Examples with MPIException mpi.MPIException used on opensource projects

Example 1 with MPIException

use of mpi.MPIException in project twister2 by DSC-SPIDAL.

the class MPIProcess method createResourcePlan.

private static ResourcePlan createResourcePlan(Config config) {
    try {
        int rank = MPI.COMM_WORLD.getRank();
        ResourcePlan resourcePlan = new ResourcePlan(MPIContext.clusterType(config), MPI.COMM_WORLD.getRank());
        String processName = MPI.getProcessorName();
        char[] processNameChars = new char[processName.length()];
        int length = processNameChars.length;
        processName.getChars(0, length, processNameChars, 0);
        IntBuffer countSend = MPI.newIntBuffer(1);
        int worldSize = MPI.COMM_WORLD.getSize();
        IntBuffer countReceive = MPI.newIntBuffer(worldSize);
        // now calculate the total number of characters
        countSend.put(length);
        MPI.COMM_WORLD.allGather(countSend, 1, MPI.INT, countReceive, 1, MPI.INT);
        int[] receiveSizes = new int[worldSize];
        int[] displacements = new int[worldSize];
        int sum = 0;
        for (int i = 0; i < worldSize; i++) {
            receiveSizes[i] = countReceive.get(i);
            displacements[i] = sum;
            sum += receiveSizes[i];
        }
        // first we need to send the expected number of characters
        // MPI.COMM_WORLD.allGather(countSend, 1, MPI.INT, countReceive, worldSize, MPI.INT);
        // now we need to send this to all the nodes
        CharBuffer sendBuffer = MPI.newCharBuffer(length);
        CharBuffer receiveBuffer = MPI.newCharBuffer(sum);
        sendBuffer.append(processName);
        // now lets receive the process names of each rank
        MPI.COMM_WORLD.allGatherv(sendBuffer, length, MPI.CHAR, receiveBuffer, receiveSizes, displacements, MPI.CHAR);
        Map<Integer, String> processNames = new HashMap<>();
        for (int i = 0; i < receiveSizes.length; i++) {
            char[] c = new char[receiveSizes[i]];
            receiveBuffer.get(c);
            processNames.put(i, new String(c));
            LOG.log(Level.FINE, String.format("Process %d name: %s", i, processNames.get(i)));
        }
        // now lets add the containers
        addContainers(config, resourcePlan, processNames);
        return resourcePlan;
    } catch (MPIException e) {
        throw new RuntimeException("Failed to communicate", e);
    }
}

Example 2 with MPIException

use of mpi.MPIException in project twister2 by DSC-SPIDAL.

the class MPIProcess method worker.

private static void worker(Config config, int rank) {
    // lets create the resource plan
    ResourcePlan resourcePlan = createResourcePlan(config);
    String containerClass = MPIContext.containerClass(config);
    IContainer container;
    try {
        Object object = ReflectionUtils.newInstance(containerClass);
        container = (IContainer) object;
        LOG.log(Level.FINE, "loaded container class: " + containerClass);
    } catch (ClassNotFoundException | InstantiationException | IllegalAccessException e) {
        LOG.log(Level.SEVERE, String.format("failed to load the container class %s", containerClass), e);
        throw new RuntimeException(e);
    }
    // lets do a barrier here so everyone is synchronized at the start
    try {
        MPI.COMM_WORLD.barrier();
        LOG.log(Level.FINE, String.format("Worker %d: the cluster is ready...", rank));
    } catch (MPIException e) {
        LOG.log(Level.SEVERE, "Failed to synchronize the workers at the start");
        throw new RuntimeException(e);
    }
    // now initialize the container
    container.init(config, rank, resourcePlan);
}

Example 3 with MPIException

use of mpi.MPIException in project twister2 by DSC-SPIDAL.

the class TWSMPIChannel method progress.

/**
 * Progress the communications that are pending
 */
public void progress() {
    // we should rate limit here
    while (pendingSends.size() > 0) {
        // post the message
        MPISendRequests sendRequests = pendingSends.poll();
        // post the send
        if (sendRequests != null) {
            postMessage(sendRequests);
            waitForCompletionSends.add(sendRequests);
        }
    }
    for (int i = 0; i < registeredReceives.size(); i++) {
        MPIReceiveRequests receiveRequests = registeredReceives.get(i);
        if (debug) {
            LOG.info(String.format("%d available receive %d %d %s", executor, receiveRequests.rank, receiveRequests.availableBuffers.size(), receiveRequests.availableBuffers.peek()));
        }
        // okay we have more buffers to be posted
        if (receiveRequests.availableBuffers.size() > 0) {
            postReceive(receiveRequests);
        }
    }
    Iterator<MPISendRequests> sendRequestsIterator = waitForCompletionSends.iterator();
    boolean canProgress = true;
    while (sendRequestsIterator.hasNext() && canProgress) {
        MPISendRequests sendRequests = sendRequestsIterator.next();
        Iterator<MPIRequest> requestIterator = sendRequests.pendingSends.iterator();
        while (requestIterator.hasNext()) {
            MPIRequest r = requestIterator.next();
            try {
                Status status = r.request.testStatus();
                // this request has finished
                if (status != null) {
                    completedSendCount++;
                    requestIterator.remove();
                } else {
                    canProgress = false;
                    break;
                }
            } catch (MPIException e) {
                throw new RuntimeException("Failed to complete the send to: " + sendRequests.rank, e);
            }
        }
        // ideally we should be able to call for each finish of the buffer
        if (sendRequests.pendingSends.size() == 0) {
            sendRequests.callback.onSendComplete(sendRequests.rank, sendRequests.edge, sendRequests.message);
            sendRequestsIterator.remove();
        }
    }
    if (false) {
        LOG.info(String.format("%d sending - sent %d comp send %d receive %d pend recv %d pending sends %d waiting %d", executor, sendCount, completedSendCount, receiveCount, pendingReceiveCount, pendingSends.size(), waitForCompletionSends.size()));
    }
    for (int i = 0; i < registeredReceives.size(); i++) {
        MPIReceiveRequests receiveRequests = registeredReceives.get(i);
        try {
            Iterator<MPIRequest> requestIterator = receiveRequests.pendingRequests.iterator();
            while (requestIterator.hasNext()) {
                MPIRequest r = requestIterator.next();
                Status status = r.request.testStatus();
                if (status != null) {
                    if (!status.isCancelled()) {
                        // LOG.info(String.format("%d Receive completed: from %d size %d %d",
                        // executor, receiveRequests.rank, status.getCount(MPI.BYTE), ++receiveCount));
                        ++receiveCount;
                        // lets call the callback about the receive complete
                        r.buffer.setSize(status.getCount(MPI.BYTE));
                        receiveRequests.callback.onReceiveComplete(receiveRequests.rank, receiveRequests.edge, r.buffer);
                        // LOG.info(String.format("%d finished calling the on complete method", executor));
                        requestIterator.remove();
                    } else {
                        throw new RuntimeException("MPI receive request cancelled");
                    }
                } else {
                    break;
                }
            }
        // this request has completed
        } catch (MPIException e) {
            LOG.log(Level.SEVERE, "Twister2Network failure", e);
            throw new RuntimeException("Twister2Network failure", e);
        }
    }
}

Example 4 with MPIException

use of mpi.MPIException in project twister2 by DSC-SPIDAL.

the class TWSMPIChannel method postMessage.

/**
 * Send a message to the given rank.
 *
 * @param requests the message
 */
private void postMessage(MPISendRequests requests) {
    MPIMessage message = requests.message;
    for (int i = 0; i < message.getBuffers().size(); i++) {
        try {
            sendCount++;
            MPIBuffer buffer = message.getBuffers().get(i);
            Request request = comm.iSend(buffer.getByteBuffer(), buffer.getSize(), MPI.BYTE, requests.rank, message.getHeader().getEdge());
            // register to the loop to make progress on the send
            requests.pendingSends.add(new MPIRequest(request, buffer));
        } catch (MPIException e) {
            throw new RuntimeException("Failed to send message to rank: " + requests.rank);
        }
    }
}

Example 5 with MPIException

use of mpi.MPIException in project twister2 by DSC-SPIDAL.

the class MPIProcess method main.

public static void main(String[] args) {
    Options cmdOptions = null;
    try {
        MPI.Init(args);
        int rank = MPI.COMM_WORLD.getRank();
        int size = MPI.COMM_WORLD.getSize();
        cmdOptions = setupOptions();
        CommandLineParser parser = new DefaultParser();
        // parse the help options first.
        CommandLine cmd = parser.parse(cmdOptions, args);
        // load the configuration
        // we are loading the configuration for all the components
        Config config = loadConfigurations(cmd, rank);
        // normal worker
        LOG.log(Level.FINE, "A worker process is starting...");
        worker(config, rank);
    } catch (MPIException e) {
        LOG.log(Level.SEVERE, "Failed the MPI process", e);
        throw new RuntimeException(e);
    } catch (ParseException e) {
        HelpFormatter formatter = new HelpFormatter();
        formatter.printHelp("SubmitterMain", cmdOptions);
        throw new RuntimeException("Error parsing command line options: ", e);
    } finally {
        try {
            MPI.Finalize();
        } catch (MPIException ignore) {
        }
    }
}