Calibrating DRAMPower Model: A Runtime Perspective from Real-System HPC Measurements

2024-11-27

arXiv.org

AI Summary

DRAM power modeling in large-scale systems suffers from insufficient accuracy: existing tools rely on idealized or worst-case parameters from vendor datasheets, failing to capture real-world variations induced by temperature, aging, and workload dynamics—leading to substantial discrepancies between measured and predicted power. This work introduces the first runtime-driven, measurement-based dynamic calibration methodology for the DRAMPower model, leveraging empirical energy measurements from a production HPC cluster and abandoning the static datasheet-parameter paradigm. Our approach employs a custom memory microbenchmark suite, hardware-level power instrumentation, current-parameter sensitivity analysis, and least-squares inversion to enable workload-adaptive calibration of critical current parameters. Experimental evaluation demonstrates that the calibrated model reduces average energy estimation error to under 5%, markedly improving fidelity. This enables high-accuracy, power-aware system design and optimization for modern HPC and datacenter environments.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success
• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization

Abstract

The escalating energy demands of main memory have become a concern in modern computing architectures, particularly in large-scale systems, due to frequent access patterns, increasing data volumes, and the lack of efficient power management strategies. Accurate modeling of DRAM power consumption is essential to address this challenge and optimize energy efficiency. However, existing modeling tools that heavily rely on vendor-provided datasheet values lead to a big discrepancy between the estimation result and the real-word power consumption. In this work, we propose a calibration towards the DRAMPower model by leveraging runtime energy measurements collected from real-system experiments. Using custom memory benchmarks and runtime data from the HPC cluster, we refine key DRAM current parameters within the DRAMPower model, aligning its predictions more closely with real-world observations. Our calibration reduces the average energy estimation error to less than 5%, demonstrating substantial improvements in modeling accuracy and making the DRAMPower model a more reliable tool for power-aware system design and optimization.

Problem

• Accurate DRAM power modeling for energy efficiency optimization
• Addressing discrepancies in existing vendor-provided DRAM power models
• Runtime calibration of DRAMPower using real-system measurements

Innovation

• Runtime calibration using real-system energy measurements
• Refining IDD values via custom memory benchmarks
• Achieving under 5% energy estimation error