T. M. Chilimbi. On the stability of temporal data reference profiles. In Proceedings of International Conference on Parallel Architectures and Compilation Techniques, Barcelona, Spain, 2001. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
Efficient Representation and Abstractions for Quantifying and.. - Chilimbi (2001) (19 citations) (Correct)
....application affects only a program s performance and not its correctness. In addition, our experiments indicate that hot data streams, when expressed in terms of the program loads and stores that generate the references, are relatively stable across program executions with different inputs [7]. 4. EXPLOITING LOCALITY Our exploitable locality abstraction, hot data streams, and our associated exploitable locality metrics can be used to improve data reference locality in at least four ways, which are described further in this section. First, they can help identify programs likely to ....
....miss rate reductions of up to 92 . In addition, preliminary results for an initial implementation of a hot data stream based prefetching optimization indicate cache miss rate improvements of 15 43 for three benchmarks when different data reference profiles were used as train and test profiles [7]. 5. EXPERIMENTAL EVALUATION This section presents results from applying our techniques to several programs and demonstrates that they produce compact representations of data reference behavior that support efficient analysis. We show that our hot data stream abstraction is useful for ....
T. M. Chilimbi. "On the stability of temporal data reference profiles. " In Microsoft Research, Technical Report MSR-TR-
Bursty Tracing: A Framework for Low-Overhead Temporal Profiling - Hirzel, Chilimbi (2001) (10 citations) Self-citation (Chilimbi) (Correct)
....such that all hot data streams together account for 90 of all traced data references. A hot data stream profile is a set of hot data streams and their regularity magnitudes. We measure the quality of a profile by computing its overlap with the profile of the trace of all data references [9]. Let P and Q be two hot data stream profiles. The overlap for each hot data stream is the minimum of its contribution to P and Q, possibly zero if it appears only in one profile. For example, if v.heat P = 3 and v.heat Q = 4 , then overlap(v, P, Q) 3 . The overlap of profiles P and Q is the ....
....overhead reduction techniques. we consider two hot data streams to be the same if they have the same access signature, which is the ordered list of instruction PCs that generated the reference sequence. This methodology follows Chilimbi s paper on the stability of temporal data reference profiles [9]. Our overlap corresponds to exact overlap in that paper. Note that because we want long streams for optimization, we measure overlap on streams of length 10 40, whereas [9] measures overlap on streams of length 2 100. 5.3 Profile Quality We evaluate how longer bursts improve the quality of ....
[Article contains additional citation context not shown here]
Trishul Chilimbi. On the stability of temporal data reference profiles. In Parallel Architectures and Compilation Techniques (PACT), 2001.
Predicting Whole-Program Locality Through Reuse Distance Analysis - Ding, Zhong (2003) (6 citations) (Correct)
No context found.
T. M. Chilimbi. On the stability of temporal data reference profiles. In Proceedings of International Conference on Parallel Architectures and Compilation Techniques, Barcelona, Spain, 2001.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC