Central vs. distributed dynamic thermal management for multi-core processors

Kadin, Michael; Reda, Sherief; Uht, Augustus K.

doi:10.1145/1531542.1531577

Cited by 19 publications

(13 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [8], [9], [10] the authors show the benefit of feedback-control approaches vs. open loop policies based on temperature thresholding heuristics. Model predictive controllers (MPC) [11] [6] outperform classic feedback controller, which cannot take into account hard constraints in the state space.…”

Section: A Related Workmentioning

confidence: 99%

“…. +β 1,1 · u 1 (k − s) + β 2,s · u 2 (k − 1) + · · · + e(k) (9) where T is the temperature of the core (the model output), s is the model order, u i (·) are the model inputs (the dissipated power of the core P EC , the ambient temperature T AMB and the temperatures of the neighbours units), e(k) is a stochastic white process with null expected value representing the model error and α i , β i, j are the identified parameters. As shown in Section II and in Eq.…”

Section: Self-calibration Routinementioning

confidence: 99%

See 1 more Smart Citation

A distributed and self-calibrating model-predictive controller for energy and thermal management of high-performance multicores

Bartolini

Cacciari

Tilli

et al. 2011

2011 Design, Automation &Amp; Test in Europe

View full text Add to dashboard Cite

High-end multicore processors are characterized by high power density with significant spatial and temporal variability. This leads to power and temperature hot-spots, which may cause non-uniform ageing and accelerated chip failure. These critical issues can be tackled on-line by closed-loop thermal and reliability management policies. Model predictive controllers (MPC) outperform classic feedback controllers since they are capable of minimizing a cost function while enforcing safe working temperature. Unfortunately basic MPC controllers rely on a-priori knowledge of multicore thermal model and their complexity exponentially grows with the number of controlled cores.In this paper we present a scalable, fully-distributed, energy-aware thermal management solution. The modelpredictive controller complexity is drastically reduced by splitting it in a set of simpler interacting controllers, each allocated to a core in the system. Locally, each node selects the optimal frequency to meet temperature constraints while minimizing the performance penalty and system energy. Global optimality is achieved by letting controllers exchange a limited amount of information at run-time on a neighbourhood basis. We address model uncertainty by supporting learning of the thermal model with a novel distributed self-calibration approach that matches well the controller architecture.

show abstract

Section: A Related Workmentioning

confidence: 99%

Section: Self-calibration Routinementioning

confidence: 99%

A distributed and self-calibrating model-predictive controller for energy and thermal management of high-performance multicores

Bartolini

Cacciari

Tilli

et al. 2011

2011 Design, Automation &Amp; Test in Europe

View full text Add to dashboard Cite

show abstract

“…At the same time, due to the increased complexity and scalability issues of modern many-core architectures, traditional centralized Dynamic Thermal Management (DTM) schemes [8] [15] can no longer be used to control these elevated temperatures. As a result, decentralized DTM schemes [6] have emerged as a new paradigm.…”

Section: Introductionmentioning

confidence: 99%

“…Moreover, due to absence of global communication in a fully decentralized system, there is an essential need for the exchange of state information across regions to achieve an optimal distribution of temperatures across the chip. A key challenge for multi-core system designers is the choice of tuning parameters for this negotiation [15]. Our proposed methodology helps the designers of decentralized DTM schemes to identify the optimal values of these tuning parameters.…”

Section: Introductionmentioning

confidence: 99%

Probabilistic Formal Verification Methodology for Decentralized Thermal Management in On-Chip Systems

Iqtedar

Hasan

Shafique

et al. 2015

2015 IEEE 24th International Conference on Enabling Technologies: Infrastructure for Collaborative Enterprises

View full text Add to dashboard Cite

Just like any other algorithm, Dynamic Thermal Management (DTM) schemes for multi-core architectures are susceptible to errors. Moreover, due to the wide spread usage and safety-critical nature of these schemes, there is a key demand for robust verification of these schemes before deployment. Traditional analysis techniques, like simulation and emulation, are inherently incomplete and therefore they cannot guarantee a complete absence of bugs. In this paper, we present a generic formal verification methodology, based of probabilistic model checking, for verifying decentralized DTM schemes. The paper provides a general modeling approach for developing a Markovian model of any decentralized DTM scheme. Moreover, we identify a set of generic probabilistic properties that can be of an interest to DTM scheme designers. For illustration purposes, the proposed methodology is used to verify Thermal Aware Agent Based Power Economy (TAPE), which is a state-of-the-art decentralized DTM scheme.

show abstract

“…Fig. 2b shows that, if tasks have different starting temperatures to begin with, they would all converge to a steady-state value by means of change in values of N and F. [5] is composed of a truly distributed control mechanism where each core in a multicore processor has its own frequency and voltage scaling controller. The approach presented by Zanini et al [6] presents an online convex optimisation process to control the frequency of cores while predicting workloads.…”

mentioning

confidence: 99%

Cellular nonlinear network controller for dynamic thermal management

Dasu¹,

Chandrasekaran

2012

Electron. Lett.

View full text Add to dashboard Cite

The dynamic thermal management problem is posed for multi-processing element (multi-PE) computing systems (which can support both dynamic re-parallelisation of tasks across PEs and their operating frequencies) as a multi-constraint optimisation problem. Proposed is a cellular nonlinear network controller as an online solver of this problem. Results obtained through a simulation model are presented followed by a brief overview of pertinent literature.Introduction: Dynamic thermal management (DTM) refers to a range of hardware and software strategies which work dynamically to control the temperature distribution on a chip. The important objectives of DTM (itemised by Brooks and Martonosi [1]) are: (a) to provide inexpensive hardware and/or software responses, (b) to reduce power dissipation, and (c) to impact performance as little as possible.Similar to many other DTM techniques, we assume that the computing platform is composed of multiple PEs, the frequency of operation of which can be controlled. Additionally we assume that such systems can support dynamic re-parallelisation of tasks across one or more PEs in either a many core architecture framework [2] or a dynamically and partially reconfigurable field programmable gate array (FPGA) [3].

show abstract

Central vs. distributed dynamic thermal management for multi-core processors

Cited by 19 publications

References 7 publications

A distributed and self-calibrating model-predictive controller for energy and thermal management of high-performance multicores

A distributed and self-calibrating model-predictive controller for energy and thermal management of high-performance multicores

Probabilistic Formal Verification Methodology for Decentralized Thermal Management in On-Chip Systems

Cellular nonlinear network controller for dynamic thermal management

Contact Info

Product

Resources

About