No abstract

Several recovery techniques for parallel iterative methods are presented. First, the implementation of checkpoints in parallel iterative methods is described and analyzed. Then a simple checkpoint-free fault-tolerant scheme for parallel iterative methods, the lossy approach, is presented. When one processor fails and all its data is lost, the system is recovered by computing a new approximate solution using the data of the nonfailed processors. The iterative method is then restarted with this new vector. The main advantage of the lossy approach over standard checkpoint algorithms is that it does not increase the computational cost of the iterative solver when no failure occurs. Experiments are presented that compare the different techniques. The fault-tolerant FT-MPI library is used. Both iterative linear solvers and eigensolvers are considered.

The frenetic development of the current architectures places a strain on the current state-of-the-art programming environments. Harnessing the full potential of such architectures has been a tremendous task for the whole scientific computing community. We present DAGuE a generic framework for architecture aware scheduling and management of micro-tasks on distributed many-core heterogeneous architectures. Applications we consider can be represented as a Direct Acyclic Graph of tasks with labeled edges designating data dependencies. DAGs are represented in a compact, problem-size independent format that can be queried on-demand to discover data dependencies, in a totally distributed fashion. DAGuE assigns computation threads to the cores, overlaps communications and computations and uses a dynamic, fully-distributed scheduler based on cache awareness, data-locality and task priority. We demonstrate the efficiency of our approach, using several micro-benchmarks to analyze the performance of different components of the framework, and a Linear Algebra factorization as a use case.

No abstract