Itkovian.net

I am a co-author for a paper accepted at PACT 2006, titled "Performance Prediction based on Inherent Program Similarity", by K. Hoste, A. Phansalkar, L. Eeckhout, A. Georges, L.K. John and K. De Bosschere. Kenneth Hoste will present it tomorrow.

The paper abstract reads as follows. A key challenge in benchmarking is to predict the performance of an application of interest on a number of platforms in order to determine which platform yields the best performance. This paper proposes an approach for doing this.
We measure a number of microarchitecture-independent characteristics from the application of interest, and relate these characteristics to the characteristics of the programs from a previously profiled benchmark suite. Based on the similarity of the application of interest with programs in the benchmark suite, we make a performance prediction of the application of interest.
We propose and evaluate three approaches (normalization, principal components analysis and genetic algorithm) to transform the raw data set of microarchitecture-independent characteristics into a benchmark space in which the relative distance is a measure for the relative performance differences.
We evaluate our approach using all of the SPEC CPU2000 benchmarks and real hardware performance numbers from the SPEC website. Our framework estimates per-benchmark machine ranks with a 0.89 average and a 0.80 worst case rank correlation coefficient.

We show that it is possible to make a very good prediction of which machine will be the best performing machine for an application of which we know only the microarchitecture independent characteristics.