tomcat8, japha 1.2.0, OracleJDK8, pjb7.0.0 and php7.0-fpm on ubuntu xenial. Test script: simple_benchmark.php. Connection time: $ba = new BridgeAdapter([]) varies between around 2ms and 7ms

Benchmark name	x1	x100	x1000	x10000	Average	Memory
New java(java.lang.String, "One")	0.14ms	4.09ms	34.17ms	282.30ms	0.03ms	12.37Kb
New java(java.math.BigInteger, 1)	0.05ms	3.28ms	32.51ms	308.27ms	0.03ms	0.37Kb
javaClass(java.sql.DriverManager)	0.09ms	0.02ms	0.14ms	1.69ms	0.00ms	3.12Kb
Method call java.lang.String->length()	0.05ms	2.14ms	21.60ms	226.35ms	0.02ms	0.34Kb
Method call String->concat("hello")	0.08ms	2.59ms	27.30ms	287.79ms	0.03ms	2.09Kb
$a = ...String->concat('hello') . ' world'	0.09ms	5.81ms	54.64ms	532.68ms	0.05ms	0.42Kb
New java(java.util.HashMap, $arr)	0.16ms	3.74ms	33.78ms	351.76ms	0.04ms	67.05Kb
Method call HashMap->get('arrKey')	0.04ms	2.71ms	23.37ms	267.88ms	0.03ms	0.39Kb
Call (string) HashMap->get('arrKey')[0]	0.08ms	5.98ms	56.99ms	566.69ms	0.06ms	0.37Kb
Iterate HashMap->get('arrKey')[0]`	0.23ms	13.40ms	133.93ms	1,252.59ms	0.13ms	2.52Kb
GetValues on HashMap	0.05ms	3.67ms	36.22ms	368.43ms	0.04ms	1.27Kb
New java(HashMap(array_fill(0, 100, true)))	0.20ms	12.04ms	122.62ms	1,202.73ms	0.12ms	0.63Kb
Pure PHP: call PHP strlen() method	0.00ms	0.00ms	0.01ms	0.07ms	0.00ms	0.37Kb
Pure PHP: concat '$string . "hello"'	0.00ms	0.00ms	0.04ms	0.31ms	0.00ms	120.37Kb

Note

Memory and average time are computed on the 11101 iterations (x1, x100...).

You'll realize that average time is always lower than x1 (or x10). There's some optimzations happening on subsequent calls that make very difficult to give sense to an average here. The best is to always refer to x1 for worst-case scenario timing.

Memory does not include the JVM side, that explains differences from pure php tests and Java one.

The figures above will vary between systems, but intuitively you might get a glimpse about how the bridge is sensitive to the number of object creations and method calls (roundtrips):

(connection time) + (number of created objects) + (number of methods) + (eventual result parsing).

Imagine a very simple case with 100 objects instantiations and 100 method calls (from the PHP side):

4ms (connection) + 4.09ms (100 new strings) + 5.81ms (100 concat methods) = ± 13ms minimal overhead (looks fine).

Imagine a bad scenario with 1.000 new objects and 10.000 method calls:

4ms (connection) + 34.1ms (1000 new objects) + 532.68ms (10000 concat methods) = ± 570ms overhead (looks too much).

The second example should be avoided if performance matters, but the first one looks not only viable but a (micro-)service would probably not do better (parsing the result might give differences - a json_decode() vs parsing bridge response... But eventually you can also get the json from the bridge as well).

As an example, generating a report with Jasper will not even require more than 10 objects and at max 100 method calls. The overhead here is clearly insignificant.