1

GreenDisc: A HW/SW Energy Optimization Framework in Globally

Distributed Computation

Marina Zapater†, José L. Ayala*, José M. Moya‡

†CEI Campus Moncloa, UCM-UPM*Universidad Complutense de Madrid

‡ Universidad Politécnica de Madrid

Marina Zapater | UCAMI 2012

Marina Zapater | UCAMI 2012 2

Outline

• Motivation• Proposed solution– The GreenDisc Platform

• Holistic Optimization Approach– Power optimization of all implied agents

• Conclusions • Future Work

Motivation

• The key to next-generation e-Health solutions are wearable personal health systems– Biomedical monitoring (European initiatives)

• World-wide sensor deployment:– Very large amount of data

• Electrocardiogram sensor (ECG), Electromyogram (EMG), O2 and CO2, temperature, lactate, movement sensors.

– Raw data must be turned into useful information

Marina Zapater | UCAMI 2012 3

FET “Guardian Angels”. European initiative: FET Flagships 2013 - http://www.ga-project.eu/home

Motivation

• Need for an accurate, integrated and long-term assessment and feedback.– Acquire, monitor and analize data 24/7

• Need for an application-specific architecture – But also provide flexibility and enough performance

• Current issues: energy and heat– High energy consumption and short battery lifespan of sensor

nodes. • Heat produced by sensor nodes for biomedical applications

– Tackle the computation needs to obtain useful information from all data• Energy consumption at the Data Center level must not put at stake

e-Health deployments

Marina Zapater | UCAMI 2012 4

Current needs and issues

Contributions

• A HW/SW energy optimization framework– Versatile platform that integrates processing, analysis and

wireless communication of biomedical data.– Both at the WSN-level and at the Data Center level

• Supports e-Health applications (and future evolutions) with lower costs and shorter time-to-market

Marina Zapater | UCAMI 2012 5

GreenDisc Platform

GreenDisc Platform

• Based on Wireless Body Sensor Networks (WBSN)– All sensors transmit to a PDA

Marina Zapater | UCAMI 2012 6

Context and System Architecture

GreenDisc Platform

• Based on Wireless Body Sensor Networks (WBSN)– All sensors transmit to a PDA

• Computation takes place at Data Centers– Data storage and processing

Marina Zapater | UCAMI 2012 7

Context and System Architecture

GreenDisc Platform

• Power optimization in the processing nodes of the WBSN– Design of embedded processors for signal processing– Optimization at the radio interface– Design automation of applications for the processing node

• Power optimization in data centers

Marina Zapater | UCAMI 2012 8

Holistic Optimization Approach

Holistic Optimization

• Design of embedded processors for signal processing– Architectural modifications considering the application

mapping, the execution profile, and the compiler optimizations

– Reducing the energy consumption of the main energy consumption sources• Selection of instruction memory architecture• Design of functional units with tunable architecture (dynamic

reconfiguration)

Marina Zapater | UCAMI 2012 9

Processing nodes of the WBSN (I)

Holistic Optimization

• Instruction memory produces highest energy consumption

• Proper selection of instruction memory architecture impacts energy consumption– SPM - scratch pad memory – CELB - central loop

buffer– CLLB- clustered loop buffer

Marina Zapater | UCAMI 2012 10

Processing nodes of the WBSN (I)

Holistic Optimization

• Power optimization in the radio interface– Reducing the amount of information to transmit:

• Framework for signal analysis to develop compressed sensing techniques for several bio-signals

• Case studies for monitoring bio-signals with a QoS study– Reduce the overhead of the transmision protocol:

• Study of the impact of tuning several parameters of the MAC layer (development of 802.15.4 MAC analytical model)

Marina Zapater | UCAMI 2012 11

Processing nodes of the WBSN (II)

Compressed sensing: signal processing technique for efficiently acquiring and reconstructing a signal. Uses the signal sparseness or compressibility in some domain, allowing the entire signal to be determined from relatively few measurements.Candes, E. J. and Wakin, M. B. (2008) An Introduction to Compressive Sampling.IEEE Signal Processing Magazine. Vol 2 (pp. 21-30)

Holistic Optimization

• Design automation of applications in the processing node– Aims to reduce the amount of data transmitted to the

backbone– Providing a generic high-level model of the architecture

• Analysis of the impact of design parameters in power consumption

• Framework for automatic design and optimization of applications

Marina Zapater | UCAMI 2012 12

Processing nodes of the WBSN (III)

Holistic Optimization

• Implementation of several resource managing techniques at different abstraction levels

• Exploiting the heterogeneity of applications and computing resources for energy minimization.– Proper usage of heterogeneity can lead to significant

energy savings• Analysis of cooling mechanisms and development of

control techniques

Marina Zapater | UCAMI 2012 13

Energy Optimization in Data Centers

Holistic Optimization

• Potential benefits of workload characterization and dynamic assignment policies

Marina Zapater | UCAMI 2012 14

Energy Optimization in Data Centers

Conclusion andFuture Work

• This paper shows how the GreenDisc platform can optimize the energy consumption of next-generation e-Health application by combining the usage of:– Optimization policies at several levels of the WBSN– Agressive energy efficiency policies at the Data Center

• Future work will integrate all steps of the platform and show the overall savings for a particular e-Health workload.

Marina Zapater | UCAMI 2012 15

Marina Zapater | UCAMI 2012 16

Questions?

Thank you for your attention

Marina ZapaterLaboratorio de Sistemas Integrados (LSI)

Universidad Politécnica de Madridmarina@die.upm.es

http://greenlsi.die.upm.es