next up previous
Next: About this document ...

Ceng 471 & Ceng 505 Parallel Computing
Final
Jan 27, 2009 17.30-19.30
Good Luck!
1 (40 Pts) Summation of numbers is performed both in serial and parallel ways. For parallel computation, the environment is the networked workstations and the sequential computation is also done in the same cluster. The following table is obtained;

TIME $nproc=1$ $nproc=2$ $nproc=3$ $nproc=4$ $nproc=5$ $nproc=6$  
100 0.013541 0.015713 0.015420 0.017393 0.018216 0.025741  
200 0.018175 0.018387 0.019951 0.022900 0.026010 0.030067  
500 0.036738 0.027887 0.032696 0.041647 0.433383 0.049059  
1000 0.072881 0.047775 0.060789 0.093839 0.097799 0.113722  
2000 0.145120 0.078808 0.111233 0.188817 0.143317 0.211288  
5000 0.365731 0.197714 0.269206 0.313151 0.371891 0.347993  
10000 0.749948 0.390106 0.507074 0.636745 0.665334 0.705855  
20000 1.571937 0.865633 1.086707 1.225067 1.273847 1.338983  

i
Complete the following tables.
Speed-Up $nproc=2$ $nproc=3$ $nproc=4$ $nproc=5$ $nproc=6$  
100 0.86 0.88 0.78      
200 0.99 0.91 0.79      
500 1.32 1.12 0.88      
1000 1.53 1.2 0.78      
2000 1.84 1.3 0.77      
5000 1.85 1.36 1.17      
10000 1.92 1.48 1.18      
20000 1.82 1.45 1.28      




Efficiency $nproc=2$ $nproc=3$ $nproc=4$ $nproc=5$ $nproc=6$  
100 0.43 0.29 0.19      
200 0.49 0.3 0.2      
500 0.66 0.37 0.22      
1000 0.76 0.4 0.19      
2000 0.92 0.43 0.19      
5000 0.92 0.45 0.29      
10000 0.96 0.49 0.29      
20000 0.91 0.48 0.32      
ii
Analyze the tables in detail.
























iii
How many processor should be used for a specific value of $N$? Why?





iv
Given the following table. The values are found as $(TIME/N)*10000$
N $nproc=2$ $nproc=3$ $nproc=4$ $nproc=5$ $nproc=6$  
100 1.571300 1.542000 1.739300 1.821600 2.574100  
200 0.919350 0.997550 1.145000 1.300500 1.503350  
500 0.557740 0.653920 0.832940 8.667660 0.981180  
1000 0.477750 0.607890 0.938390 0.977990 1.137220  
2000 0.394040 0.556165 0.944085 0.716585 1.056440  
5000 0.395428 0.538412 0.626302 0.743782 0.695986  
10000 0.390106 0.507074 0.636745 0.665334 0.705855  
20000 0.432817 0.543354 0.612534 0.636924 0.669492  


Can we estimate the $N$ value(s) for each column that communication time overheads computation time?
2 (30 Pts) Answer the following questions, choose only 5 of them.
v
Describe the Flynn's classification for computers. Which type of the computer we have made use of?
vi
Is it possible to have a system efficiency (E) of greater than %100? Discuss.
vii
Describe Blocking and Nonblocking Message-Passing.
viii
Compare the point-to-point and collective communications.
ix
What is embarrassingly parallel computations? Explain by an example.
x
What is the data parallel computations? Example it.
xi
Describe load balancing.
xii
What could be your criteria to choose a shared- or distributed-memory programming technique.
xiii
What is the difference between SIMD and SPMD models?
xiv
What are the main advantages of the compute clusters (Beowulf clusters)?
xv
Discuss the concept of communication overhead.
3 (30 Pts) Discuss the parallel programming/computing aspects of your presentation topic.


next up previous
Next: About this document ...
Cem Ozdogan 2009-01-31