PTF - Performance Instance Resources Optimization - QCP1

Overview

  • The primary objective of testing was to evaluate the performance of the Baseline MCPT Environment configuration while attempting to optimize costs by adjusting instance types and reducing the number of instances. The tests were designed to compare the performance outcomes across different configurations, including variations in instance types and counts within multiple Auto Scaling Groups (ASGs). By systematically modifying these variables, the goal was to maintain or improve the performance observed in the baseline configuration while achieving cost efficiency.

PERF-962 - Getting issue details... STATUS  

Summary

  • Through a series of experiments involving different placement strategies, instance types, and total instance counts, we found that the performance remained consistent when using these configurations: 
    • three c7g.large instances dedicated to the okapi service alongside five r7g.2xlarge instances for all other services, with the CPU parameter set to 2 for all services.
    • five r7g.2xlarge instances for all services, with the CPU parameter set to 2 for all services.
  • Optimized environment configurations offers a 20-40% cost reduction compared to the existing setup, making it a more economical option without compromising on performance.
  • Configurations with three c7g.large instances for the okapi service and five r7g.2xlarge instances for all other services show the best performance across all experiments.
  • In fact, some workflows show better performance with this new setup than correct infrastructures.
  • The CPU utilization on EC2 level better now - around 30-60%, previously it was under 20%.

AWS Configuration Costs

ClusterInstance TypeCost per Month
(USD)
Number of InstancesTotal Cost per Cluster
(USD)
QCP1 m6g.2xlarge$221.7610$2,217.60
MCPT m6g.2xlarge$221.7614$3,104.64
Optimized Infrastructure
Two Auto Scaling Groups
c7g.large$52.203$1,698.84
r7g.2xlarge$308.455
Optimized Infrastructure
One Auto Scaling Groups
r7g.2xlarge$308.455$1,542.25


Cost Comparison (Before vs After)

Cluster

Previous Total Cost
(USD)

New Total Cost
(USD)

Percentage Saving
(%)

QCP1$2,217.60$1,698.8423.39%
MCPT$3,104.64$1,698.8445.28%

Test Runs

Test #DescriptionStatus
Test 1Instance type: m6g.2xlargeInstances count: 10Completed
Test 2Instance type: m6g.2xlargeInstances count: 10 (Repeat Test 1)Completed
Test 3Used two autoscaling groups, 1st with 3 Instance Type: c7g.large for okapi service and 5 Instance Type: m6g.2xlarge for others services.Completed
Test 4Used two autoscaling groups, 1st with 3 Instance Type: c7g.large for okapi service and 5 Instance Type: m6g.2xlarge for others services (Repeat Test 3).Completed
Test 5CPU=2 was set for all modules, used two autoscaling groups, 1st with 3 Instance Type: c7g.large, 3 of them for okapi service and 5 Instance Type: r7g.xlarge for others modules.Completed
Test 6CPU=2 was set for all modules, used two autoscaling groups, 1st with 3 Instance Type: c7g.large, 3 of them for okapi service and 5 Instance Type: r7g.xlarge for others modules (Repeat Test 5).Completed
Test 7CPU=2 was set for all modules except CPU=2048 for mod-search, used two autoscaling groups, 1st with 3 Instance Type: c7g.large, 3 of them for okapi service and 5 Instance Type: r7g.xlarge for others modules.Completed
Test 8

CPU=2 was set for all modules, used ONE autoscaling group with 5 Instance Type:  c7g.large for all services.


Test 9CPU=2 was set for all modules, used ONE autoscaling group with 5 Instance Type:  c7g.large for all services (Repeat Test 8).
Test 10CPU=2 was set for all modules except CPU=2048 for mod-search, used ONE autoscaling group with 5 Instance Type:  c7g.large for all services.
Test 11

CPU=2 was set for all modules except CPU=2048 for mod-search, used ONE autoscaling group with 5 Instance Type:  c7g.large for all services (Repeat Test 10).


Test 12CPU=2 was set for all modules except CPU=2048 for mod-search, used ONE autoscaling group with 5 Instance Type:  c7g.large for all services (Repeat Test 11).

Test Results

This table contains durations for all Workflows. 

WorkflowsTest 1

Test 2
(Repeat 1)

Test 3Test 4
(Repeat 3)
Test 5 Test 6
(Repeat 5)
Test 7Test 8Test 9
(Repeat 8)
Test 10Test 11
(Repeat 10)
Test 12
(Repeat 11)

Average response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
ErrorsAverage response time
(milliseconds)
Errors
DATA IMPORT0:52:03
0:44:55
0:46:07
0:47:09
0:51:41
0:58:35
1:00:03
0:43:53
0:47:06
0:55:30
0:45:46
0:44:09
DATA EXPORT0:58:11
0:44:43
0:47:41
0:50:32
0:38:59
0:45:53
0:48:26not finished for main0:45:41
0:48:49
0:56:49
0:42:16
0:44:26
CICO_TC_Check-In Controller11630%9480%9320%9580%8490%8950%9400%9120%9930%11760%8920%9670%
CICO_TC_Check-Out Controller16970%14810%14080%14280%13180%13180%13670%13450%14450%16750%13710%14670%
DE_Exporting MARC Bib records workflow25280%38180%36750%28300%19180%22230%18650%33630%24200%18720%33980%50330%
ILR_TC: Create ILR10230%8740%7300%8040%6240%6620%8020%8200%7670%11160%8770%7800%
ILR_TC: Get ItemId1320%1310%1070%1140%1120%1090%1090%1160%1270%1650%1360%1300%
MSF_TC: mod search by auth query48300%48356%54806%765810%757019%21750%24740%609410%411912%29380%554611%707518%
MSF_TC: mod search by boolean query4690%6213%14562%10275%18765%2561%6050%13435%5274%3330%11977%12636%
MSF_TC: mod search by contributors4750%16553%18056%17143%34678%4960%5030%19664%7556%4650%20905%20368%
MSF_TC: mod search by filter query2290%8442%7032%11370%15732%2170%2620%11002%4293%2380%9444%10686%
MSF_TC: mod search by keyword query2280%5193%11614%12458%13124%1980%2560%11443%4342%2500%9794%10188%
MSF_TC: mod search by subject query5770%18463%18253%16347%23282%4450%5390%21633%8407%4730%14766%19198%
MSF_TC: mod search by title query21410%32513%37701%45982%48878%20380%22940%42855%32146%20540%41457%44379%
DI_TC: Importing MARC records workflow Transaction &{tenant}10594510%204230%9356260%9504030%10478360%179460%121042933%172580%9553490%22382080%189740%187620%
PRV_TC: View Patron record Group3470%3100%2480%2380%2130%2840%3130%2650%2540%3550%2420%2730%
ULR_TC: Users loan Renewal Transaction10594510%18520%15280%17170%15510%15950%17490%15260%17610%41170%14960%16170%


Comparison

This graph shows the durations of all workflows compared between the best test results.

Part 1

 

Part 2

Part 3



Test №1

Introduction: Test 1: The Baseline QCP1 Environment configuration was applied, qcp1_MiniMaster.jmx script was run

Objective: The objective of test was to evaluate the performance of the MCPT environment by applying the baseline configuration.

Instance CPU Utilization

Service CPU Utilization

Here we can see that mod-inventory-b modules used 73% CPU.

Service Memory Utilization

Here we can't see any sign of memory leaks on every module. Memory shows stable trend.


Kafka metrics


OpenSearch metrics


DB CPU Utilization

DB CPU was 97% average.

DB Connections

Max number of DB connections was 1250 in maximum.

DB load

                                                                                                                    

Top SQL-queries

Test №2

Introduction: The Baseline QCP1 Environment configuration was applied, qcp1_MiniMaster.jmx script was run (Repeat Test 1).

Objective: The objective of this test was to validate the consistency of performance observed in Test 1 by repeating the same configuration.

Results: Results was almost the same for all workflows.

Instance CPU Utilization

Service CPU Utilization

Here we can see that mod-inventory-b modules used 102% CPU in maximum.

Service Memory Utilization

Here we can't see any sign of memory leaks on every module. Memory shows stable trend.


Kafka metrics


OpenSearch metrics

DB CPU Utilization

DB CPU was 95% average 

DB Connections

Max number of DB connections was 1250.

DB load

                                                                                                                    

Top SQL-queries

Test №3-4

Introduction: Test 3: The Baseline QCP1 Environment configuration was applied, used two autoscaling groups, 1st with 3 Instance Type: c7g.large for okapi service and 5 Instance Type:  c7g.large for others services, qcp1_MiniMaster.jmx script was run. Test 3: Repeat Test 1.

Objective: The objective of test was to evaluate the performance of the QCP1 environment by applying two distinct autoscaling groups with different instance types: c7g.large for the Okapi service and r7g.xlarge for all other services. By running test, the goal was to determine if this modified configuration could achieve similar or improved performance compared to the baseline, while potentially optimizing resource allocation and cost. 

Results: We observed nearly identical performance results almost for all the workflows compared to the Baseline configuration and repeated it in Test №4.

Instance CPU Utilization

Service CPU Utilization

Here we can see that mod-inventory-b used 78k% of the CPU.

Service Memory Utilization

Here we can't see any sign of memory leaks on every module. Memory shows stable trend.


Kafka metrics

OpenSearch metrics

DB CPU Utilization

DB CPU was 97% maximum.

DB Connections

Max number of DB connections was 2200.

DB load

                                                                                                                    

Top SQL-queries

Test №5-6-7

Introduction: The Baseline QCP1 Environment configuration was applied, and CPU=2 was set for all modules, used two autoscaling groups, 1st with 3 Instance Type: c7g.large for okapi service and 5 Instance Type:  c7g.large for others services, qcp1_MiniMaster.jmx script was run.

Objective: Repeat tests to confirm performance.

Results: Performance was confirmed at the same like baseline.

Instance CPU Utilization


Service CPU Utilization

Here we can see that mod-data-export-b used 82k% of the CPU power of parameter CPU=2.

Service Memory Utilization

Here we can't see any sign of memory leaks on every module. Memory shows stable trend.

Kafka metrics


OpenSearch metrics

DB CPU Utilization

DB CPU was 97%.

DB Connections

Max number of DB connections was 1250.

DB load

                                                                                                                    

Top SQL-queries

Test №8-9

Introduction:  The Baseline QCP1 Environment configuration was applied, and CPU=2 was set for all modules, used ONE autoscaling group with 5 Instance Type:  c7g.large for all services, qcp1_MiniMaster.jmx script was run.

Objective: The goal of test is to put okapi service in one tier with all services and compare performance with previous result and baseline result.

Results: Performance was confirmed at the same like baseline.

Instance CPU Utilization

Service CPU Utilization

Here we can see that mod-data-export-b used 36k% AVG of the CPU power of parameter CPU=2.