Reference no: EM131040249 , Length: word count:1200

Question:

I am looking for an answers to those questions. i would like to have the solution for them with fall explanationsQuestion 1: Identify the choice that best completes the statement or answers the question.

1. Scalability is a measure of a parallel systems capacity to increase speedup in proportion to the problem size.(T/F)

2. if a problem size w has a serial component ws. w ws is an upper bound on its speedup. no matter how many processing elements are used. (T/F)

3. A hypercube with 2nd nodes can he regarded as a d-dimensional mesh with two, nodes in each dimension. (T/F)

4. A linear array or a ring) composed of 2nd nodes can be embedded into a d-dimensional hypercube by mapping node i of the linear array onto node. (T/F)

5. In cut through routing, all small messages (flits) go the same routine. (T/F)

6. As Ions as we increase the problem size, the parallel efficiency can he improved. (T/F)

7. There may have many approaches to decompose an application into tasks. (T/F)

8. Recursive decomposition is not suitable for questions that can be solved by divide-and -conquer technique. (T/F)

9. In store and forward routing a message can only be passed to the next node after completely received. (T/F)

10. Centralized load-balancing schemes arc usual easier to implement than distributed schemes, but may have limited satiability. (T/F)

11. Exploratory decomposition is a powerful and commonly used method for deriving concurrency) in algorithms that operate on large data structures, (T/F)

12. Efficiency is a measure that captures the relative benefit of solving a problem in parallel. (T/F)

13. Which of the following statements about high performance computing is incorrect?
a. both parallel computing and grid computing belong to high performance computing.
b. High performance computing can save computing time, solve large problem, provide concurrency.
e. High performance computing is needed because the memory speed is growing faster than the CPU speed.
d. John Von Neumann is not the one who first proposed high performance computing concept.

14. Which of the following statements about message passing and shared address space platforms is incorrect?
a. Shared space and message passing here means data communication models.
b. We cannot do message passing on a shared address space platforms.
c. Message passing requires less hardware support
d. Both uniform memory access and non-uniform memory access machines are shared address space machines.

15. Which of the following, in statements about decomposition methods is incorrect?
a. Recursive decomposition is not suitable for questions that can be solved by divide and conquer technique.
b) Different decomposition methods may produce the same result.
c. Decomposition can be regarded the first step fur doing the parallel computation.
d. Output data decomposition sometimes fails because the output is unknown.

16. Which if the following statements about message passing and shared address space platforms is incorrect?
a. Shared space and message passing here means data communication models
b. We cannot do maize passing on a shared address space platforms
c. Message passing requires less hardware support
d. Both uniform memory access and non-uniform memory access machines arc shared address space machines.

17. Which of the following statements about decomposition is incorrect?
a. There may have approaches to decompose an application into tasks,
b. We usually need many decomposition methods working together to get a good decomposition.
c. After the decomposition, we get man independent tasks, which can be executed at the same time.
d. After The decomposition we need to map obtained tasks to the different physical processors. We should consider both the communication and concurrency during the mapping.

18. Which of the following statements about network connection is incorrect?
a. Network Connection in high performance computing only means processor to processor connection.
b. Bandwidth is the bottleneck for the bus.
c. Crossbar networking is cost inefficient.
d. Multistage network is a tradeoff between bus and crossbar network.

19. Which of the following statements about the message routing methods is incorrect?
a. In store and forward routing a message can only he passed to the next node after completely received.
b. In packet routing, all packets go the same routine.
c. In cut through routing, all small messages (flits) go the same routine,
d. When designing a cluster computer (many PCs connected together). we generally, use cut-through routing, for message passing.

20. Assume you halve many processors Each processor contains a value b1. You want to read an integer value a from the keyboard and then compute and output the sum of ax bi. Which of the following functions cannot do this job?
a. MPI_Send, MP1_Recv
b. MPI Beast, MPI_Reduce
c. MPI Send. MPI Gather
d. None of the above

21. There are 16 processors in the following graph. Each node is a processor. They are connected a mesh network. Now there is a message going to be broadcasted from one node to all other nodes.

a. Labe the node in the graph from 0 to 15.
b. No suppose the node 0 contains the message. Show how to broadcast the message.
c. Compute the cost of one-to-all broadcast.

22. Compute efficiency of adding n numbers on an n-processor hypercube.

23 A parallel program has some experiment data below.

Number of processors	CPU Time in school
1	100
2	15
4	22
8	12
16	10

We consider one processor case as the sequential runtime. Compute The following value in the 16 processors Case.

a. Parallel cost

b. Overhead

c. Speedup

d. Is there a superliner speedup here?

24. On a p-node hypercube. The time to complete the entire all to all broadcast procedure is

T - _r=iΣlog^r(t, 2^t-1t_n). Show the details to get T = t, log p + t_um_p)

25. Give an example of how increasing processor utilization increase inter processor communication.