CS-802-C High Performance Computing - May 2022 Question Paper

Under each question, How to answer lists the points an answer should cover, outlined by CampusPrep from the syllabus. Check the details against your textbook.

1. 7 Marks
   aDraw the block diagram of a typical cache-based microprocessor.
   How to answer

   • Define cache-based microprocessor and its need for hiding memory latency
   • Show CPU core, L1/L2/L3 cache, MMU, bus interface and main memory blocks
   • Indicate data / address / control paths between blocks
   • Brief function of each block in one line
2. 7 Marks
   bDifferentiate Multi-core, Multithreaded and Vector processors.
   How to answer

   • Define multi-core, multithreaded and vector processors
   • Compare on architecture, execution model and parallelism type
   • Compare on use cases and examples
   • Summarize differences in tabular form
3. 14 Marks
   What is Parallelism? Discuss the different variants of parallelism with the help of an example in each case.
   How to answer

   • Define parallelism and its need in HPC
   • Describe variants e.g. data, task, bit-level, instruction-level parallelism
   • Give one example for each variant
   • State benefits / applications
4. 7 Marks
   aDiscuss C++ optimizations.
   How to answer

   • State need for C++ level optimization in HPC
   • Discuss compiler optimizations, inlining, loop unrolling
   • Discuss memory, cache, STL and object-oriented overhead optimizations
   • Give examples of optimization flags / techniques
5. 7 Marks
   bWhat do you mean by scalar profiling? Explain function and line-based profiling.
   How to answer

   • Define scalar profiling and its purpose
   • Explain function-based profiling with metrics collected
   • Explain line-based profiling with metrics collected
   • State use in performance tuning
6. 14 Marks
   Explain the blocking and rolling strategies with the help of multiplication of a sparse matrix with a vector.
   How to answer

   • Define sparse matrix-vector multiplication and its memory access problem
   • Explain blocking strategy for cache reuse with example
   • Explain rolling / strip-mining strategy with example
   • Compare benefits in locality and performance
7. 7 Marks
   aWhat is OpenMP? Explain the concepts of data scoping, work-sharing for loops and loop scheduling.
   How to answer

   • Define OpenMP as shared-memory parallel API
   • Explain data scoping clauses: shared, private, etc.
   • Explain work-sharing for loops: parallel for constructs
   • Explain loop scheduling: static, dynamic, guided
8. 7 Marks
   bWhat is non-blocking point-to-point communication? How it differs from blocking communication?
   How to answer

   • Define blocking vs non-blocking point-to-point communication
   • Explain MPI_Isend/Irecv, wait/test semantics
   • Differentiate on overlap, synchronization and buffer use
   • Give example scenario for each
9. 14 Marks
   Discuss the performance problems that can be raised with OpenMP shared-memory programming, also discuss how they can be circumvented?
   How to answer

   • Define OpenMP shared-memory programming model
   • List performance problems: race, false sharing, overhead, load imbalance
   • Explain cause and effect of each problem
   • Discuss circumvention: synchronization, padding, scheduling, affinity
10. 7 Marks
    aHow to reduce the communication overhead? Differentiate non-blocking and asynchronous communication.
    How to answer

    • List techniques to reduce communication overhead: aggregation, overlap, locality
    • Explain each technique briefly
    • Define non-blocking vs asynchronous communication
    • Differentiate on semantics and completion / synchronization
11. 7 Marks
    bDraw the flowchart for distributed-memory parallelization of the Jacobi algorithm.
    How to answer

    • Recall Jacobi iterative method and parallel decomposition
    • Show initialization, halo exchange, computation and convergence check steps
    • Draw flowchart with decision for convergence / max iterations
    • Label distributed-memory send/receive operations
12. 14 Marks
    Write a short note on the following (Any two)

    1. a)Moore's law
    2. b)Scalability Laws
    3. c)Bandwidth-based performance modelling
    4. d)Virtual topologies
    How to answer

    • Select any two topics from the four given options
    • Give definition, background and key points for each
    • Mention formulas / laws / relevance to HPC where applicable
    • Conclude with significance / application