Improve base performance

[motin]

According to a recent benchmark, Propel2 does not perform well in the most basic conditions when compared to other popular ORMs:

orm	time (ms)	memory (KB)
doctrine	77.57	1297.48
propel2	36.83	1178.08
eloquent	25.89	671.20
yii1	14.14	800.39
fuel	9.26	381.07
yii2	7.20	818.16
phalcon	5.66	149.42

The executed code that took almost 37 ms to execute was:

$post = PostQuery::create()->findPk($id);

echo $post->getTitle() . '<br>' . "\n";
$comments = $post->getComments();
echo $comments[0]->getBody() . '<br>' . "\n";

This should be reduced so that Propel 2 remains competitive in simple benchmarks like the one mentioned above.

[motin]

There was an old discussion on the subject (#478), but it did not include the ORMs above that were found faster than Propel.

Another benchmark (https://github.com/Big-Shark/forked-php-orm-benchmark): Results updated 2015-12-08:

PHP CLI 5.6.4 with opcode cache

Library	Insert	findPk	complex	hydrate	with	memory usage	time
PDO	49	45	0	33	92	775,264	0.22
LessQL	153	154	0	114	135	5,232,880	0.57
YiiM	211	138	0	79	201	9,961,472	0.67
YiiMWithCache	217	148	0	88	222	9,961,472	0.71
Yii2M	603	338	0	187	252	14,155,776	1.46
Yii2MArrayHydrate	585	327	0	120	220	14,155,776	1.31
Yii2MScalarHydrate	571	321	0	105	102	14,155,776	1.17
Propel20	218	105	0	263	323	10,747,904	0.98
Propel20WithCache	156	68	0	228	261	10,747,904	0.77
Propel20FormatOnDemand	151	67	0	221	264	11,010,048	0.76
DoctrineM	252	280	0	342	193	17,301,504	1.55
DoctrineMWithCache	266	272	0	332	189	17,039,360	1.54
DoctrineMArrayHydrate	263	282	0	226	141	17,825,792	1.35
DoctrineMScalarHydrate	242	292	0	192	123	17,825,792	1.25
DoctrineMWithoutProxies	243	266	0	253	251	16,777,216	1.43
Eloquent	419	254	0	126	218	11,534,336	1.09
EloquentWithoutEvent	380	260	0	124	232	11,534,336	1.06

HHVM CLI 3.10.1 (Corresponding roughly to an up-to-date PHP 5.6)

Library	Insert	findPk	complex	hydrate	with	memory usage	time
PDO	54	30	0	37	92	783,680	0.23
LessQL	165	194	0	78	135	10,316,584	0.66
YiiM	333	194	0	81	494	7,267,424	1.33
YiiMWithCache	337	191	0	81	465	7,286,040	1.30
Yii2M	722	272	0	103	175	9,025,400	1.90
Yii2MArrayHydrate	702	273	0	100	165	9,033,272	1.83
Yii2MScalarHydrate	700	291	0	90	70	8,997,160	1.74
Propel20	545	169	0	771	512	9,740,696	2.30
Propel20WithCache	481	135	0	736	485	9,807,688	2.14
Propel20FormatOnDemand	479	128	0	675	463	9,822,696	2.05
DoctrineM	634	647	0	1009	342	19,639,560	5.11
DoctrineMWithCache	624	674	0	1047	339	19,635,128	5.07
DoctrineMArrayHydrate	629	642	0	804	249	18,447,232	4.73
DoctrineMScalarHydrate	618	634	0	747	203	17,366,336	4.55
DoctrineMWithoutProxies	620	629	0	849	357	19,402,032	4.85
Eloquent	589	273	0	98	203	14,652,488	1.53
EloquentWithoutEvent	547	264	0	99	206	14,572,112	1.45

PHP CLI 7.0.0 GA

Library	Insert	findPk	complex	hydrate	with	memory usage	time
PDO	38	36	0	33	86	1,720,432	0.20
LessQL	80	80	0	64	80	10,109,208	0.32
YiiM	128	76	0	53	147	6,291,456	0.45
YiiMWithCache	109	73	0	50	150	6,291,456	0.42
Yii2M	246	125	0	75	113	8,388,608	0.64
Yii2MArrayHydrate	233	128	0	63	104	8,388,608	0.60
Yii2MScalarHydrate	235	119	0	50	50	8,388,608	0.51
Propel20	142	74	0	163	190	6,291,456	0.63
Propel20WithCache	86	41	0	124	141	6,291,456	0.44
Propel20FormatOnDemand	85	43	0	118	138	6,291,456	0.44
DoctrineM	120	150	0	189	122	16,777,216	0.97
DoctrineMWithCache	118	146	0	194	112	16,777,216	0.96
DoctrineMArrayHydrate	121	144	0	146	85	16,777,216	0.90
DoctrineMScalarHydrate	118	235	0	183	91	16,777,216	1.01
DoctrineMWithoutProxies	120	151	0	170	190	16,777,216	1.06
Eloquent	209	137	0	84	137	8,388,608	0.65
EloquentWithoutEvent	221	148	0	85	144	8,388,608	0.66

[staabm]

Propel will for sure be perf optimzed but first the featureset itself needs to be completed.

[marcj]

Does anyone know if YII and Eloquent do lazy loading here? https://github.com/kenjis/php-orm-benchmark/blob/master/fuel/app/classes/controller/orm/yii2.php#L37 and https://github.com/kenjis/php-orm-benchmark/blob/master/fuel/app/classes/controller/orm/eloquent.php#L36.

[Big-Shark]

@marcj yes, both

[marcj]

Real lazy loading or are they eager loaded relations through configuration?

[Big-Shark]

@marcj It's really lazy loading

[marcj]

Interesting, then we should look what the hell we're doing currently wrong. :) I guess some caches aren't hit or we load too much eager.

[Big-Shark]

@maikg I think we should not pay attention to this test, it shows the speed initialization propel, and performed only 2 queries

[motin]

@Big-Shark @marcj Agree that is more important to pay attention to the second benchmark, which performs the following:

1. Mass insertion: Tests model objects and save() operations.
2. Retrieve By Pk: Tests basic hydration
3. Complex Query an OR but no hydration: Tests Query parsing
4. Basic Query with 5 results: Tests hydration and collections
5. Query with join: Tests join hydration

It gives a better albeit far from perfect idea of how Propel2 performs in real world applications.

[dereuromark]

Is this still relevant, is there anythign that can be done here as action points or can it be closed?