Package mondrian.test

Class PerformanceTest

java.lang.Object

TestCase

mondrian.test.FoodMartTestCase

mondrian.test.PerformanceTest

public class PerformanceTest
extends FoodMartTestCase

Various unit tests concerned with performance.

Since:

August 7, 2006

Author:

jhyde

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	PerformanceTest.CounterUdf	User-defined function that counts how many times it has been invoked.
(package private) static class	PerformanceTest.Statistician	Collects statistics for a test that is run multiple times.

Nested classes/interfaces inherited from class mondrian.test.FoodMartTestCase

FoodMartTestCase.QueryAndResult

Field Summary

Fields

Modifier and Type	Field	Description
static org.apache.log4j.Logger	LOGGER	Certain tests are enabled only if logging is enabled at debug level or higher.

Fields inherited from class mondrian.test.FoodMartTestCase

propSaver

Constructor Summary

Constructors

Constructor	Description
PerformanceTest(String name)

Method Summary

Modifier and Type	Method	Description
void	testBigResultsWithBigSchemaPerforms()	Tests performance of a larger schema with a large number of result cells.
void	testBugMondrian1242()	Test for MONDRIAN-1242, "Slicer size is exponentially inflating the cell requests".
void	testBugMondrian550()	Test case for Bug MONDRIAN-550, "Performance bug with NON EMPTY and large axes".
void	testBugMondrian550Tuple()	As testBugMondrian550() but with tuples on the rows axis.
void	testBugMondrian639()	Test case for Bug MONDRIAN-639, "RolapNamedSetEvaluator anon classes implement Iterable, causing performance regression from 2.4 in FunUtil.count()".
void	testBugMondrian843()	Test case for Bug MONDRIAN-843, where Filter is inefficient.
void	testBugMondrian981()	Testcase for bug MONDRIAN-981, "Poor performance when >=2 hierarchies are access-controlled with rollupPolicy=partial".
void	testInMemoryCalc()	Runs a query that performs a lot of in-memory calculation.
void	testMondrianBug641()	Test case for Bug MONDRIAN-641, "Large NON EMPTY result performs poorly with ResultStyle.ITERABLE".
void	testStablePartialSort()	Tests performance of FunUtil.stablePartialSort(java.util.List<T>, java.util.Comparator<T>, int).
void	testVeryLargeExplicitSet()	Tests performance when an MDX query contains a very large explicit set.

Methods inherited from class mondrian.test.FoodMartTestCase

allMember, assertAxisReturns, assertAxisThrows, assertBooleanExprReturns, assertExprReturns, assertExprThrows, assertQueriesReturnSimilarResults, assertQueryReturns, assertQueryThrows, assertSize, cubeByName, execute, executeExpr, executeQuery, executeSingletonAxis, genderMembersIncludingAll, getConnection, getDimensionWithName, getTestContext, isDefaultNullMemberRepresentation, isGroupingSetsSupported, member, productMembersPotScrubbersPotsAndPans, storeMembersCAAndOR, storeMembersUsaAndCanada, tearDown, verifySameNativeAndNot, warehouseMembersCanadaMexicoUsa

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

LOGGER

public static final org.apache.log4j.Logger LOGGER

Certain tests are enabled only if logging is enabled at debug level or higher.

Constructor Detail

PerformanceTest

public PerformanceTest(String name)

Method Detail

testBugMondrian550

public void testBugMondrian550()

Test case for Bug MONDRIAN-550, "Performance bug with NON EMPTY and large axes".

testBugMondrian550Tuple

public void testBugMondrian550Tuple()

As testBugMondrian550() but with tuples on the rows axis.

testMondrianBug641

public void testMondrianBug641()

Test case for Bug MONDRIAN-641, "Large NON EMPTY result performs poorly with ResultStyle.ITERABLE". Runs in ~10 seconds with ResultStyle.LIST, 99+ seconds with ITERABLE (on DELL Latitude D630).

testVeryLargeExplicitSet

public void testVeryLargeExplicitSet()

Tests performance when an MDX query contains a very large explicit set.

testBugMondrian639

public void testBugMondrian639()

Test case for Bug MONDRIAN-639, "RolapNamedSetEvaluator anon classes implement Iterable, causing performance regression from 2.4 in FunUtil.count()".

testBigResultsWithBigSchemaPerforms

public void testBigResultsWithBigSchemaPerforms()

Tests performance of a larger schema with a large number of result cells. Runs in 186 seconds without nonAllPositions array in RolapEvaluator. Runs in 14 seconds when RolapEvaluator.getProperty uses getNonAllMembers. The performance boost gets more significant as the schema size grows.

testInMemoryCalc

public void testInMemoryCalc()

Runs a query that performs a lot of in-memory calculation.

Timings (branch / change / host / DBMS / jdk / timings (s) / mean):

• mondrian-3.2 13366 marmalade oracle jdk1.6 592 588 581 571 avg 583
• mondrian-3.2 13367 marmalade oracle jdk1.6 643 620 631 671 avg 641
• mondrian-3.2 13397 marmalade oracle jdk1.6 604 626
• mondrian-3.2 13467 marmalade oracle jdk1.6 610 574
• mondrian-3.2 13489 marmalade oracle jdk1.6 565 561 579 596 avg 575
• mondrian-3.2 13490 marmalade oracle jdk1.6 607 611 581 605 avg 601
• mondrian-3.2 xxxxx marmalade oracle jdk1.6 562 583 541 522 avg 552
• mondrian-3.2 14036 marmalade oracle jdk1.6 451 433
• mondrian 14036 marmalade oracle jdk1.6 598 552
• mondrian 14037 marmalade oracle jdk1.6 626 596
• mondrian 14052 marmalade oracle jdk1.6 454
• mondrian 14770 marmite mysql jdk1.7 > 30 minutes

testBugMondrian843

public void testBugMondrian843()

Test case for Bug MONDRIAN-843, where Filter is inefficient.

testBugMondrian981

public void testBugMondrian981()

Testcase for bug MONDRIAN-981, "Poor performance when >=2 hierarchies are access-controlled with rollupPolicy=partial".

testBugMondrian1242

public void testBugMondrian1242()

Test for MONDRIAN-1242, "Slicer size is exponentially inflating the cell requests". This case just checks correctness; a similar case in PerformanceTest checks performance.

testStablePartialSort

public void testStablePartialSort()

Tests performance of FunUtil.stablePartialSort(java.util.List<T>, java.util.Comparator<T>, int).

"Pedro's algorithm" was supplied as MONDRIAN-1288, "Optimize stable partial sort when dataset is huge and limit is small".

Parameters: N (number of elements in list), L (limit; number of elements to return)

Conclusions:

• Array sort is better when L is almost as large as N
• If L is small, Pedro's algorithm is best; but it grows with O(L^2) and is unusable for L > 10,000
• If L is less than N / 20, Julian's algorithm is best
• For L larger than N / 20, array sort is best
• Each algorithm improves number of comparisons: Julian's algorithm is the best, with N + L log L + comparisons