Performance Improvements in 4.8

Last modified 02 May 2023 20:14 +02:00

Since 4.8

This functionality is available since version 4.8.

After optimizing the model-level post-processing of retrieved objects, we tried to measure the overall performance of midPoint. The semi-real performance test, developed as part of midScale project, was used.

Preliminary Results

The preliminary (indicative) results were obtained from the execution of the test on a dedicated iron, an older development PC. They may not be precise, as the number of objects involved was quite low (up to thousands), in order to gather the data quickly.

More detailed tests are being prepared. But let us present those preliminary results anyway.

Figure 1. Preliminary performance results

The chart compares average processing time for an account (during "initial import" scenarios) and then for a user (during "recomputation" scenarios). Individual scenarios differ in the size and complexity of the simulated deployment:

Table 1. Characterization of scenarios
Scenario	Source systems	Source mappings	Target systems	Target mappings per system	Assignments per user	Number of accounts/users
`1s-1m-0t-0m-0a-10ku`	1	1	0	-	0	10,000
`1s-50m-0t-0m-0a-5ku`	1	50	0	-	0	5,000
`1s-50m-5t-50m-0a-2500u`	1	50	5	50	0	2,500
`1s-50m-5t-50m-20a-1ku`	1	50	5	50	20	1,000

Table 2. MidPoint versions used
Version	Source code	Comment
`support-4.4`	427a0cdf	Current on Apr 27, 2023
`support-4.6`	d4139f25	Current on Apr 27, 2023
`support-4.7`	d71980b6	Current on Apr 27, 2023
`master`	midPoint fba1e2c9, prism fe9b732a	Current on Apr 27, 2023
`prism-fix-1`	midPoint the same, prism 531be184	Improved `PrismSchemaImpl.findTypeDefinitionByCompileTimeClass`.
`prism-fix-2`	midPoint the same, prism 1d5f8911	Improved `ItemImpl.equals`.
`prism-fix-3`	midPoint the same, prism e84729be	Removed two iterative definition lookups.
`prism-fix-4`	midPoint 23e237aa, prism 382f87be	Not freezing already frozen objects. Skipping checking equivalent prism values during parsing from repo.
`cloning-1`	midPoint f8e8d6b1	Reduced the number of object clone operations. The effect on the performance seems to be lower than expected.
`cached-marks`	midPoint 0f2062df	Enabled global caching of `MarkType` objects. The system performance test uses a custom caching configuration that didn’t contain the caching of marks (introduced in 4.7). However, it seems that the difference between caching disabled and enabled is quite small.
`qualified-names`	midPoint ed6351ad, prism 9c684889	The presence of unqualified names in `path` expression evaluator leads to sequential scans through prism container values. Before trying to fix this seriously in midPoint, this commit tries to measure the effect of using fully qualified names. (In addition to a minor prism improvement that should slightly improve the performance in this case.)
`misc-1`	midPoint e40f8794, prism ed32e894	Improved XMLGregorianCalendar cloning, avoided slow `PrismContainerValue.getProperties()` calls, partially improved definitions lookups and others.

Details

Testing scripts

Listing 1: Example of testing script for 4.7

cd /home/pavol/mp47
export DB="-Dtest.config.file=test-config-new-repo.xml -Dmidpoint.repository.jdbcUrl=jdbc:postgresql://localhost:5432/midpoint-test-47 -Dmidpoint.repository.jdbcUsername=midpoint -Dmidpoint.repository.jdbcPassword=..."

/home/pavol/perf/tests.sh

Listing 2: Common testing script

#!/bin/bash
export GIT_COMMIT=$(git rev-parse HEAD)
export BRANCH=$(git branch --show-current)

PREFIX="integration-test -pl :story -nsu -Pextratest -DenableAssertions=false -Dit.test=TestSystemPerformance $DB -Dsources.multi-attr-values=5 -DtaskTimeout=7200000 $EXTRA"

SOURCES_1_50="-Dsources.resources=1 -Dsources.single-mappings=50 -Dsources.multi-mappings=50"
SOURCES_3_50="-Dsources.resources=3 -Dsources.single-mappings=50 -Dsources.multi-mappings=50"

TARGETS_3_50="-Dtargets.resources=3 -Dtargets.single-mappings=50 -Dtargets.multi-mappings=50"
TARGETS_5_50="-Dtargets.resources=5 -Dtargets.single-mappings=50 -Dtargets.multi-mappings=50"

TARGETS_0="-Dtargets.resources=0"
ROLES_0="-Droles.business.count=0 -Droles.technical.count=0 -Droles.assignments.count=0 -Droles.inducements.count=0"

SCHEMA_10="-Dschema.single-valued-properties=10 -Dschema.multi-valued-properties=10"
SCHEMA_50="-Dschema.single-valued-properties=50 -Dschema.multi-valued-properties=50"
SCHEMA_100="-Dschema.single-valued-properties=100 -Dschema.multi-valued-properties=100"
SCHEMA_200="-Dschema.single-valued-properties=200 -Dschema.multi-valued-properties=200"

# Assuming 8x vCPU -> 6 threads
THREADS="-Dimport.threads=6 -Dimport.no-op-runs=0 -Dreconciliation.runs=0 -Drecomputation.threads=6"

# Scale (build variable)
S=5

#
# Group 1: Single source, increasing number of mappings: 1, 10, 50, 100, 200 (both single and multi)
#

PREFIX_G1="$PREFIX -Dsources.resources=1 $TARGETS_0 $ROLES_0 $THREADS"

mvn $CLEAN $PREFIX_G1 -Dsources.single-mappings=1   -Dsources.multi-mappings=1   $SCHEMA_10  -Dsources.accounts=$((2000*S))
#mvn        $PREFIX_G1 -Dsources.single-mappings=10  -Dsources.multi-mappings=10  $SCHEMA_10  -Dsources.accounts=$((2000*S))
mvn        $PREFIX_G1 -Dsources.single-mappings=50  -Dsources.multi-mappings=50  $SCHEMA_50  -Dsources.accounts=$((1000*S))
#mvn        $PREFIX_G1 -Dsources.single-mappings=100 -Dsources.multi-mappings=100 $SCHEMA_100 -Dsources.accounts=$((1000*S))
#mvn        $PREFIX_G1 -Dsources.single-mappings=200 -Dsources.multi-mappings=200 $SCHEMA_200 -Dsources.accounts=$((1000*S))

#
# Group 2: Increasing number of sources (having 50 + 50 mappings): 5, 10, 20
#

PREFIX_G2="$PREFIX -Dsources.single-mappings=50 -Dsources.multi-mappings=50 $SCHEMA_50 $TARGETS_0 $ROLES_0 $THREADS"

#mvn $PREFIX_G2 -Dsources.resources=3  -Dsources.accounts=$((500*S))
#mvn $PREFIX_G2 -Dsources.resources=5  -Dsources.accounts=$((500*S))
#mvn $PREFIX_G2 -Dsources.resources=10 -Dsources.accounts=$((500*S))

#
# Group 3: Increasing number of targets (having 1 source, 50 + 50 mappings): 5, 10, 20
#

PREFIX_G3="$PREFIX $SOURCES_1_50 $SCHEMA_50 $ROLES_0 $THREADS -Dtargets.single-mappings=50 -Dtargets.multi-mappings=50"

mvn $PREFIX_G3 -Dtargets.resources=5  -Dsources.accounts=$((500*S))
#mvn $PREFIX_G3 -Dtargets.resources=10 -Dsources.accounts=$((500*S))

#
# Group 4: Increasing number of assignments (having 1 source, 5 targets, 50 + 50 mappings; each BR has 2 TRs): 5, 10, 20, 100
#

PREFIX_G4="$PREFIX $SOURCES_1_50 $TARGETS_5_50 $SCHEMA_50 $THREADS -Droles.business.count=100 -Droles.technical.count=500 -Droles.inducements.count=2"

#mvn $PREFIX_G4 -Droles.assignments.count=5   -Dsources.accounts=$((250*S))
#mvn $PREFIX_G4 -Droles.assignments.count=10  -Dsources.accounts=$((200*S))
mvn $PREFIX_G4 -Droles.assignments.count=20  -Dsources.accounts=$((200*S))
#mvn $PREFIX_G4 -Droles.assignments.count=100 -Dsources.accounts=$((100*S))

#
# Group 5: Misc
#
PREFIX_G5="$PREFIX $SOURCES_3_50 $TARGETS_3_50 $SCHEMA_50 $THREADS -Droles.business.count=100 -Droles.technical.count=500 -Droles.inducements.count=2"

#mvn $PREFIX_G5 -Droles.assignments.count=50 -Dsources.accounts=$((150*S))

The -DenableAssertions=false turns off the evaluation of Java assertion statements.

Was this page helpful?

YES NO

Thanks for your feedback