Space-efficient time-series call-path profiling of parallel applications

• DOI: 10.1145/1654059.1654097
• Publication Date: 2009-11-17T13:30:15Z
• Authors: Zoltán Szebenyi; Felix Wolf; Brian J. N. Wylie
• Keywords: Profiling (computer programming); Call graph

The performance behavior of parallel simulations often changes considerably as the simulation progresses --- with potentially process-dependent variations temporal patterns. While call-path profiling is an established method linking a problem to context in which it occurs, call paths reveal only little information about evolution phenomena. However, generating profiles separately for thousands iterations may exceed available buffer space especially when tree large and more than one metric collected. In this paper, we present runtime approach semantic compression based on incremental clustering series single-iteration that scales terms number without sacrificing important details. Our offers low overhead by using condensed version profile data calculating distances accounts making all decisions locally.