Devoxx 2014: Conference day 1 Notes (12/11/2014)

keynote

jboss

• containers
• microservices
• “Camel”

oracle

• 20 years java (1995)
• java 8 Brian Goetz
  • lamba’s
  • stream() / parallelStream() (List / Hashset api)
  • default methods
• next
  • —> jigsaw module system
    • small devices (compact profile)
    • security mechs
    • startup performance
    • blocked usage of sun*internal stuff
    • part of openjdk
  • Poject Valhalla
    • cache misses / align curren tmemory
    • sepcialized generics
    • value types
    • var handles
  • Project Panama
• IoT Jasper Potters

---

Devoxx 2014: Pico Services with Java EE 7 on Java 8 and Docker

Adam Bien

base concepts:

• JAX-RS
• @Inject
• Business component: Boundary Control Entity pattern

setup demo

• netbeans
• glassfish v4
• simple pom.xml: javax / javaee-api 7.0
• JAX-RS
• @Stateless EJB (bean.xml in JEE6)
• docker:
  • test environment
  • expose ports
  • jenkins in docker
  • “ambassador pattern” -> used for real data

---

60 useful Linux commands in 15 minutes

Pierre-Antoine Grégoire

---

Sparky guide to bug-free JavaScript

Mite Mitreski

• debugger; statement: force debugger
• console.table clean format in console
• console.trace + Objects.observe who changed object?
• async debug (full stactrace) async
• DOM mutation Observer
• log with google analytics gaq.push() :-/

---

Let’s Get Physical: I/O Programming with Java on the Raspberry Pi using Pi4J

Robert Savage

• Pi4j:
  • open source / Low level IO
  • event based
  • java + JNI (C)
• IO interfaces
  • digital interfaceGPIO/ PWM
  • data UART, serial, SPI, I²C
  • analog interfaces: extra hardware
• models
  • a/b: 21 GPIO
  • b+ 28 GPIO
  • compute module: 46

---

Spotify - audio delivery at scale

Niklas Gustavsson

• commoditiy servers
  • high availibility, robustnest
  • colocated
• backend services
  • ‘atomic’
  • self-executing jar
  • wired toghether with propriatary protocol (zeromq-based)
• clients: keep socket open
• Service Discovery: DNS records
• robustnest / scalability:
  • storage: Cassandra
• teams
  • autonomuous org.
  • general principles / rules (programming language etc.)
• content distribution
  • attempt to do local caching (depends on device)
  • “storage resolve” service -> file location
  • CDN’s
    • audio-files are not really ‘standard’ content for CDN’s
    • multiple providers
    • performance is correlated with amount of data served from CDN
• measure!
  • latency
  • buffering
• others:
  • no p2p
  • only desktop clients (battery / bandwidth constraints)
  • prefetching
  • head files: first few seconds (unencrypted on CDN)

---

Concurrency in Enterprise Java

Alexander Heusingfeld

Java SE 5 classic Thread -> java.util.concurrent

• ExecutorServices
• no trx’s
• etc.

patterns of software stability

• timeouts
• bulkheads: separate app components
• curcuit breaker: failfast on thread starvation
• steady state: free unused resources

JSR 236: javax.enterprise.concurrent

• containermanaged threads
• ManagedThreadFactory
  • extends ThreadFactory from SE
  • ManageableThread
  • use ExecutorServices from SE
  • fork/join pool not supported
• ManagedExecutorService
  • jndi lookup
  • lifecycle api’s disabled
  • List >executor.invokeAll()
• ManagedScheduledExecutorService
  • schedule tasks with trigger (e.g. executor.scheduleAtFixedRate())
• ContextService
• ManagedTaskListener -> event management
  • ManagedExecutors class

runtime remarks:

• wildfly (JBoss): special commanline options
• some gotcha’s as this is new technology
  • check RejectionPolicy
  • Java 8 Streams API (parallelStream -> uses ForkJoinPool internally, not compatible with ManagedThreads)
• running on < Java8: Guava — ListeningExecutorService -> Callbacks
• Netflix Hystrix: stability patterns (e.g. Circuit-Breaker)
  • custom threadpool
  • hystrix.plugin.HystrixConcurrencyStrategy

Summary JSR 236:

• good mix of convenience /stability
• potential for optimisations
  • incompat. with ForkJoinPool
  • Runtime.availableProcessors()
• see https://github.com/jee-samples/javaee7-samples

---

Predictability and Performance

Richard Warburton

“be predictable”

Branch prediction:

• cpu-cycles: fetch, decode, exec, writeback
• branch -> stall -> predict
  • static: no dynamic data considered
    • forward: not taken
    • backware: taken
  • local / global
  • performance event counters:
    • specific registers -> can store prediction info
• demo time perf java ...

Locality

• arrays -> strides
• good data layout principles
  • primitive collections: HPPC, FastUtil, GNU Trove)
  • arrays instead of linked lists
  • hashtable instead of tree
  • custom data-structures: judy arrays, kD-Trees, Z-Order Curves
• java heap layout issues: no location guarantee (GC etc)
  • off heap structures (libraries)

Storage

• seeking vs sequential reads
• fragmentation: preallocate
• sector misalignment: offset logical vs physical sector
• iotop

---

The end of traditional enterprise IT

Ron Van Kemenade (CIO ING group)