The ROI of Quarkus for Spring Boot Applications

Sanil Nambiar Session APD23 | June 2nd, 2026 Holly Cummins,
STSM, Quarkus The ROI of Quarkus for Spring Boot Applications

Agenda -What is Quarkus? -Why Quarkus? -Can we quantify that?
-The developer experience ROI -Getting to Quarkus from Spring Boot

@holly_cummins #Quarkus #IBM #RedHat 2022

@holly_cummins #Quarkus #IBM #RedHat “I can’t bring up the microservices
in my Java dev stack 2022

@holly_cummins #Quarkus #IBM #RedHat “I can’t bring up the microservices
in my Java dev stack … on my brand new Apple laptop with a M1 chip and - fi ntech CTO 2022

@holly_cummins #RedHat most java web frameworks are resource-hungry they hog
cpu and ram

@holly_cummins #RedHat most java web frameworks are verbose there is
a lot of typing

@holly_cummins #Quarkus #IBM #RedHat in production, it’s worse

@holly_cummins #Quarkus #IBM #RedHat in production, it’s worse example microservices
maths:

maths: Lufthansa maintenance scheduling system

maths: Lufthansa maintenance scheduling system single service: ½ core + 1 GB RAM

maths: Lufthansa maintenance scheduling system single service: ½ core + 1 GB RAM HA → 3x instances

maths: Lufthansa maintenance scheduling system single service: ½ core + 1 GB RAM HA → 3x instances ~100 microservices

maths: Lufthansa maintenance scheduling system single service: ½ core + 1 GB RAM HA → 3x instances ~100 microservices = 150 cores + 300 GB RAM

@holly_cummins #Quarkus #IBM #RedHat why is this happening? mismatch between
what we need and what the platform is optimised for

@holly_cummins #Quarkus #IBM #RedHat java was not designed for the
cloud

@holly_cummins #RedHat old Java frameworks were optimised for …

@holly_cummins #RedHat long-lived processes old Java frameworks were optimised for
…

@holly_cummins #RedHat long-lived processes annual (!) deployments old Java frameworks
were optimised for …

@holly_cummins #RedHat long-lived processes annual (!) deployments late-binding old Java
frameworks were optimised for …

@holly_cummins #RedHat long-lived processes annual (!) deployments late-binding re-con fi
gurable without restart old Java frameworks were optimised for …

enter … quarkus

a Java framework that gets you going faster, faster. enter
… quarkus

what is quarkus?

hibernate

hibernate rest

hibernate rest jackson

hibernate netty rest jackson

hibernate netty rest langchain4j jackson

hibernate netty rest langchain4j jackson etc

hibernate netty rest langchain4j jackson etc runtime infrastructure

hibernate netty rest langchain4j jackson etc runtime infrastructure build infrastructure

netty etc runtime infrastructure build infrastructure hibernate rest langchain4j jackson
- An integrated collection of your favourite libraries (Hibernate, Jackson, dependency injection, Netty, etc) - A super-scalable core, build on Eclipse Vert.x - Minimises unnecessary work at runtime

e ffi ciency

e ffi ciency build-time principle

e ffi ciency build-time principle reactive core

e ffi ciency developer experience (also e ffi ciency) build-time
principle reactive core

principle reactive core live reload

principle reactive core live reload continuous testing

principle reactive core minimising boilerplate live reload continuous testing

principle reactive core minimising boilerplate live reload continuous testing dev ui

why is quarkus novel?

@holly_cummins #RedHat application frameworks were optimised for dynamism

@holly_cummins #RedHat application frameworks optimised for dynamism dynamism has a
cost

@holly_cummins #RedHat paying a dynamism tax … even though the
app is not dynamic

@holly_cummins #RedHat a highly dynamic runtime in a container is
pointless

@holly_cummins #RedHat most apps are immutable now

@holly_cummins #Quarkus #IBM #RedHat how does a java framework start?

build time

build time runtime

packaging (maven, gradle…) build time runtime

build time runtime

</> build time runtime load and parse • config files • properties • yaml • xml • etc.

</> build time runtime

@ @ </> build time runtime • classpath scanning and annotation discovery • attempt to load class to enable/disable features

@ @ </> build time runtime

@ @ </> build time runtime build a metamodel of the world

@ @ </> build time runtime

@ @ </> build time runtime start • thread pools • I/O • etc.

@ @ </> build time runtime ready to do work!

@holly_cummins #Quarkus #IBM #RedHat what if we start the application
more than once? @ @ </>

more than once? @ @ </> @ @ </>

more than once? @ @ </> @ @ </> @ @ </>

more than once? @ @ </> @ @ </> @ @ </> @ @ </>

more than once? @ @ </> @ @ </> @ @ </> @ @ </> so much work gets redone every time

@holly_cummins Hibernate speed example: JTA auto-wiring

@holly_cummins Hibernate speed example: JTA auto-wiring Class.forName(“LikelyJTAImplementation”);

@holly_cummins Hibernate speed example: JTA auto-wiring Class.forName(“LikelyJTAImplementation”); Class.forName(“APossibleJTAImplementation”);

@holly_cummins Hibernate speed example: JTA auto-wiring Class.forName(“LikelyJTAImplementation”); Class.forName(“APossibleJTAImplementation”); Class.forName(“AnotherJTAImplementation”);

@holly_cummins Hibernate speed example: JTA auto-wiring Class.forName(“LikelyJTAImplementation”); Class.forName(“APossibleJTAImplementation”); Class.forName(“AnotherJTAImplementation”); …

@holly_cummins Hibernate speed example: JTA auto-wiring Class.forName(“LikelyJTAImplementation”); Class.forName(“APossibleJTAImplementation”); Class.forName(“AnotherJTAImplementation”); Class.forName(“NicheJTAImplementation”);
…

Class.forName(“VeryNicheJTAImplementation”); …

Class.forName(“VeryNicheJTAImplementation”); … ~129 auto-wiring attempts

Class.forName(“VeryNicheJTAImplementation”); … ~129 auto-wiring attempts every single start.

@holly_cummins it’s not just JTA this happens for lots of
internal service bindings

@holly_cummins JVM footprint example: Hibernate

@holly_cummins JVM spends time loading classes for specific databases JVM
class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database footprint example: Hibernate

class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database class for unused database turns out they’re never used footprint example: Hibernate

turns out they’re never used JIT spends time unloading classes footprint example: Hibernate

@holly_cummins Hibernate example: ~500 classes which are only useful if
you're running an Oracle database loaded and then unloaded

@holly_cummins Hibernate example: ~500 classes which are only useful if
you're running an Oracle database loaded and then unloaded every single start.

@holly_cummins the true cost of loaded classes isn’t just memory
+ start time

+ start time method dispatching:

@holly_cummins interface the true cost of loaded classes isn’t just
memory + start time method dispatching:

@holly_cummins unused implementation the one we want interface unused implementation
unused implementation the true cost of loaded classes isn’t just memory + start time method dispatching:

@holly_cummins unused implementation the one we want interface megamorphic call
slow dispatching unused implementation unused implementation the true cost of loaded classes isn’t just memory + start time method dispatching:

+ start time the one we want interface

+ start time the one we want monomorphic call fast dispatching interface

@holly_cummins how do we fix all this?

@holly_cummins #Quarkus #IBM #RedHat @ @ </> build time runtime
what if we initialize at build time?

start • thread pools • I/O • etc. what if we initialize at build time?

ready to do work! start • thread pools • I/O • etc. what if we initialize at build time?

@holly_cummins #Quarkus #IBM #RedHat @ @ </> repeated starts are
now efficient

@holly_cummins #Quarkus #IBM #RedHat @ @ </> repeated starts are
now efficient less wasted work

@holly_cummins #Quarkus #IBM #RedHat the Quarkus way enables native compilation
native (graalvm) @ @ </> jvm build time

to the code! @holly_cummins #RedHat native binaries with Quarkus

#Quarkus @holly_cummins 2020

#Quarkus @holly_cummins hey, wanna see quarkus? hey, wanna see quarkus?
hey, wanna see quarkus? hey, wanna see quarkus? hey, wanna see quarkus? 2020

hey, wanna see quarkus? hey, wanna see quarkus? hey, wanna see quarkus? uhh … are you supposed to shut down applications after using them? 2020

hey, wanna see quarkus? hey, wanna see quarkus? hey, wanna see quarkus? uhh … are you supposed to shut down applications after using them? 120 instances (!) 2020

@holly_cummins

@holly_cummins doing more up-front

@holly_cummins doing more up-front - speeds up start

@holly_cummins doing more up-front - speeds up start - shrinks
memory footprint

@holly_cummins doing more up-front - speeds up start - shrinks
memory footprint - improves throughput (!)

can we quantify that?

spoiler:

2 - 3x infra savings spoiler:

to the code! @holly_cummins #RedHat performance comparisons

https://github.com/quarkusio/spring-quarkus-perf-comparison

quarkus has great performance

quarkus starts fast

quarkus is lean with memory

quarkus has super throughput

how does all that impact costs?

low load pack many instances onto a node memory is
the limiting factor how much hardware does an app need? high load horizontally scale to support load max throughput is the limiting factor

machine (1GiB RAM)

machine (1GiB RAM) 300 MiB RAM

machine (1GiB RAM) 300 MiB RAM 300 MiB RAM

machine (1GiB RAM) 300 MiB RAM 300 MiB RAM 300
MiB RAM

MiB RAM 300 MiB RAM

MiB RAM

assumptions: EC2 c6i.xlarge instance, us-east-1, $0.17 per hour low load
scenario: cost per instance per year

quarkus spring p r i 0.17 p r i 1489.2
assumptions: EC2 c6i.xlarge instance, us-east-1, $0.17 per hour low load scenario: cost per instance per year

jvm US$64.75 US$135.38 assumptions: EC2 c6i.xlarge instance, us-east-1, $0.17 per hour low load scenario: cost per instance per year

jvm US$64.75 US$135.38 native US$19.86 US$64.75 assumptions: EC2 c6i.xlarge instance, us-east-1, $0.17 per hour low load scenario: cost per instance per year

jvm US$64.75 US$135.38 native US$19.86 US$64.75 assumptions: EC2 c6i.xlarge instance, us-east-1, $0.17 per hour low load scenario: cost per instance per year 2.1x

jvm US$64.75 US$135.38 native US$19.86 US$64.75 assumptions: EC2 c6i.xlarge instance, us-east-1, $0.17 per hour low load scenario: cost per instance per year 2.1x 3.3x

p r i p r i jvm native quarkus 0.17
1489.2 US$6,475 US$1,986 spring US$13,538 US$6,475 low load scenario: cost per hundred instances per year assumptions: EC2 c6i.xlarge instance, us-east-1, $0.17 per hour

10,000 tps

10,000 tps machine 15,000 tps

10,000 tps 20,000 tps machine 15,000 tps

10,000 tps 20,000 tps machine 15,000 tps machine 15,000 tps
machine 15,000 tps

10,000 tps 20,000 tps 30,000 tps machine 15,000 tps machine
15,000 tps machine 15,000 tps

10,000 tps 20,000 tps 30,000 tps machine 15,000 tps machine
15,000 tps machine 15,000 tps machine 15,000 tps machine 15,000 tps

10,000 tps 20,000 tps 30,000 tps 40,000 tps machine 15,000
tps machine 15,000 tps machine 15,000 tps machine 15,000 tps machine 15,000 tps

10,000 tps 20,000 tps 30,000 tps 40,000 tps machine 15,000
tps machine 15,000 tps machine 15,000 tps machine 15,000 tps machine 15,000 tps machine 15,000 tps machine 15,000 tps machine 15,000 tps

native 2.5x

jvm 1.7x native 2.5x

native $85k savings/year

jvm $16k savings/year native $85k savings/year

#Quarkus @holly_cummins we also see better response times

#Quarkus @holly_cummins users care (a lot) about response time we
also see better response times

can we trust this data?

what we see in the fi eld matches what we
see in the lab.

@holly_cummins #Quarkus #IBM #RedHat remember the airline scheduling application?

@holly_cummins #Quarkus #IBM #RedHat remember the airline scheduling application? “[With
Quarkus], we can run 3 times denser deployments without sacrificing availability and response times of services. ” – Thorsten Pohl

@holly_cummins #Quarkus #IBM #RedHat remember the airline scheduling application? “[With
Quarkus], we can run 3 times denser deployments without sacrificing availability and response times of services. ” – Thorsten Pohl 3x less hardware

https://www.reddit.com/r/SpringBoot/comments/1tj5tsn/comment/on04x23/

https://www.reddit.com/r/SpringBoot/comments/1tj5tsn/comment/on04x23/ 2.7x less memory

is quarkus greener?

@holly_cummins #Quarkus #IBM #RedHat

@holly_cummins #Quarkus #IBM #RedHat does being faster and lighter mean
quarkus is greener?

to the code! @holly_cummins #RedHat energy comparisons

@holly_cummins #RedHat Setup: • REST + CRUD • large heap
• RAPL energy measurement • multiple instances to support high load  Assumptions: • US energy mix Source: John O’Hara experiment 1: RAPL measurements

• RAPL energy measurement • multiple instances to support high load  Assumptions: • US energy mix Source: John O’Hara experiment 1: RAPL measurements quarkus on JVM has the lowest carbon … because it has the highest throughput

• RAPL energy measurement • multiple instances to support high load  Assumptions: • US energy mix Source: John O’Hara experiment 1: RAPL measurements quarkus on JVM has the lowest carbon … because it has the highest throughput ~½ the energy requirements

@holly_cummins #RedHat density Source: Clement Esco ff i er experiment
2: cloud Setup: • 800 requests/second, over 20 days • SLA > 99% • AWS instances Assumptions: • Costs are for us-east-1 data centre

@holly_cummins #RedHat interpolated carbon metrics – hardware and electricity Setup:
• 800 requests/second, over 20 days • SLA > 99% Assumptions: • 50% load Source: Clement Esco ffi er x Teads cloud carbon impact of framework choice

• 800 requests/second, over 20 days • SLA > 99% Assumptions: • 50% load Source: Clement Esco ffi er x Teads cloud carbon impact of framework choice the carbon is lower because the cost is lower

• 800 requests/second, over 20 days • SLA > 99% Assumptions: • 50% load Source: Clement Esco ffi er x Teads cloud carbon impact of framework choice the carbon is lower because the cost is lower ~½ the carbon

developer experience

developers cost money too

@holly_cummins #Quarkus #IBM #RedHat quarkus enables faster feedback

@holly_cummins #Quarkus #IBM #RedHat mvn quarkus:dev zero-con fi g live
coding

@holly_cummins #Quarkus #IBM #RedHat tests are run on every code
change “reverse code coverage” means only relevant tests run mvn quarkus:dev continuous testing

@holly_cummins #Quarkus #IBM #RedHat developer UI

to the code! @holly_cummins #RedHat dev mode and continuous testing

@holly_cummins #Quarkus #IBM #RedHat with quarkus, developers have to type
less boilerplate

@holly_cummins #Quarkus #IBM #RedHat with quarkus, developers have to read
less boilerplate

@holly_cummins #Quarkus #IBM #RedHat zero-con fi g testcontainers

@holly_cummins #Quarkus #IBM #RedHat testcontainers integration … without quarkus @TestConfiguration(proxyBeanMethods
= false) public class ContainersConfig { @Bean @ServiceConnection public PostgreSQLContainer<?> postgres() { return new PostgreSQLContainer<>(DockerImageName.parse("postgres:14")); } } public class TestApplication { public static void main(String[] args) { SpringApplication .from(MySpringDataApplication::main) .with(ContainersConfig.class) .run(args); } } @Import(ContainersConfig.class)

@holly_cummins #Quarkus #IBM #RedHat testcontainers integration … without quarkus

@holly_cummins #Quarkus #IBM #RedHat zero-config testcontainers integration the only thing
you need to do to make testcontainers work is not con fi gure anything # configure your datasource quarkus.datasource.db-kind = postgresql quarkus.datasource.username = sarah quarkus.datasource.password = connor quarkus.datasource.jdbc.url = jdbc:postgresql://localhost:5432/mydatabase # drop and create the database at startup quarkus.hibernate-orm.database.generation = drop-and-create

you need to do to make testcontainers work is not con fi gure anything # drop and create the database at startup quarkus.hibernate-orm.database.generation = drop-and-create

you need to do to make testcontainers work is not con fi gure anything

you need to do to make testcontainers work is not con fi gure anything quarkus also auto-invokes fl yway and liquibase

@holly_cummins #Quarkus #IBM #RedHat @ApplicationScoped public class GreetingRepository { public
Entity findByName(int name) { return find("name", name).firstResult(); } void persist(Entity entity) {} void delete(Entity entity) {} Entity findById(Id id) {} List<Entity> list(String query, Sort sort, Object... params) { return null; } Stream<Entity> stream(String query, Object... params) { return null; } long count() { return 0; } long count(String query, Object... params) { return 0; } } example: panache + hibernate

@holly_cummins #Quarkus #IBM #RedHat example: panache + hibernate @ApplicationScoped public
class GreetingRepository implements PanacheRepository<Greeting> { public Entity findByName(int name) { return find("name", name).firstResult(); } }

@holly_cummins #Quarkus #IBM #RedHat DAO example: panache + hibernate @ApplicationScoped
public class GreetingRepository implements PanacheRepository<Greeting> { public Entity findByName(int name) { return find("name", name).firstResult(); } } repository pattern

@holly_cummins #Quarkus #IBM #RedHat example: panache + hibernate

@holly_cummins #Quarkus #IBM #RedHat example: panache + hibernate active record
pattern @Entity public class Greeting extends PanacheEntity { public String name; public LocalDate issued; @Version public int version; public static List<Greeting> getTodaysGreetings() { return list("date", LocalDate.now()); } }

@holly_cummins #Quarkus #IBM #RedHat can we quantify that?

https://www.reddit.com/r/SpringBoot/comments/1tj5tsn/comment/on04x23/

https://www.reddit.com/r/SpringBoot/comments/1tj5tsn/comment/on04x23/ 40% less code

Christos Sotiriou, DXL technical lead at Vodafone Greece

2019: Moved digital experience to quarkus (from Spring Boot) Christos
Sotiriou, DXL technical lead at Vodafone Greece

2019: Moved digital experience to quarkus (from Spring Boot) •
Startup times 40s ➡ 12s Christos Sotiriou, DXL technical lead at Vodafone Greece

Startup times 40s ➡ 12s • RAM 800 MB ➡ 360 MB Christos Sotiriou, DXL technical lead at Vodafone Greece

Startup times 40s ➡ 12s • RAM 800 MB ➡ 360 MB • Developer productivity ⬆ 30-40%– Christos Sotiriou, DXL technical lead at Vodafone Greece

Startup times 40s ➡ 12s • RAM 800 MB ➡ 360 MB • Developer productivity ⬆ 30-40%– 30-40% more productive Christos Sotiriou, DXL technical lead at Vodafone Greece

Quarkus Insights #192, https://www.youtube.com/watch?v=YNxz3KaTz5Y

Quarkus Insights #192, https://www.youtube.com/watch?v=YNxz3KaTz5Y ~2x more productive

2-3x less dev costs Quarkus Insights #192, https://www.youtube.com/watch?v=YNxz3KaTz5Y

“Quarkus is an absolute joy”

#RedHat @holly_cummins job satisfaction is the no. 1 predictor of
organizational performance

migration (the i in roi)

@holly_cummins #Quarkus #IBM #RedHat

@holly_cummins #Quarkus #IBM #RedHat compatibility libraries

@holly_cummins #Quarkus #IBM #RedHat <dependency> <groupId>io.quarkus</groupId> <artifactId>quarkus-spring-web</artifactId> </dependency> <dependency> <groupId>io.quarkus</groupId>
<artifactId>quarkus-spring-data-jpa</artifactId> </dependency> compatibility libraries

@holly_cummins #Quarkus #IBM #RedHat compatibility libraries

@holly_cummins #Quarkus #IBM #RedHat migration compatibility libraries

@holly_cummins #Quarkus #IBM #RedHat migration compatibility libraries tooling

@holly_cummins #Quarkus #IBM #RedHat migration compatibility libraries tooling ai skills

to the code! @holly_cummins #RedHat spring compatibility libraries

@holly_cummins #Quarkus #IBM #RedHat “After a week of development with
Quarkus, I was able to regain the same level of productivity as when I was developing with Spring Boot.” – Fawaz Paraïso, Decathlon

the double-win

@holly_cummins #RedHat

@holly_cummins #RedHat co-benefits

@holly_cummins #RedHat co-benefits the double win

@holly_cummins #RedHat co-benefits the double win win-win

@holly_cummins #RedHat co-benefits the double win win-win twofer

@holly_cummins #RedHat co-benefits the double win win-win twofer überwinden

@holly_cummins #RedHat

@holly_cummins #RedHat cheaper at low load

@holly_cummins #RedHat cheaper at low load cheaper at high load

greener

greener more productive

greener more productive happier

@holly_cummins #Quarkus #IBM #RedHat tl;dpa (too long didn’t pay attention)
deployment density lower cloud bill frictionless development experience Medium Nano auto-provision services zero-con fi g live coding continuous testing developer UI greener happy, productive people

@holly_cummins #Quarkus #IBM #RedHat stl;dpa (still too long didn’t pay
attention) e ffi ciency quarkus can be packed ~2x more densely than spring boot quarkus can handle ~2x more tra ffi c on the same hardware developer experience quarkus creates joy quarkus boosts productivity by

attention) e ffi ciency quarkus can be packed ~2x more densely than spring boot quarkus can handle ~2x more tra ffi c on the same hardware developer experience quarkus creates joy quarkus boosts productivity by 2x less infra

attention) e ffi ciency quarkus can be packed ~2x more densely than spring boot quarkus can handle ~2x more tra ffi c on the same hardware developer experience quarkus creates joy quarkus boosts productivity by 40% more productive 2x less infra

The ROI of Quarkus for Spring Boot Applications

The ROI of Quarkus for Spring Boot Applications

More Decks by Holly Cummins

Other Decks in Programming

Featured

Transcript