Wir unternehmen IT. Efficient use of multi-core processors with lambdas and streams Dr. Fabian Stäber 

Slide 09.09.2014 www.consol.de 2 λ λ λ λ λ 

Slide Example 26 March 2014 www.consol.de 3 Business Data Aggregation 

Slide Example • Boilerplate – pom.xml – RestConfig (modern web.xml) • Model – Contract – State enum – TestDataGenerator • View – JavaScript, HTML • Controller – Dashboard. 26 March 2014 www.consol.de 4 

Slide • Boilerplate – pom.xml – RestConfig (modern web.xml) • Model – Contract – State enum – TestDataGenerator • View – JavaScript, HTML • Controller – Dashboard. 4.0.0 de.consol.research lambda 1.0-SNAPSHOT war 1.8 1.8 false javax javaee-api 7.0 lambda Example 26 March 2014 www.consol.de 5 javax javaee-web-api 7.0 

Slide package de.consol.research; import ... @ApplicationPath("api") public class RestConfig extends Application {} Example • Boilerplate – pom.xml – RestConfig (modern web.xml) • Model – Contract – State enum – TestDataGenerator • View – JavaScript, HTML • Controller – Dashboard. 26 March 2014 www.consol.de 6 @ApplicationPath("api") public class RestConfig extends Application {} 

Slide Example • Boilerplate – pom.xml – RestConfig (modern web.xml) • Model – Contract – State enum – TestDataGenerator • View – JavaScript, HTML • Controller – Dashboard. 26 March 2014 www.consol.de 7 package de.consol.research; import java.time.LocalDate; public class Contract { private final State state; private final int priceInCent; private final LocalDate date; public Contract(State state, int priceInCent, LocalDate date) { this.state = state; this.priceInCent = priceInCent; this.date = date; } public State getState() { return state; } public int getPriceInCent() { return priceInCent; } public LocalDate getDate() { return date; } } public class Contract { private final State state; private final int priceInCent; private final LocalDate date; 

Slide Example • Boilerplate – pom.xml – RestConfig (modern web.xml) • Model – Contract – State enum – TestDataGenerator • View – JavaScript, HTML • Controller – Dashboard. 26 March 2014 www.consol.de 8 package de.consol.research; public enum State { BW, // Baden-Württemberg BY, // Bayern BE, // Berlin BB, // Brandenburg HB, // Bremen HH, // Hamburg HE, // Hessen MV, // Mecklenburg-Vorpommern NI, // Niedersachsen NW, // Nordrhein-Westfalen RP, // Rheinland-Pfalz SL, // Saarland SN, // Sachsen ST, // Sachsen-Anhalt SH, // Schleswig-Holstein TH // Thüringen } public enum State { BW, // Baden-Württemberg BY, // Bayern BE, // Berlin BB, // Brandenburg 

Slide Example • Boilerplate – pom.xml – RestConfig (modern web.xml) • Model – Contract – State enum – TestDataGenerator • View – JavaScript, HTML • Controller – Dashboard. 26 March 2014 www.consol.de 9 package de.consol.research; import ... public class TestDataGenerator { public static List makeTestData() { return Arrays.asList( new Contract(BY, 2000, now()), new Contract(BY, 3500, now().minusDays(1)), new Contract(NW, 7000, now().minusMonths(1)), new Contract(HE, 400, now().minusMonths(1) .minusDays(2)) ); } } public class TestDataGenerator { public static List makeTestData() { return Arrays.asList( new Contract(BY, 2000, now()), new Contract(BY, 3500, now().minusDays(1)), new Contract(NW, 7000, now().minusMonths(1)), 

Slide Example • Boilerplate – pom.xml – RestConfig (modern web.xml) • Model – Contract – State enum – TestDataGenerator • View – JavaScript, HTML • Controller – Dashboard. 26 March 2014 www.consol.de 10 

Slide Example • Boilerplate – pom.xml – RestConfig (modern web.xml) • Model – Contract – State enum – TestDataGenerator • View – JavaScript, HTML • Controller – Dashboard. 26 March 2014 www.consol.de 11 package de.consol.research; import ... @Path("dashboard") public class Dashboard { private List contracts = TestDataGenerator.makeTestData(); } @Path("dashboard") public class Dashboard { private List contracts = TestDataGenerator.makeTestData(); } 

Slide Stream API 26 March 2014 www.consol.de 12 Business Data Aggregation with Stream API 

Slide GET /lambda/api/dashboard/total-sales Stream API 26 March 2014 www.consol.de 13 Task 1: Total Sales 12900 

Slide Stream API 26 March 2014 www.consol.de 14 @GET @Path("total-sales") @Produces(APPLICATION_JSON) public int totalSales() { return contracts .stream() .filter(contract -> contract.getDate().isAfter(now().minusYears(1))) .mapToInt(Contract::getPriceInCent) .sum(); } 

Slide Stream API 26 March 2014 www.consol.de 15 Task 2: Sales by State GET /lambda/api/dashboard/sales-by-state { "BY": 5500, "NW": 7000, "HE": 400 } 

Slide Stream API 26 March 2014 www.consol.de 16 @GET @Path("sales-by-state") @Produces(APPLICATION_JSON) public Map salesByState() { return contracts .stream() .filter(contract -> contract.getDate().isAfter(now().minusYears(1))) .collect(Collectors.groupingBy( Contract::getState, Collectors.summingInt(Contract::getPriceInCent))); } 

Slide Stream API 26 March 2014 www.consol.de 17 @GET @Path("sales-by-state") @Produces(APPLICATION_JSON) public Map salesByState() { return contracts .stream() .filter(contract -> contract.getDate().isAfter(now().minusYears(1))) .collect(Collectors.groupingBy( Contract::getState, Collectors.summingInt(Contract::getPriceInCent))); } Can we re-use Lambda expressions? 

Slide Stream API 26 March 2014 www.consol.de 18 private Predicate last12months = contract -> contract.getDate().isAfter(now().minusYears(1)); @GET @Path("sales-by-state") @Produces(APPLICATION_JSON) public Map salesByState() { return contracts .stream() .filter(last12months) .collect(Collectors.groupingBy( Contract::getState, Collectors.summingInt(Contract::getPriceInCent))); } 

Slide GET /lambda/api/dashboard/sales-per-month Stream API 26 March 2014 www.consol.de 19 Task 3: Sales per Month { “2014-03": 5500, “2014-02": 7400 } 

Slide Stream API 26 March 2014 www.consol.de 20 @GET @Path("sales-per-month") @Produces(APPLICATION_JSON) public Map salesPerMonth() { return contracts .stream() .filter(last12months) .collect(Collectors.groupingBy( c -> c.getDate().getYear() + "-" + c.getDate().getMonthValue(), Collectors.summingInt(Contract::getPriceInCent))); } 

Slide Stream API Stream API 26 March 2014 www.consol.de 21 Get things done in few lines of code… …really  ? 

Slide Stream API 26 March 2014 www.consol.de 22 public int totalSales() { return contracts .stream() .mapToInt(Contract::getPriceInCent) .sum(); } public int totalSales() { int resultInCent = 0; for (Contract contract : contracts) { resultInCent += contract.getPriceInCent(); } return resultInCent; } Java 8 Java 7 

Slide Stream API 26 March 2014 www.consol.de 23 Why all that hype? 

Slide Stream API 26 March 2014 www.consol.de 24 Faster !!! 

Slide How To Make Things Faster • Keep Data in RAM • Avoid DB Roundtrips • Snappy Lightweight AJAX Front-End • Use Multiple Cores 26 March 2014 www.consol.de 25    ? ? 

Slide 09.09.2014 www.consol.de 26 λ λ λ λ λ 

Slide Use Multiple Cores parallelize 26 March 2014 www.consol.de 27 

Slide Use Multiple Cores 26 March 2014 www.consol.de 28 public int totalSales() { return contracts .stream() .mapToInt(Contract::getPriceInCent) .sum(); } public int totalSales() { int resultInCent = 0; for (Contract contract : contracts) { resultInCent += contract.getPriceInCent(); } return resultInCent; } Java 8 Java 7 

Slide Use Multiple Cores 26 March 2014 www.consol.de 29 public int totalSales() { return contracts .stream() .mapToInt(Contract::getPriceInCent) .sum(); } public int totalSales() { int resultInCent = 0; for (Contract contract : contracts) { resultInCent += contract.getPriceInCent(); } return resultInCent; } Do It Yourself synchronized, volatile, Lock, Executor, Callable, … Let the Library Do It For You 

Slide Use Multiple Cores 26 March 2014 www.consol.de 30 public int totalSales() { return contracts .parallelStream() .mapToInt(Contract::getPriceInCent) .sum(); } public int totalSales() { int resultInCent = 0; for (Contract contract : contracts) { resultInCent += contract.getPriceInCent(); } return resultInCent; } Do It Yourself synchronized, volatile, Lock, Executor, Callable, … Let the Library Do It For You 

Slide Use Multiple Cores 26 March 2014 www.consol.de 31 // Set of all states in contract data Set states = new HashSet(); contracts .parallelStream() .map(Contract::getState) .forEach(s -> states.add(s)); Thread Safe Operations Not Thread Safe 

Slide Use Multiple Cores 26 March 2014 www.consol.de 32 // Set of all states in contract data Set states = contracts .parallelStream() .map(Contract::getState) .collect(Collectors.toSet()); Thread Safe Operations Thread Safe  

Slide Use Multiple Cores Under the Hood (internal stream API implementation) 26 March 2014 www.consol.de 33 

Slide Under the Hood 26 March 2014 www.consol.de 34 (2 + 8 + 7 + 3) + (3 + 5 + 1 + 7) (2 + 8) + (7 + 3) 7 + 3 2 + 8 3 + 5 1 + 7 (3 + 5) + (1 + 7) 

Slide Under the Hood 26 March 2014 www.consol.de 35 if (my portion of the work is small enough) do the work directly else split my work into two pieces invoke the two pieces and wait for the results Java 7: Fork/Join Framework 

Slide Under the Hood 26 March 2014 www.consol.de 36 public class LegacyContractsData implements Iterable { private State[] states; private int[] pricesInCent; private LocalDate[] dates; // ... } Can I use this with my legacy data? 

Slide Under the Hood Tell the stream API: • how to split your data into chunks • how to iterate over a chunk. 26 March 2014 www.consol.de 37 Implement a Spliterator. Can I use this with my legacy data? 

Slide Under the Hood 26 March 2014 www.consol.de 38 public class MyOwnSpliterator implements Spliterator { @Override public Spliterator trySplit() { // split data into chunks } @Override public boolean tryAdvance(Consumer action) { // iterate over a chunk } @Override public long estimateSize() { // optional metadata } @Override public int characteristics() { // optional metadata } } 

Slide Use Multiple Cores 26.03.2014 www.consol.de 39 Awesome! Amazing! But: How much faster is it really? 

Slide Use Multiple Cores Benchmarking 26 March 2014 www.consol.de 40 

Slide public static void main(String[] args) { long start = System.nanoTime(); for (int i=0; i<10; i++) { dashboard.totalSales(); // sequential version } long duration = System.nanoTime() - start; System.out.println("10 sequential calls took " + duration + "ns."); // ... same with parallel version here. } Use Multiple Cores 26.03.2014 www.consol.de 41 Naive and probably wrong. vm warmup dead code elimination constant folding 

Slide Use Multiple Cores 26.03.2014 www.consol.de 42 JMH to the Rescue! @Benchmark @BenchmarkMode(Mode.Throughput) @OutputTimeUnit(TimeUnit.MILLISECONDS) public int myBenchmark() { return dashboard.totalSales(); } 

Slide Use Multiple Cores 26.03.2014 www.consol.de 43 10 Contracts: Sequential 8 times better 10,000 Contracts: Parallel 2 times better 

Slide 09.09.2014 www.consol.de 44 λ λ λ λ λ λ 

Slide Summary 26 March 2014 www.consol.de 45 Efficient use of multi-core processors with lambdas and streams: • Library does it for you • Use thread safe operations • Implement Spliterators • Evaluate your results 

Slide Summary 26 March 2014 www.consol.de 46 Slides: https://speakerdeck.com/fstab/ Example code: https://github.com/consol/ Twitter: @fstabr