DidaWiki

This is the list of arguments actually presented during the course in the A.A. 2009-2010.

• Evolution of computer architectures (multi/many core, clusters, grids, clouds) and urgency of effective parallel/distributed programming models and tools
• Problems in classical parallel/distributed programming
  • Distinction between distributed and parallel based on computational grain
  • Definition of mapping and scheduling problems
  • Communications and synchronization
  • Computation grain
  • Load balancing, reliability, security and autonomic management
• Definition of structured parallel/distributed programming model
  • Patterns/Skeletons
  • Key features of a pattern
  • User programmers and system programmer views
  • Layered implementation
  • Pros and cons
• Skeleton/patterns for parallel/distributed programming
  • Techniques for discovering patterns
  • Classification of patterns (data, stream, control parallel)
  • Typical patterns
  • Cole's manifesto principles (with extensions)
• Skeleton frameworks
  • Definition of algorithmic skeleton
  • Language based vs. library based skeleton frameworks
• Implementation of skeleton frameworks
  • Implementation template technology
  • Macro data flow technology
• Fixed vs. variable skeleton set frameworks: how to handle expandability of the skeleton set using macro data flow implementation
• Advanced features
  • Portability (virtual machine based vs. compiler based)
  • Adaptivity
    • General principles
    • Behavioural skeletons with autonomic management of non functional features
    • Hierarchical management
    • Multiconcern management
• Performance modelling of structured parallel/distributed computations
  • Measures of interest
    • Service and completion time
    • Speedup and scalability
    • Efficiency, isoefficiency
  • Performance models for farm, pipeline and data parallel skeletons
  • Performance models using queue theory results
  • How to derive a performance model for a generic skeleton.
• Design patterns
  • Concept of design pattern
  • Parallel design patterns
  • Sample parallel patterns
  • Perspectives and differences with algorithmic skeletons
• Software components
  • Definition
  • Component composition (use/provide ports, other kind of ports)
  • Usage, usage in a structured framework
  • Sample component frameworks (Java Beans, Fractal, SCA). Sample SCA code
  • Perspectives and differences with algorithmic skeletons and design patterns.
• Workflows
  • Definition
  • Typical usage
  • Scheduling and fault tolerance
  • Perspectives and differences with algorithmic skeletons and design patterns.
• Low level frameworks to support structured parallel/distributed programming
  • POSIX-TCP/IP (targets clusters and network of workstations
    • Client server sample implementation
    • Tools and command line utilities (ssh, scp, rsync)
    • Implementation of collective operations with UDP
    • Sample code in C/C++ and Java
  • OpenMP (targets multicores)
    • Directives and run time primitives
    • Sample code
    • Sample implementation of structured code
  • RPC/RMI (targeting clusters and networks of workstations)
    • Concepts
    • Sample code in C (SUN rpcgen), Java (RMI) and ProActive)
  • Measuring and modelling POSIX/TCP platform
    • Mechanisms used to measure time (wall clock, clock, per process)
    • Simple modelling of TCP communications (approximating communication costs at the steady state)
• Wireless networks
  • Concept
  • Infrastructure and ad hoc modes
  • Channel access protocols
  • 802.11
  • Usage in structured context
• Peer-to-peer techniques
  • Concept
  • Centralized, distributed and hierarchical approaches
  • Typical algorithms and applications
  • Usage in structured contexts
• Client server model vs. peer to peer: perspectives
• Concepts and perspectives on parallel programming from the Berkeley paper “A View of the Parallel Computing Landscape”
• Tools and frameworks:
  • Sample skeleton framework: Muesli. Skeletons provided, installation, sample code.
  • Sample skeleton framework: Calcium/ProActive. Skeletons provided, installation, sample code.
  • Sample component framework: ProActiveGCM
  • Sample message passing framework: MPI. Point to point primitives, collective primitives, compilation and run on clusters.
  • Sample thread framework: OpenMPI.