我从未见过的MPJ Express错误

Question

我得到了这个运行时错误：

MPJ Express (0.35) is started in the cluster configuration
Starting process <0> on <Tornado>
Starting process <1> on <Predator>
mpi.MPIException: Error in SimplePacker : count <1> is less than length <2>
        at mpi.SimplePackerChar.unpack(SimplePackerChar.java:105)
        at mpi.Comm.recv(Comm.java:1305)
        at mpi.Comm.Recv(Comm.java:1255)
        at PingPongVariousLengths.main(PingPongVariousLengths.java:29)
        at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.
java:57)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAcces
sorImpl.java:43)
        at java.lang.reflect.Method.invoke(Method.java:601)
        at runtime.daemon.Wrapper.execute(Wrapper.java:165)
        at runtime.daemon.Wrapper.main(Wrapper.java:180)
java.lang.reflect.InvocationTargetException
        at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.
java:57)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAcces
sorImpl.java:43)
        at java.lang.reflect.Method.invoke(Method.java:601)
        at runtime.daemon.Wrapper.execute(Wrapper.java:165)
        at runtime.daemon.Wrapper.main(Wrapper.java:180)
Caused by: mpi.MPIException: mpi.MPIException: mpi.MPIException: Error in Simple
Packer : count <1> is less than length <2>
        at mpi.Comm.Recv(Comm.java:1259)
        at PingPongVariousLengths.main(PingPongVariousLengths.java:29)
        ... 6 more
Caused by: mpi.MPIException: mpi.MPIException: Error in SimplePacker : count <1>
 is less than length <2>
        at mpi.Comm.recv(Comm.java:1317)
        at mpi.Comm.Recv(Comm.java:1255)
        ... 7 more
Caused by: mpi.MPIException: Error in SimplePacker : count <1> is less than leng
th <2>
        at mpi.SimplePackerChar.unpack(SimplePackerChar.java:105)
        at mpi.Comm.recv(Comm.java:1305)
        ... 8 more

我不明白这是什么意思，

这就是导致它的代码：

import mpi.* ;

class PingPongVariousLengths {

    static public void main(String[] args) {

        MPI.Init(args);
        int myrank = MPI.COMM_WORLD.Rank();
        int tag = 99;
        int maxlen = 104857600; //200 megabytes     104857600 characters * 2 bytes per character = 209715200 bytes total, or 200 megabytes
        int minlen = 1; // 2 bytes
        char [] sendbuff = new char [maxlen];
        char [] recvbuff = new char [maxlen];
        long speedKbps;
        long speedMbps;
        long durationseconds;
int MAX_LOOPS = 20;

for (int len = minlen; len <= maxlen; len *= 2) {
        if (myrank == 0) {
                durationseconds = 0;
                for (int i = 0; i < MAX_LOOPS; i++) {
                        long startTime = System.nanoTime();           
                        MPI.COMM_WORLD.Send(sendbuff, 0, len, MPI.CHAR, 1, tag);
                        MPI.COMM_WORLD.Recv(recvbuff, 0, len, MPI.CHAR, 1, tag);
                        long endTime = System.nanoTime();
                        long duration = endTime - startTime;
                        durationseconds = durationseconds + (duration* 10-9);
                }
                durationseconds = durationseconds / MAX_LOOPS;
                System.out.println("Average time for the ping to be sent and recived of " + (len*2) + " bytes is " + durationseconds + " seconds");
                double transferRateMb = ((len*524288.0) / durationseconds );
                System.out.println("average transferRate (megabytes) : " + transferRateMb + " megabytes per second");
        } else if (myrank == 1) {
                MPI.COMM_WORLD.Recv(recvbuff, 0, len, MPI.CHAR, 0, tag);
                MPI.COMM_WORLD.Send(recvbuff, 0, len, MPI.CHAR, 0, tag);
        }
}

        MPI.Finalize();
    }
}

导致错误的原因是什么?如何解决？

EDITTTTT

将minlength更改为2

import mpi.* ;

class PingPongVariousLengths {

    static public void main(String[] args) {

        MPI.Init(args);
        int myrank = MPI.COMM_WORLD.Rank();
        int tag = 99;
        int maxlen = 104857600; //200 megabytes     104857600 characters * 2 bytes per character = 209715200 bytes total, or 200 megabytes
        int minlen = 2; // 2 bytes
        char [] sendbuff = new char [maxlen];
        char [] recvbuff = new char [maxlen];
        long speedKbps;
        long speedMbps;
        long durationseconds;
int MAX_LOOPS = 20;

for (int len = minlen; len <= maxlen; len *= 2) {//len=*2 doubles the ping size each time
        if (myrank == 0) {
                durationseconds = 0;
                for (int i = 0; i < MAX_LOOPS; i++) {
                        long startTime = System.nanoTime();           
                        MPI.COMM_WORLD.Send(sendbuff, 0, len, MPI.CHAR, 1, tag);
                        MPI.COMM_WORLD.Recv(recvbuff, 0, len, MPI.CHAR, 1, tag);
                        long endTime = System.nanoTime();
                        long duration = endTime - startTime;
                        durationseconds = durationseconds + (duration* 10-9);// Converts nanoseconds to seconds
                }
                durationseconds = durationseconds / MAX_LOOPS;
                 //double transferRate = ((len*2.0) / durationseconds ) ; //amount of data in bytes transferred in 1 second. Currently returning 0 for every result
                //System.out.println("transferRate: " + transferRate + " bytes per second");
                System.out.println("Average time for the ping to be sent and recived of " + (len*2) + " bytes is " + durationseconds + " seconds");
                double transferRateMb = ((len*524288.0) / durationseconds );
                System.out.println("average transferRate (megabytes) : " + transferRateMb + " megabytes per second");
        } else if (myrank == 1) {
                MPI.COMM_WORLD.Recv(recvbuff, 0, len, MPI.CHAR, 0, tag);
                MPI.COMM_WORLD.Send(recvbuff, 0, len, MPI.CHAR, 0, tag);
        }
}

        MPI.Finalize();
    }
}

我得到了这个错误：

PongVariousLengths
MPJ Express (0.35) is started in the cluster configuration
Starting process <0> on <Tornado>
Starting process <1> on <Predator>
mpi.MPIException: Error in SimplePacker : count <2> is less than length <4>
        at mpi.SimplePackerChar.unpack(SimplePackerChar.java:105)
        at mpi.Comm.recv(Comm.java:1305)
        at mpi.Comm.Recv(Comm.java:1255)
        at PingPongVariousLengths.main(PingPongVariousLengths.java:25)
        at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.
java:57)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAcces
sorImpl.java:43)
        at java.lang.reflect.Method.invoke(Method.java:601)
        at runtime.daemon.Wrapper.execute(Wrapper.java:165)
        at runtime.daemon.Wrapper.main(Wrapper.java:180)
java.lang.reflect.InvocationTargetException
        at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.
java:57)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAcces
sorImpl.java:43)
        at java.lang.reflect.Method.invoke(Method.java:601)
        at runtime.daemon.Wrapper.execute(Wrapper.java:165)
        at runtime.daemon.Wrapper.main(Wrapper.java:180)
Caused by: mpi.MPIException: mpi.MPIException: mpi.MPIException: Error in Simple
Packer : count <2> is less than length <4>
        at mpi.Comm.Recv(Comm.java:1259)
        at PingPongVariousLengths.main(PingPongVariousLengths.java:25)
        ... 6 more
Caused by: mpi.MPIException: mpi.MPIException: Error in SimplePacker : count <2>
 is less than length <4>
        at mpi.Comm.recv(Comm.java:1317)
        at mpi.Comm.Recv(Comm.java:1255)
        ... 7 more
Caused by: mpi.MPIException: Error in SimplePacker : count <2> is less than leng
th <4>
        at mpi.SimplePackerChar.unpack(SimplePackerChar.java:105)
        at mpi.Comm.recv(Comm.java:1305)
        ... 8 more

编辑2

好的，经过一些试验和错误后，我在第19行注释掉了'//len *= 2)‘，注释掉了这个，程序会运行，但它总是以2个字节运行，并且在所需的20个循环之后没有停止，所以我想这就是问题所在，但是这个问题该如何解决呢？

Answer 1

您的异常似乎是由此block of code:引发的

  public void unpack(mpjbuf.Buffer mpjbuf, int length, Object buf,
                  int offset, int count) throws MPIException {

    if(count * numEls < length) {
      throw new MPIException ("Error in SimplePacker : count <"+
          (count*numEls)+"> is less than length <"+length+">");
    }

所以看起来count * numEls比length小。这一切似乎都指向您的Recv()调用(我假设是第29行)：

MPI.COMM_WORLD.Recv(recvbuff, 0, len, MPI.CHAR, 1, tag);

或

MPI.COMM_WORLD.Recv(recvbuff, 0, len, MPI.CHAR, 0, tag);

所以“计数”(len)小于长度( 2)。您将minlen设置为1(并不是注释中所说的"2“)，请尝试将其设置为2。

Answer 2

您的代码的主要问题是，您错过了一个内部循环，它在使用rank=1的进程中从i=0运行到MAX_LOOP。由于外部循环的长度不同，您使用rank=0的进程将发送len=2消息，而使用rank=1的进程将使用len=4发送消息，因为它处于下一次外部循环迭代中。如果插入print语句，您将看到第一条消息已成功完成。但对于第二次迭代，它在长度上是未匹配的。以下是代码修复：

for (int len = minlen; len <= maxlen; len *= 2) {
        if (myrank == 0) {
                durationseconds = 0;
                for (int i = 0; i < MAX_LOOPS; i++) {
                        long startTime = System.nanoTime();
                        System.out.println("Processor 0 printing len="+len);
                        MPI.COMM_WORLD.Send(sendbuff, 0, len, MPI.CHAR, 1, tag);
                        MPI.COMM_WORLD.Recv(recvbuff, 0, len, MPI.CHAR, 1, tag);
                        long endTime = System.nanoTime();
                        long duration = endTime - startTime;
                        durationseconds = durationseconds + (duration* 10-9);
                }
                durationseconds = durationseconds / MAX_LOOPS;
                System.out.println("Average time for the ping to be sent and recived of " + (len*2) + " bytes is " + durationseconds + " seconds");
                double transferRateMb = ((len*524288.0) / durationseconds );
                System.out.println("average transferRate (megabytes) : " + transferRateMb + " megabytes per second");
        } else if (myrank == 1) {
                for(int i =0; i < MAX_LOOPS; i++){
                        System.out.println("Processor 1 printing len="+len);
                        MPI.COMM_WORLD.Recv(recvbuff, 0, len, MPI.CHAR, 0, tag);
                        MPI.COMM_WORLD.Send(recvbuff, 0, len, MPI.CHAR, 0, tag);
                }
        }
}