Grid'5000 user report for Jens Gustedt

Jump to: navigation, search

User information

Jens Gustedt (users, user, account-manager, nancy, ml-users user)
More user information in the user management interface.

Experiments

  • benchmarking of par::step (Programming) [achieved]
    Description: We are benchmarking our library parXXL (formerly SSCRAP) that is conceived for large scale coarse grained computing. This concerns a variaty of basic algorithms in the layer par::step which will be benched as widely as possible on differently composed platforms.
    Results:
    More information here
  • benchmarking of par::cell and par::cellnet (Networking) [achieved]
    Description: The par::cell layer of parXXL (formerly PARCEL) implements cellular networks on parallel and distributed platforms. It is now ported to sit on top of par::mem (what previously was SSCRAP) such that it benefits from the variaty of architectures and the efficiency of that parXXL provides. Thereby we mix aspects of fine grained programming (cells) with coarse grained architectures.
    Results:
    More information here
  • Wrekavoc a tool for emulating heterogeneity (Other) [in progress]
    Description: Computer science and especially heterogeneous distributed computing is an experimental science. Simulation, emulation, or in-situ implementation are complementary methodologies to conduct experiments in this context. In this paper we address the problem of defining and controlling the heterogeneity of a platform. We evaluate the proposed solution, called Wrekavoc, with micro-benchmark and by implementing algorithms of the literature.
    Results: In this work we propose an emulation tool called Wrekavoc. The goal of Wrekavoc is to define and control the heterogeneity of a given platform by degrading CPU, network or memory capabilities of each node composing this platform. Our current implementation of Wrekavoc have been tested on an homogeneous cluster. We have shown that configuring a set of nodes is very fast. Micro-benchmarks show that we are able to independently degrade CPU, bandwidth and latency to the desired values. Tests on algorithms of the literature (load balancing and matrix multiplication) confirm the previous results and show that Wrekavoc is a suitable tool for developing, studying and comparing algorithms in heterogeneous environments.
  • Distributed Generation of Large Scale Permutations (Application) [achieved]
    Description:

    For testing and benchmarking the generation of large random input data with known probability distributions is crucial. In [GUSTEDT:2008:INRIA-00000900:2], we show how to uniformly distribute data at random in two related settings: coarse grained parallelism and external memory. In contrast to previously known work for parallel setups, our method is able to fulfill the three criteria of uniformity, work-optimality and balance among the processors simultaneously. To guarantee the uniformity we investigate the matrix of communication requests between the processors. We show that its distribution is a generalization of the multivariate hypergeometric distribution and we give algorithms to sample it efficiently in the two settings.

    If instead of shuffling existing data we constrain ourselves to produce permutations of integers 0, ..., n for some large number n, solutions that are much more efficient become possible. First, we are able to show that the communication mentioned above can be improved by using adapted compression techniques. In particular the information theoretic lower bound of Theta(n log p) (instead of Theta(n log n)) for the overall communication can be matched up to a constant factor. Second, if instead of communicating data (integers in that case) we generate them in place, we achieve a scheme that allows for a trade-off between the number of random bits (ie the quality of the target distribution) and the total running time of the algorithm. This approach is presented in [GUSTEDT:2008:INRIA-00312131:1].


    Results:
  • Ordered Read-Write Locks (Programming) [achieved]
    Description: This experiment uses ORWL as control structure to compare block- and optimized layouts.
    Results: The experiments tend to show better results (i.e. smaller execution time) for optimized-layout over block-layout while allowing to scale the amount of computing threads (ORWL control structure).

Publications

Collaborations

    Success stories and benefits from Grid'5000

    • Success stories
      • First test for sorting with SSCRAP on GdX
      • Scalability tests for ParCel.
      • Generation of permutations on a large scale

    last update: 2011-06-22 11:20:00

    Personal tools
    Namespaces

    Variants
    Views
    Actions
    Wiki special pages