Multiprocessor Architectures

Semestr: Summer

Range: 2+2s

Completion:

Credits: 4

Programme type: Undefined

Study form: Fulltime

Course language:

Summary:

The aim of the course is to give students details on architectures of the state-of-the-art high-performance multiprocessor systems, namely those belonging to the category of parallel and scalable servers. Models of memory coherence and consistency will be explained. SW and HW techniques for their implementation will be given for computers with shared and with distributed memory. Cache coherence protocols for distributed memory systems will be extended to virtual shared memory solutions. Also, HW and SW tools for synchronization operations will be explained, namely the locks and barriers.

Keywords:

Course syllabus:

1. Models for memory coherence and consistency
2. Bus-based cache coherence protocols in shared-memory systems
3. Synchronization locks in shared-memory systems
4. Scalable synchronization locks
5. SW barriers in shared-memory systems
6. SW barriers in distributed-memory systems
7. HW barriers in distributed-memory systems
8. HW and SW techniques for virtual shared memory
9. Directory-based cache-coherence schemes
10. Cache-based cache-coherence schemes
11. Hierarchical cache-coherence schemes
12. Fast communication protocols supporting virtual shared memory
13. Dynamically reconfigurrable clusters of workstations
14. Protocols implementing partial memory consistency

Seminar syllabus:

1. SW examples of models of memory coherence and consistency
2. Case studies of bus-based cache-coherence protocols
3. Correctness proof of synchronization locks in shared memory
4. Correctness proof of synchronization distributed locks
5. HW and SW tradeoffs of synchronization lock implementations
6. Correctness proof of SW barriers
7. Comparison of SW and HW implementation of barriers
8. A case study of a directory-based cache-coherence scheme
9. A case study of a cache-based cache-coherence scheme
10. A case study of a hierarchical cache-coherence scheme
11. Implementation tradeoffs of coherence schemes
12. A case study of a communication protocol supporting virtual shared memory
13. A case study of a dynamically reconfigurrable cluster of workstations
14. A case study of an architecture with partial memory consistency

Literature:

1. P. Culler et al: Parallel Computer Architecture, Morgan Kaufmann, USA,
ISBN 1-55860-343-3

Examiners:

Lecturers:

Instructors: