Supplying your own file and process names, write .sh script to be run under sbatch supporting 7 tasks, running on 4 nodes and 1 command line argument.
Ans: Similar examples can be found in the SLURM tab.
(5 pts.)
Setting aside mpirun why would this not work for OpenMP?
Ans :OpenMP spawns Threads not Processes.
Discuss the role of MPI_Irecv and MPI_Test in allowing non-blocking inter-task MPI communication.
Ans: Execution continues past MPI_Irecv without waiting for the corresponding Send. MPI_Test is used to check that the corresponding Send has been Received.
Encode the Virtual Task Topology shown in the following image by filling in the question marks in the code outline below it.
int reorder= ?; /* The processes need to be reordered to improve performance.*/
int index[] = { ?};
int edges[] = { ?};
MPI_Comm MPI_Tree_World;
MPI_Graph_create(MPI_COMM_WORLD, ?, index, edges, reorder,&MPI_Tree_World);
The Answer
Provide a struct that would be supported by the following MPI code
MPI_Datatype MPI_MYTYPE;
MPI_Datatype type[2] = { MPI_CHAR,MPI_INT};
int blocklen[2] = {30, 2};
MPI_Aint disp[2];
MPI_Aint start_address;
MPI_Aint address;
MPI_Get_address( &example, &start_address);
MPI_Get_address( &example.field1 , &address);
disp[0] = address - start_address;
MPI_Get_address( &example.field2 , &address);
disp[1] = address - start_address;
MPI_Type_create_struct(2, blocklen, disp, type, &MPI_MYTYPE);
MPI_Type_commit(&MPI_MYTYPE);
MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
Answer: Something like:
struct mytpe {char field1[30];
int field2[2];
};
- In general, what is the major limiting factor coding using Pseudorandom number generators?
At some point Psudorandom number generators cycle through the same sequence of values.
- Using Pseudorandom number generators for Cluster Computing generates an additional complication. What is this complication?
- In the class notes there are 2 approaches to solving this additional complication. What are they and
what are their limitations?
Ans: 1. Use different seeds for each process. You need to be sure the seeds do not produce related lists.
2. Chunk the list for a given seed. This shortens the repeat cycle.
Discuss the Algorithmic Complexity of factoring a product of two primes to attack RSA.
Ans: The number of divisions you need to perform growns as the square root of the number you are trying to factor.
In particular, using the simple factoring method presented in class, suppose I double the number of nodes available to break an RSA instance in half the time, what "arithmetic" might an RSA practitioner use defeat my efforts?