Examples with ClientRouter org.nd4j.parameterserver.distributed.logic.ClientRouter used on opensource projects

Example 1 with ClientRouter

use of org.nd4j.parameterserver.distributed.logic.ClientRouter in project nd4j by deeplearning4j.

the class VoidParameterServerStressTest method testPerformanceUnicast1.

/**
 * This is one of the MOST IMPORTANT tests
 */
@Test
public void testPerformanceUnicast1() {
    List<String> list = new ArrayList<>();
    for (int t = 0; t < 1; t++) {
        list.add("127.0.0.1:3838" + t);
    }
    VoidConfiguration voidConfiguration = VoidConfiguration.builder().unicastPort(49823).numberOfShards(list.size()).shardAddresses(list).build();
    VoidParameterServer[] shards = new VoidParameterServer[list.size()];
    for (int t = 0; t < shards.length; t++) {
        shards[t] = new VoidParameterServer(NodeRole.SHARD);
        Transport transport = new RoutedTransport();
        transport.setIpAndPort("127.0.0.1", Integer.valueOf("3838" + t));
        shards[t].setShardIndex((short) t);
        shards[t].init(voidConfiguration, transport, new SkipGramTrainer());
        assertEquals(NodeRole.SHARD, shards[t].getNodeRole());
    }
    VoidParameterServer clientNode = new VoidParameterServer(NodeRole.CLIENT);
    RoutedTransport transport = new RoutedTransport();
    ClientRouter router = new InterleavedRouter(0);
    transport.setRouter(router);
    transport.setIpAndPort("127.0.0.1", voidConfiguration.getUnicastPort());
    router.init(voidConfiguration, transport);
    clientNode.init(voidConfiguration, transport, new SkipGramTrainer());
    assertEquals(NodeRole.CLIENT, clientNode.getNodeRole());
    final List<Long> times = new CopyOnWriteArrayList<>();
    // at this point, everything should be started, time for tests
    clientNode.initializeSeqVec(100, NUM_WORDS, 123, 25, true, false);
    log.info("Initialization finished, going to tests...");
    Thread[] threads = new Thread[4];
    for (int t = 0; t < threads.length; t++) {
        final int e = t;
        threads[t] = new Thread(() -> {
            List<Long> results = new ArrayList<>();
            int chunk = NUM_WORDS / threads.length;
            int start = e * chunk;
            int end = (e + 1) * chunk;
            for (int i = 0; i < 200; i++) {
                long time1 = System.nanoTime();
                INDArray array = clientNode.getVector(RandomUtils.nextInt(start, end));
                long time2 = System.nanoTime();
                results.add(time2 - time1);
                if ((i + 1) % 100 == 0)
                    log.info("Thread {} cnt {}", e, i + 1);
            }
            times.addAll(results);
        });
        threads[t].setDaemon(true);
        threads[t].start();
    }
    for (int t = 0; t < threads.length; t++) {
        try {
            threads[t].join();
        } catch (Exception e) {
        }
    }
    List<Long> newTimes = new ArrayList<>(times);
    Collections.sort(newTimes);
    log.info("p50: {} us", newTimes.get(newTimes.size() / 2) / 1000);
    // shutdown everything
    for (VoidParameterServer shard : shards) {
        shard.getTransport().shutdown();
    }
    clientNode.getTransport().shutdown();
}

Example 2 with ClientRouter

use of org.nd4j.parameterserver.distributed.logic.ClientRouter in project nd4j by deeplearning4j.

the class RoutedTransportTest method testMessaging1.

/**
 * This test
 *
 * @throws Exception
 */
@Test
public void testMessaging1() throws Exception {
    List<String> list = new ArrayList<>();
    for (int t = 0; t < 5; t++) {
        list.add("127.0.0.1:3838" + t);
    }
    VoidConfiguration voidConfiguration = // this port will be used only by client
    VoidConfiguration.builder().shardAddresses(list).unicastPort(43120).build();
    // first of all we start shards
    RoutedTransport[] transports = new RoutedTransport[list.size()];
    for (int t = 0; t < transports.length; t++) {
        Clipboard clipboard = new Clipboard();
        transports[t] = new RoutedTransport();
        transports[t].setIpAndPort("127.0.0.1", Integer.valueOf("3838" + t));
        transports[t].init(voidConfiguration, clipboard, NodeRole.SHARD, "127.0.0.1", voidConfiguration.getUnicastPort(), (short) t);
    }
    for (int t = 0; t < transports.length; t++) {
        transports[t].launch(Transport.ThreadingModel.DEDICATED_THREADS);
    }
    // now we start client, for this test we'll have only one client
    Clipboard clipboard = new Clipboard();
    RoutedTransport clientTransport = new RoutedTransport();
    clientTransport.setIpAndPort("127.0.0.1", voidConfiguration.getUnicastPort());
    // setRouter call should be called before init, and we need
    ClientRouter router = new InterleavedRouter(0);
    clientTransport.setRouter(router);
    router.init(voidConfiguration, clientTransport);
    clientTransport.init(voidConfiguration, clipboard, NodeRole.CLIENT, "127.0.0.1", voidConfiguration.getUnicastPort(), (short) -1);
    clientTransport.launch(Transport.ThreadingModel.DEDICATED_THREADS);
    // we send message somewhere
    VoidMessage message = new IntroductionRequestMessage("127.0.0.1", voidConfiguration.getUnicastPort());
    clientTransport.sendMessage(message);
    Thread.sleep(500);
    message = transports[0].messages.poll(1, TimeUnit.SECONDS);
    assertNotEquals(null, message);
    for (int t = 1; t < transports.length; t++) {
        message = transports[t].messages.poll(1, TimeUnit.SECONDS);
        assertEquals(null, message);
    }
    /**
     * This is very important part, shutting down all transports
     */
    for (RoutedTransport transport : transports) {
        transport.shutdown();
    }
    clientTransport.shutdown();
}

Example 3 with ClientRouter

use of org.nd4j.parameterserver.distributed.logic.ClientRouter in project nd4j by deeplearning4j.

the class VoidParameterServerStressTest method testPerformanceUnicast2.

/**
 * This is second super-important test for unicast transport.
 * Here we send non-blocking messages
 */
@Test
@Ignore
public void testPerformanceUnicast2() {
    List<String> list = new ArrayList<>();
    for (int t = 0; t < 5; t++) {
        list.add("127.0.0.1:3838" + t);
    }
    VoidConfiguration voidConfiguration = VoidConfiguration.builder().unicastPort(49823).numberOfShards(list.size()).shardAddresses(list).build();
    VoidParameterServer[] shards = new VoidParameterServer[list.size()];
    for (int t = 0; t < shards.length; t++) {
        shards[t] = new VoidParameterServer(NodeRole.SHARD);
        Transport transport = new RoutedTransport();
        transport.setIpAndPort("127.0.0.1", Integer.valueOf("3838" + t));
        shards[t].setShardIndex((short) t);
        shards[t].init(voidConfiguration, transport, new SkipGramTrainer());
        assertEquals(NodeRole.SHARD, shards[t].getNodeRole());
    }
    VoidParameterServer clientNode = new VoidParameterServer();
    RoutedTransport transport = new RoutedTransport();
    ClientRouter router = new InterleavedRouter(0);
    transport.setRouter(router);
    transport.setIpAndPort("127.0.0.1", voidConfiguration.getUnicastPort());
    router.init(voidConfiguration, transport);
    clientNode.init(voidConfiguration, transport, new SkipGramTrainer());
    assertEquals(NodeRole.CLIENT, clientNode.getNodeRole());
    final List<Long> times = new CopyOnWriteArrayList<>();
    // at this point, everything should be started, time for tests
    clientNode.initializeSeqVec(100, NUM_WORDS, 123, 25, true, false);
    log.info("Initialization finished, going to tests...");
    Thread[] threads = new Thread[4];
    for (int t = 0; t < threads.length; t++) {
        final int e = t;
        threads[t] = new Thread(() -> {
            List<Long> results = new ArrayList<>();
            int chunk = NUM_WORDS / threads.length;
            int start = e * chunk;
            int end = (e + 1) * chunk;
            for (int i = 0; i < 200; i++) {
                Frame<SkipGramRequestMessage> frame = new Frame<>(BasicSequenceProvider.getInstance().getNextValue());
                for (int f = 0; f < 128; f++) {
                    frame.stackMessage(getSGRM());
                }
                long time1 = System.nanoTime();
                clientNode.execDistributed(frame);
                long time2 = System.nanoTime();
                results.add(time2 - time1);
                if ((i + 1) % 100 == 0)
                    log.info("Thread {} cnt {}", e, i + 1);
            }
            times.addAll(results);
        });
        threads[t].setDaemon(true);
        threads[t].start();
    }
    for (int t = 0; t < threads.length; t++) {
        try {
            threads[t].join();
        } catch (Exception e) {
        }
    }
    List<Long> newTimes = new ArrayList<>(times);
    Collections.sort(newTimes);
    log.info("p50: {} us", newTimes.get(newTimes.size() / 2) / 1000);
    // shutdown everything
    for (VoidParameterServer shard : shards) {
        shard.getTransport().shutdown();
    }
    clientNode.getTransport().shutdown();
}