Reference no: EM133040021

COMP1680 Clouds, Grids and Virtualisation - University of Greenwich

Learning Outcome 1: Characterise and critically evaluate high performance computing based architectures and their suitability for given applications.

Learning Outcome 2: Implement and execute applications using shared and distributed memory programming paradigms.

Learning Outcome 3: Describe and critically discuss the roles and applications of cloud and grid computing.

Part 1: Parallel processing using cloud computing

The company you work for is looking at investing in a new system for running parallel code, both MPI and OpenMP. They are considering either using a cloud computing platform or purchasing their own HPC equipment. You are required to write a short report analyzing the different platforms and detailing your recommendations. The report will go to both the Head of IT and the Head of Finance and so the report should be aimed at a non-specialist audience. Assume the company is a medium sized consultancy with around 50 consultants, who will likely utilize an average of 800 CPU hours each a month. Your report should include:
1) A definition of cloud computing and how it can be beneficial.
2) An analysis of the advantages and disadvantages of the different commercial platforms over a traditional HPC.
3) A cost analysis, assume any on site HPC will likely need dedicated IT support
4) Your recommendations to the company.
5) References

Part 2: To complete this assignment you will need the source code

You are provided with a two C program codes (called jacobi2d.c and gauss2d.c) that solve a rectangular 2 dimensional heat conductivity problem using the Jacobi and Gauss-Seidel iterative methods.

This code can be compiled and linked to produce a conventional executable files called jacobiSerial and gaussSerial by using the following commands:
gcc jacobi2d.c -o jacobiSerial gcc gauss2d.c -o gaussSerial
To run the executable type in the executable name: jacobiSerial or gaussSerial
As you implement each of the following 4 steps make sure that you retain and do not overwrite previous versions of your solutions.

Step 1
You are required to compute a temperature distribution for a rectangular 2D problem with boundary conditions set at top 100°C, bottom 20°C, left 30°C and right 40°C with a range of problem sizes. To do this you are required to modify the codes to:
• reflect the boundary conditions described above
• report the execution time Record the run-time of your code under a range of problem sizes using different levels of compiler optimization.
Be advised that:
• it is possible that aggressive optimization will break the code
• you will need to stop the results from printing if you are to obtain realistic measurements of the execution time.

Step 2
You are then required to modify the applications you created in step 1 to produce a basic parallel version of the codes using OpenMP. The following commands will compile your parallel version on a platform that has OpenMP installed:
gcc -fopenmp jacobiOpenmp.c -o jacobiOpenmp gcc -fopenmp gaussOpenmp.c -o gaussOpenmp
The parallel codes must include timers to report the parallel run-time of the code. This version must be tested to establish correct operation using 1, 2, 4 and 8 threads/processors, regardless of performance. (These versions may run on any platform you choose as performance is not an issue at this stage.)
Include in your report, the result for a 20x20 problem size for 1,2,4 and 8 processors to demonstrate the code works correctly.
Run the Gauss-Seidel code for only 1 iteration using 1 and 2 threads for a 20x20 problem size. Output the result along with the timings. Discuss the differences in the solutions.

Step 3
Using the cms-grid machines you are to run performance tests with the OpenMP implementation you created in step 2. This will require that you remove most of the print output from the code and increase the problem size to provide sufficient work to demonstrate useful speedup. You are expected to provide speedup results:
• for a range of problem sizes, you are unlikely to see much speedup for small domains, use at least 100x100+
• for a range of number of threads (from 2 up to 8 threads) In calculating the speedup of your parallel code you should use the optimized single processor version of your code you produced in step 1 and compare to this. You will need to apply similar compiler optimizations to your parallel code. Please list your runtimes in a suitable unit.

Step 4
Using different OpenMP directives and clauses you are to further modify your OpenMP application to improve the parallel performance. You are expected to provide results that permit comparison with those you obtained in steps 2 and 3. Comment on the differences between optimising the Jacobi and Gauss-Seidel Methods.

Your report is required to provide details of your implementation of steps 1 to 4 as described above. The report should include discussion of your solutions and provide a clear description of; the code changes you have implemented, your compilation and execution processes and your test cases. For steps 3 and 4 you are expected to provide tabular and graphical results. Comment on the differences between the two methods and the effect on parallelisation. Your zip file should provide suitably named source code files for each of your implementations.

Note: Need PART 1 only for now

Attachment:- Clouds, Grids and Virtualisation.rar