skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Addressing Thermal and Performance Variability Issues in Dynamic Processors

Abstract

As CMOS scaling nears its end, parameter variations (process, temperature and voltage) are becoming a major concern. To overcome parameter variations and provide stability, modern processors are becoming dynamic, opportunistically adjusting voltage and frequency based on thermal and energy constraints, which negatively impacts traditional bulk-synchronous parallelism-minded hardware and software designs. As node-level architecture is growing in complexity, implementing variation control mechanisms only with hardware can be a challenging task. In this paper we investigate a software strategy to manage hardwareinduced variations, leveraging low-level monitoring/controlling mechanisms.

Authors:
 [1];  [2];  [3];  [3];  [3];  [3];  [1];  [1]
  1. Argonne National Lab. (ANL), Argonne, IL (United States)
  2. Univ. Carlos III de Madrid (Spain)
  3. Northwestern Univ., Evanston, IL (United States)
Publication Date:
Research Org.:
Argonne National Lab. (ANL), Argonne, IL (United States)
Sponsoring Org.:
USDOE Office of Science (SC)
OSTI Identifier:
1353371
Report Number(s):
ANl/MCS-TM-368
134372
DOE Contract Number:
AC02-06CH11357
Resource Type:
Technical Report
Country of Publication:
United States
Language:
English
Subject:
97 MATHEMATICS AND COMPUTING

Citation Formats

Yoshii, Kazutomo, Llopis, Pablo, Zhang, Kaicheng, Luo, Yingyi, Ogrenci-Memik, Seda, Memik, Gokhan, Sankaran, Rajesh, and Beckman, Pete. Addressing Thermal and Performance Variability Issues in Dynamic Processors. United States: N. p., 2017. Web. doi:10.2172/1353371.
Yoshii, Kazutomo, Llopis, Pablo, Zhang, Kaicheng, Luo, Yingyi, Ogrenci-Memik, Seda, Memik, Gokhan, Sankaran, Rajesh, & Beckman, Pete. Addressing Thermal and Performance Variability Issues in Dynamic Processors. United States. doi:10.2172/1353371.
Yoshii, Kazutomo, Llopis, Pablo, Zhang, Kaicheng, Luo, Yingyi, Ogrenci-Memik, Seda, Memik, Gokhan, Sankaran, Rajesh, and Beckman, Pete. Wed . "Addressing Thermal and Performance Variability Issues in Dynamic Processors". United States. doi:10.2172/1353371. https://www.osti.gov/servlets/purl/1353371.
@article{osti_1353371,
title = {Addressing Thermal and Performance Variability Issues in Dynamic Processors},
author = {Yoshii, Kazutomo and Llopis, Pablo and Zhang, Kaicheng and Luo, Yingyi and Ogrenci-Memik, Seda and Memik, Gokhan and Sankaran, Rajesh and Beckman, Pete},
abstractNote = {As CMOS scaling nears its end, parameter variations (process, temperature and voltage) are becoming a major concern. To overcome parameter variations and provide stability, modern processors are becoming dynamic, opportunistically adjusting voltage and frequency based on thermal and energy constraints, which negatively impacts traditional bulk-synchronous parallelism-minded hardware and software designs. As node-level architecture is growing in complexity, implementing variation control mechanisms only with hardware can be a challenging task. In this paper we investigate a software strategy to manage hardwareinduced variations, leveraging low-level monitoring/controlling mechanisms.},
doi = {10.2172/1353371},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Wed Mar 01 00:00:00 EST 2017},
month = {Wed Mar 01 00:00:00 EST 2017}
}

Technical Report:

Save / Share:
  • Traditional estimates of health risk are typically inflated, particularly if cancer is the dominant endpoint and there is fundamental uncertainty as to mechanism(s) of action. Risk is more realistically characterized if it accounts for joint uncertainty and interindividual variability after applying a unified probabilistic approach to the distributed parameters of all (linear as well as nonlinear) risk-extrapolation models involved. Such an approach was applied to characterize risks to potential future residents posed by trichloroethylene (TCE) in ground water at an inactive landfill site on Beale Air Force Base in California. Variability and uncertainty were addressed in exposure-route-specific estimates of appliedmore » dose, in pharmacokinetically based estimates of route-specific metabolized fractions of absorbed TCE, and in corresponding biologically effective doses estimated under a genotoxic/linear (MA{sub g}) vs. a cytotoxic/nonlinear (MA{sub c}) mechanistic assumption for TCE-induced cancer. Increased risk conditional on effective dose was estimated under MA{sub G} based on seven rodent-bioassay data sets, and under MA, based on mouse hepatotoxicity data. Mean and upper-bound estimates of combined risk calculated by the unified approach were <10{sup -6} and <10{sup -4}, respectively, while corresponding estimates based on traditional deterministic methods were >10{sup -5} and >10{sup -4}, respectively. It was estimated that no TCE-related harm is likely occur due any plausible residential exposure scenario involving the site. The unified approach illustrated is particularly suited to characterizing risks that involve uncertain and/or diverse mechanisms of action.« less
  • Traditional estimates of health risk are typically inflated, particularly if cancer is the dominant endpoint and there is fundamental uncertainty as to mechanism(s) of action. Risk is more realistically characterized if it accounts for joint uncertainty and interindividual variability within a systematic probabilistic framework to integrate the joint effects on risk of distributed parameters of all (linear as well as nonlinear) risk-extrapolation models involved. Such a framework was used to characterize risks to potential future residents posed by trichloroethylene (TCE) in ground water at an inactive landfill site on Beale Air Force Base in California. Variability and uncertainty were addressedmore » in exposure-route-specific estimates of applied dose, in pharmacokinetically based estimates of route-specific metabolized fractions of absorbed TCE, and in corresponding biologically effective doses estimated under a genotoxic/linear (MA{sub G}) vs. a cytotoxic/nonlinear (MA{sub c}) mechanistic assumption for TCE-induced cancer. Increased risk conditional on effective dose was estimated under MA{sub G} based on seven rodent-bioassay data sets, and under MA{sub c} based on mouse hepatotoxicity data. Mean and upper-bound estimates of combined risk calculated by the unified approach were <10{sup -6} and 10{sup -4}, respectively, while corresponding estimates based on traditional deterministic methods were >10{sup -5} and 10{sup -4}, respectively. It was estimated that no TCE-related harm is likely to occur due to any plausible residential exposure scenario involving the site. The systematic probabilistic framework illustrated is particularly suited to characterizing risks that involve uncertain and/or diverse mechanisms of action.« less
  • This study was designed to accomplish two objectives. The first was to provide to the US Air Force and the regulatory community quantitative procedures that they might want to consider using for addressing uncertainty and variability in exposure to better characterize potential health risk. Such methods could be used at sites where populations may now or in the future be faced with using groundwater contaminated with low concentrations of the chemical trichloroethylene (TCE). The second was to illustrate and explain the application of these procedures with respect to available data for TCE in ground water beneath an inactive landfill sitemore » that is undergoing remediation at Beale Air Force Base in California. The results from this illustration provide more detail than the more traditional conservative deterministic, screening-level calculations of risk, also computed for purposes of comparison. Application of the procedures described in this report can lead to more reasonable and equitable risk-acceptability criteria for potentially exposed populations at specific sites.« less
  • Conservative deterministic, screening-level calculations of exposure and risk commonly are used in quantitative assessments of potential human-health consequences from contaminants in environmental media. However, these calculations generally are based on multiple upper-bound point estimates of input parameters, particularly for exposure attributes, and can therefore produce results for decision makers that actually overstate the need for costly remediation. Alternatively, a more informative and quantitative characterization of health risk can be obtained by quantifying uncertainty and variability in exposure. This process is illustrated in this report for a hypothetical population at a specific site at Beale Air Force Base in California, wheremore » there is trichloroethylene (TCE) contaminated ground water and a potential for future residential use. When uncertainty and variability in exposure were addressed jointly for this case, the 95th-percentile upper-bound value of individual excess lifetime cancer risk was a factor approaching 10 lower than the most conservative deterministic estimate. Additionally, the probability of more than zero additional cases of cancer can be estimated, and in this case it is less than 0.5 for a hypothetical future residential population of up to 26,900 individuals present for any 7.6-y interval of a 70-y time period. Clearly, the results from application of this probabilistic approach can provide reasonable and equitable risk-acceptability criteria for a contaminated site.« less
  • The directive identifies a senior Superfund official responsible for reviewing and addressing specific issues at NPL (National Priorities List) sites that cannot be resolved at the Regional level and identifies criteria for NPL site referrals to this official.