Logo des Repositoriums

Auflistung nach Autor:in "Eitschberger, Patrick"

1 - 3 von 3
  • Zeitschriftenartikel
    Energy-Efficient Static Scheduling of Streaming Task Collections with Malleable Tasks
    (PARS-Mitteilungen: Vol. 30, Nr. 1, 2013) Kessler, Christoph; Eitschberger, Patrick; Keller, Jörg
    We investigate the energy-efficiency of streaming task collections with parallelizable or malleable tasks on a manycore processor with frequency scaling. Streaming task collections differ from classical task sets in that all tasks are running concurrently so that cores typically run several tasks that are scheduled round-robin on user level. A stream of data flows through the tasks and intermediate results are forwarded to other tasks like in a pipelined task graph. We first show the equivalence of task mapping for streaming task collections and normal task collections in the case of continuous frequency scalingunder reasonable assumptions for the user-level schedulerif a makespani.e. a throughput requirement of the streaming applicationis given and the energy consumed is to be minimized. We then show that in the case of discrete frequency scalingit might be necessary for processors to switch frequenciesand that idle times still can occurin contrast to continuous frequency scaling. We formulate the mapping of (streaming) task collections on a manycore processor with discrete frequency levels as an integer linear program. Finallywe propose two heuristics to reduce energy consumption compared to the previous results by improved load bal- ancing through the parallel execution of a parallelizable task. We evaluate the effects of the heuristics analytically and experimentally on the Intel SCC.
  • Zeitschriftenartikel
    Energy-Efficient Static Scheduling of Streaming Task Collections with Malleable Tasks
    (PARS: Parallel-Algorithmen, -Rechnerstrukturen und -Systemsoftware: Vol. 30, No. 1, 2013) Kessler, Christoph; Eitschberger, Patrick; Keller, Jörg
    We investigate the energy-efficiency of streaming task collections with parallelizable or malleable tasks on a manycore processor with frequency scaling. Streaming task collections differ from classical task sets in that all tasks are running concurrently, so that cores typically run several tasks that are scheduled round-robin on user level. A stream of data flows through the tasks and intermediate results are forwarded to other tasks like in a pipelined task graph. We first show the equivalence of task mapping for streaming task collections and normal task collections in the case of continuous frequency scaling, under reasonable assumptions for the user-level scheduler, if a makespan, i.e. a throughput requirement of the streaming application, is given and the energy consumed is to be minimized. We then show that in the case of discrete frequency scaling, it might be necessary for processors to switch frequencies, and that idle times still can occur, in contrast to continuous frequency scaling. We formulate the mapping of (streaming) task collections on a manycore processor with discrete frequency levels as an integer linear program. Finally, we propose two heuristics to reduce energy consumption compared to the previous results by improved load balancing through the parallel execution of a parallelizable task. We evaluate the effects of the heuristics analytically and experimentally on the Intel SCC.
  • Zeitschriftenartikel
    Optimizing Parallel Runtime of Cryptanalytic Algorithms by Selecting Between Word-Parallel and Bit-Serial Variants of Program Parts
    (PARS-Mitteilungen: Vol. 33, Nr. 1, 2016) Eitschberger, Patrick; Keller, Jörg
    Cryptanalytic algorithms such as dictionary attacks, that test huge numbers of keys to decrypt a ciphertext to a certain plaintext, need lots of computational resources and efficient coding, but allow large scale parallelism such as many-cores plus GPUs. Some attacks have profited from a bit-serial data representation, that allows SIMD-like coding per thread and increases the degree of parallelism. We investigate the question how to decide for distinct parts of such algorithms whether to code them in a bit-serial or normal word-parallel manner. Given bit-serial and word-parallel variants for each part of the cryptographic algorithm, we benchmark the runtime of the variants, and additionally the runtime of the conversion between the different data rep- resentations. Then we model the resulting variant selection problem as a direct graph — in the fashion of a global composition optimization problem — and find the optimal runtime by computing the shortest path from source to sink node. We evaluate our approach with the Advanced Encryption Standard (AES) and demonstrate runtime advantages.