Showing papers by "Rudolf Eigenmann published in 2014"

Reliable and Efficient Distributed Checkpointing System for Grid Environments

Abstract: In Fine-Grained Cycle Sharing (FGCS) systems, machine owners voluntarily share their unused CPU cycles with guest jobs, as long as their performance degradation is tolerable. However, unpredictable evictions of guest jobs lead to fluctuating completion times. Checkpoint-recovery is an attractive mechanism for recovering from such "failures". Today's FGCS systems often use expensive, high-performance dedicated checkpoint servers. However, in geographically distributed clusters, this may incur high checkpoint transfer latencies. In this paper we present a distributed checkpointing system called Falcon that uses available disk resources of the FGCS machines as shared checkpoint repositories. However, an unavailable storage host may lead to loss of checkpoint data. Therefore, we model the failures of a storage host and develop a prediction algorithm for choosing reliable checkpoint repositories. We experiment with Falcon in the university-wide Condor testbed at Purdue and show improved and consistent performance for guest jobs in the presence of irregular resource availability.

\pi Abstraction: Parallelism-Aware Array Data Flow Analysis for OpenMP

Abstract: Array data flow analysis (ADFA) is a classical method for collecting array section information in sequential programs. When applying ADFA to parallel OpenMP programs, array access information needs to be analyzed in loops whose iteration spaces are partitioned across threads. The analysis involves symbolic expressions that are functions of the original loop iteration spaces, subscript expressions and thread numbers. Adequate representations of, and operations on, these expressions can be critical for the accuracy of the analysis. This paper presents a new ADFA compiler framework for OpenMP programs. We introduce the \(\pi \) operator to abstractly represent the parallelism effects in array section expressions and improve the accuracy of the cross-thread analysis during data flow computation. We also present a novel delayed symbolic evaluation technique that enables all array section operations in the data flow computation to be performed fully accurately. Using four NAS OpenMP benchmarks, we show that the \(\pi \) operator improves array section operations’ accuracy (i.e., reduces conservative operations) during data flow computation by \(66\) %, on average, compared to the best alternative. In addition, it reduces the number of terms, and thus the complexity of computed array sections by \(33\) %, on average. We also show that delayed symbolic evaluation eliminates conservative operations and does so without significant increase in complexity when combined with \(\pi \) operators.