As a guest user you are not logged in or recognized by your IP address. You have
access to the Front Matter, Abstracts, Author Index, Subject Index and the full
text of Open Access publications.
A main issue of fault tolerance techniques for High-Performance applications, based on checkpointing and rollback recovery, is related to the feasibility of statically analyzing the induced overhead. In this paper we show that, under the hypothesis of a general unstructured parallelism model (e.g. MPI, OpenMP), it is difficult, where not impossible, to achieve such an analysis. To overcome this issue we propose an approach to fault tolerance based on structured parallel programming (e.g. divide-and-conquer and parallel sort). We show which are the gains of assuming such a programming model in the case of data parallel programs: we introduce an optimized checkpointing protocol and we compare it with a Communication-Induce-Checkpointing (CIC) protocol, representing one of the most advanced solution for general “unstructured” parallelism models.
This website uses cookies
We use cookies to provide you with the best possible experience. They also allow us to analyze user behavior in order to constantly improve the website for you. Info about the privacy policy of IOS Press.
This website uses cookies
We use cookies to provide you with the best possible experience. They also allow us to analyze user behavior in order to constantly improve the website for you. Info about the privacy policy of IOS Press.