Thermal model identification of supercomputing nodes in production environment

Diversi, Roberto; Bartolini, Andrea; Beneventi, Francesco; Benini, Luca

doi:10.1109/iecon.2016.7793664

Cited by 5 publications

(17 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To extract the models all the above mentioned works [22], [21], [23] rely on the capability of testing the system with Pseudo Random Binary Sequences (PRBS) workloads, where each core, synchronously with the thermal response measurements (with a regular sub-second sampling time) can be forced to execute at in ether a low workload/power (idle) state or high workload/power (power virus) state to emulate a Gaussian distribution of the power stimulus. The binary workload is chosen as it allows to pre-characterize precisely the power consumption of each of the two workload states and thus to create input vectors not affected by measurement noise in conjunction with an exciting workload.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Robust Identification of Thermal Models for In-Production High-Performance-Computing Clusters With Machine Learning-Based Data Selection

Pittino

Diversi

Benini

et al. 2020

IEEE Trans. Comput.-Aided Des. Integr. Circuits Syst.

Self Cite

View full text Add to dashboard Cite

Power and thermal management are critical components of High-Performance-Computing (HPC) systems, due to their high power density and large total power consumption. The assessment of thermal dissipation by means of compact models directly from the thermal response of the final device enables more robust and precise thermal control strategies as well as automated diagnosis. However, when dealing with large scale systems "in production", the accuracy of learned thermal models depends on the dynamics of the power excitation, which depends also on the executed workload, and measurement nonidealities, such as quantization. In this paper we show that, using an advanced system identification algorithm, we are able to generate very accurate thermal models (average error lower than our sensors quantization step of 1 • C) for a large scale HPC system on real workloads for very long time periods. However, we also show that: 1) not all real workloads allow for the identification of a good model; 2) starting from the theory of system identification it is very difficult to evaluate if a trace of data leads to a good estimated model. We then propose and validate a set of techniques based on machine learning and deep learning algorithms for the choice of data traces to be used for model identification. We also show that deep learning techniques are absolutely necessary to correctly choose such traces up to 96% of the times.

show abstract

Section: Related Workmentioning

confidence: 99%

“…In order to evaluate the performance of the model, as in [23], we rely on a Kalman filter for the prediction of the temperature, using the identified thermal model. The filter is based on the following state space representation of the model (3)-(4):…”

Section: B Thermal Model Of a Corementioning

confidence: 99%

Robust Identification of Thermal Models for In-Production High-Performance-Computing Clusters With Machine Learning-Based Data Selection

Pittino

Diversi

Benini

et al. 2020

IEEE Trans. Comput.-Aided Des. Integr. Circuits Syst.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Two important features of these models are the possibility of obtaining asymptotically unbiased estimates of their parameters by means of least squares and the absence of stability problems of the associated optimal one step-ahead predictors [10]. Nevertheless, it has been shown in [9] and [11] that the classic MISO ARX model (1.1) is not able to describe properly the thermal dynamics of the system because the estimated models are characterized by relevant negative poles and/or complex conjugate poles. This is in contrast with the physics of thermal systems, where only real positive poles can exist.…”

Section: The Self-learning Policymentioning

confidence: 99%

“…. A loss function based on the IV approach is minimized along to get consistent estimates of , and , see [11] for the details. Finally, the identified models can be transformed into a state-space representation of the noisy ARX model (1.5) to be used in the ILP problem to proactively select the optimal usage of the resources.…”

Section: The Self-learning Policymentioning

confidence: 99%

See 1 more Smart Citation

Self-Aware Thermal Management for High-Performance Computing Processors

et al. 2018

Self Cite

View full text Add to dashboard Cite

Processors for high performance computing and server workload are today thermally constrained. To preserve a safe working temperature, state-of-the-art processors for this market segment integrates many cores on the same die and feature fine-grain power management and thermal management feedback loops implemented in hardware. However, to keep the control policy simple, these controllers fail in taking advantage on the underlining thermal heterogeneity, long thermal transients and specific user mode. In this paper, we present a self-aware thermal management framework for making each chip self-aware of its thermal and workload peculiarities and use it to maximize performance and thermal sustainability.

show abstract