Introduction to Java performance tuning

Introduction to Java Performance Tuning

Marouane Gazanayi@mgazanayi

API Dictionary

Prescriptions safety checkingPMSIRecos...

hoptimal/eVidalvidal.fr

MobileAnd much more...

WORA!

Write Once, Run Anywhere

Main concepts

javac

JVM

Main concepts

javac

scalac JVM

Main concepts

javac

JVMscalac

clojure compiler

And...

No more manual memory management

Signed: GC

SERIOUSLY

Awesome!

Generational hypothesis

Objects age

nb o

f ob

ject

s

Garbage Collector

Manages the Heap by removing "dead" objects

Garbage Collector

Garbage Collector

It is also responsible for "defragmenting" the Heap

Garbage Collector

Garbage Collector

Most employ a "stop the world" collection, meaning the running application must stop processing while the GC is engaged.

Garbage Collector

Therefore frequent or large amounts of collection affect the performance of your application.

Garbage CollectorLot of JVM options, at least always run with these

Heap

S0 S1eden

Young Generation

Survivor space

Heap

TenuredS0 S1eden

Old GenerationYoung Generation

Survivor space

Heap

Tenured Permanent

Permanent Generation

S0 S1eden

Old GenerationYoung Generation

Survivor space

Heap

Tenured Permanent

Metaspace

S0 S1eden

Old GenerationYoung Generation

Survivor space

Heap: Example

TenuredS0 S1eden

Heap: Example

11

TenuredS0 S1eden

Heap: Example

2

TenuredS0 S1eden

Heap: Example

31

TenuredS0 S1eden

Heap: Example

42

TenuredS0 S1eden

Heap: Example

3

TenuredS0 S1eden

Garbage Collector basics

SweepCompact

Copy

Garbage Collector sweep

Garbage Collector compact

Garbage Collector copy

Garbage Collector(s)

Serial GCParallel CollectorParallec CompactionConcurrent Mark Sweep CollectorG1

Garbage Collector(s)

Azul ZingShenandoah

...

HotSpot & JIT

Java source code

javac

.class file

Code cache

RuntimeHotSpot

Profiler JIT

Method cacheclass loader

java

What is performance?

The efficiency with which something reacts or

fulfills its intended purpose

What is performance?

Perceived Performance Issues

Some issues can be fixed by simply providing more information to the user

Response Time

How Long does it take for a system to react to a user's request

Throughput

How much data can be processed by the system within a given amount of time?

Load

Amount of work assigned at a given time

Utilization

Measure of how much of a system’s capacity is being consumed at a givenwork loads

Availability

Amount of time that an application is available to the end user

Technique

Estimate theoretical values and compare to measured values

Response timePoo

rAc

cept

able

Time

Load

Response timePoo

rAc

cept

able

Time

Load

ThroughputPoo

rAc

cept

able

Throughput

Load

ThroughputPoo

rAc

cept

able

Throughput

Load

ThroughputThroughput

Load

ThroughputThroughput

Load

CPU

100% CPU

CPU

100% CPU

Performance tips

A tip is a workaround for a

known problem

-Xmx == -Xms turns off adaptive sizing

Components that affect Performance

CPU

● Other processes running that are monopolizing CPU time● Single vs Parallel Processing

Input/Output

● File System ● Database ● Network Data Stores

The performance Tuning Lifecycle

1 Define Requirements

2 Measure / Test (create baseline)

3 Identify Bottleneck

4 Implements Fixes

5 Measure / Test (New Baseline)

6 Compare to Old Baseline & Requirements

7 If necessary repeat steps 3-6

The performance Tuning workflowsys cpu > 10%

user cpu

idle > 0% Memory efficient? GC

logs

system profiling: netstat, mpstat, iostat sar strace, gc logs, strace, etc...

App/CPU profilingMemory profiling, size frequency, life span,...

GC tuning, pool sizes, collectors...

Thread starvationThread dump

Application

NoneJVM

SystemYes

No

NoNo

Yes Yes

That’s all folks!