Green Wireless Technology Panel Presentation
Professor Sandeep K. S. Gupta
p
p
IMPACT
(Intelligent Mobile Pervasive Autonomic Computing & Technologies) LAB(
http://impact.asu.edu
)
School of Computing, Informatics, Decision Systems Engineering
Arizona State University
Arizona State University
[email protected]
Sandeep Gupta, IEEE Senior Member
• Heads
Use-inspired, Human-centric research in distributed cyber-physical systems
@
School of Computing & Informatics
Use inspired, Human centric research in distributed cyber physical systems
Pervasive Health Monitoring
Thermal Management for Data Centers
Criticality Aware-Systems ID Assurance Intelligent Container Mobile Ad-hoc Networks
BEST PAPER AWARD: Security Solutions for Pervasive HealthCare – ICISIP 2006
BOOK: Fundamentals of Mobile and Pervasive Computing, Publisher: McGraw-Hill Dec. 2004
HealthCare ICISIP 2006.
• TCP Chair
•TCP Co-Chair:
GreenCom’07
• Area Editor
Also for IEEE TPDS WINETEmail:
[email protected]
;
IMPACT Lab URL: http://impact.asu.edu;
Two
extreme examples of
Greenness
1. Embedded networks, e.g. Body Area Networks (BANs), consisting
of embedded and inter‐communicating sensor devices
EKG EEG BPg
– Constrained resources • Principally battery powered with limited available energy – Sustainabilityis the main issue • Un‐interrupted operations, e.g. sensing & communication, for long periods Required Approaches SpO2 BP Base Station – Required Approaches • Energy‐efficient algorithm design • Scavenging energy from the environment (e.g. human body) Motion Sensor Body Area Network (BAN)2.
Large‐scale networked systems, e.g. data centers, integrating a
large number of computing resources for service provisioning
– Power density increases ( ) • Circuit density increases by a factor of 3 every 2 years • Energy efficiency increases by a factor of 2 every 2 years • Effective power density increases by a factor of 1.5 every 2 years [Keneth Brill: The Invisible Crisis in the Data Center] – Total Cost of Ownership (TCO)rising• Data Center TCO doubles every three yearsData Center TCO doubles every three years
• Cooling the data center can cost up to half of the total electricity bill • Power Usage Efficiency has to be reduced [Uptime Institute] – Required Approaches • Energy‐efficient thermal‐aware resource management C di t d t f ti d li • Coordinated management of computing and cooling resources Data Center
Green Data Centers
• System Requirements
– Equipment Safety
• Equipment operating temperature should be within a manufacturer‐specified redline temperature– Service Level Agreements (SLAs)
• The throughput and turn‐around time should Typical Data Center Design g p meet the user requirements• Energy Issues
– Computing Energy: Trade‐off with meeting the SLAs Cooling Energ Trade off ith eq ipment safet – Cooling Energy: Trade‐off with equipment safety • Stems from the data center thermal‐issues• Thermal‐ Issues
–Heat recirculation
Green data centers call for
Coordinated
Thermal‐aware resource management
to reduce cooling and computing
Heat recirculation
• Hot air from the equipment air outlets is fed back to the equipment air inlets –Hot spots
Eff t f H t R i l tienergy consumption while meeting the
equipment safety and SLAs
• Effect of Heat Recirculation • Areas in the data center with alarmingly high temperature –Consequence
• Cooling has to be set very low to have all inlet temperatures within the redline for safetyEcosystem of Datacenters
Ecosystem of Datacenters
Different task assignments lead to different power
consumption distributions
consumption distributions
Different power consumption distributions lead to
different temperature distributions
different temperature distributions
Different temperature distributions lead to different total
energy costs
80 100 3000 3500 100gy
1 2 3 4 5 0 20 40 60 1 2 S3 0 500 1000 1500 2000 2500 1 2 S2 S3 0 20 40 60 80 5 6 7 8 9 10 11 12 S1 S3 S5 2 3 4 5 S1 S2 2 3 4 S1Server load
Power consumption
Temperature
7
Server load
Coordinated Resource Management for Green Data Centers
Energy reduction in data centers has three
Energy‐reduction in data centers has three
directions
1. Thermal‐aware workload management • Schedule and place jobs such that low power servers having less impact on the cooling demand are used 2. Server Power Management • Operate under‐utilized servers at low power states (e.g. turn‐off idle servers) 3. Cooling Management• Dynamically set the cooling to the highest temperaturesDynamically set the cooling to the highest temperatures that meets the equipment safety
There is a spatio‐temporal workload schedule that minimizes the total energy (cooling + computing) demand. Find it and perform dynamic cooling
management and server power management with it to
Energy Savings Over Current Practice
minimize the total energy consumption while meeting the SLAs and equipment safety!
Online Data Center Thermal Management
Data Center-Level Thermal Models
To enable on-line real-time thermal-aware job schedulingPower Characterization
Characterize the power consumption of
a given workload (CPU, memory, disk
• fast (analytical, non CFD based) • non-evasive (machine-learning)
g
(
,
y,
etc) on a given server machine
Model the thermal impact of
Thermal Management Infrastructure
multicore systems Sensor Data Gathering Service Data Center Monitoring Performance Monitoring Service & Services for Data Centers
Thermal-aware
Job Scheduling
On-line job scheduling algorithm to minimize peak air inlet
Non-Invasive Thermal Evaluation Fast Thermal Evaluation Service Thermal/Power & Performance Correlation Service
Resource & to minimize peak air inlet
temperature, thus minimizing the cost of cooling. Policy Enforcement Thermal Management Policy Enforcement Service Thermal Control Policies Resource & Server Management OS-Level Services Performance Monitoring
Service Control Policies
PI: Sandeep K. S. Gupta [email protected]
Job Scheduling Service
Cluster
Management Resource Job Queues
Queues Cooling Control Service Air-flow Control Service Facility Management
http://impact.asu.edu
Green
Body Area Networks (BANs)
• System Requirements
– Sustainability
• Uninterrupted BAN operations for long periods
EEG Sensors Uninterrupted BAN operations for long periods essential for medical care
– Operational Safety
• Side‐effects of BAN operations (e.g. heat SpO2 EKG BP Base S i Base Station dissipation) should be within a threshold– Security
• Sensitive medical data collection should be secure as per legal requirements (HIPAA)Station Motion Typical BAN Design secure as per legal requirements (HIPAA) • Secure wireless inter‐sensor communication is essential
• Energy Issues
SensorGreen BANs call for
energy analysis of
the computing and communication
ti
d
il bilit
f
gy
– Computing and Communication Energy: Trade‐off with security and sustainability • Limited battery powerS i f th h b d i ti l
operations
and
availability of energy
for potential scavenging
to ensure
sustainability and security.
– Scavenging energy from the human body is essential • Security adds computation and communication overhead – Energy analysis of security primitives important – Reduced power consumption in the sensors can enable safety by reducing potential heat dissipationSustainable Physiological Value‐based Security for BANs
h i l i l i
l b
d
Ph i l i l i l Ph i l i l i l• Physiological signal based Key
Agreement (PKA)
– Perform key agreement by combining
f f Physiological signal Physiological signal features key features ky g
y
g
cryptographic primitives with signal
processing
Sender Receiverhide Un‐hide key PKA P P fil (R di ON V l Si 5000) 0 04 0.05 0.06
PKA Power Profile (Radio-ON, Vault Size = 5000)
Receiver(Radio ON) 0.02 0.03 0.04 Sender(Radio ON) Scavenging Technique Source Power Gain
• The max power required by PKA
1 2 3 4 5 6 7 8 9
0.01
Sensing FFT Peak + Quant
Poly Gen +
Eval Add Chaff Vault Tx/Rx
Lagrangian Interpolation Ackn Tx/Rx q Body Heat Latent heat of vaporization of perspiration 200mW – 320mW
Respiration Chest Expansion while ~420mW
The max power required by PKA
(58 mW) low enough to be
sustained by prominent energy
scavenging techniques
Respiration Chest Expansion while breathing420mW Ambulation Arm & Leg Movement 1.W‐1.6W Photovoltaic Cells Photovoltaic Cells 100mW/cm2
scavenging techniques
Venkatasubramanian et al Green and Sustainable Security Solutions for BANs, BSN’08
Safety: Minimize Tissue Heating
Safety: Minimize Tissue Heating
•
Medical sensors implanted/worn by
Medical sensors implanted/worn by
human need to be safe.
•
Sensor activity causes heating in the
tissue.
H
i
d b RF i d
i
– Heating caused by RF inductive
powering
– Radiation from wireless
communication
Tissue Blow-up– Power dissipation of circuitry
•
Goal:
minimize tissue heating.
•
Two solutions:
Communication scheduling
for
– Communication scheduling
for
minimizing thermal effects:
• Rotate cluster leader – balance energy usage + distribute heat dissipation H ti dissipation– Thermal aware routing:
route
around thermal hotspots
Cluster leader Heating ZoneBSN Scheduling
Medium 1(free space) ε1,µ1, σ1Medium 2(Body tissue) ε2,µ2, σ2
g
Requirement
SAR = σ E
2/ ρ (W/kg)
E = induced Electric Field
Incident Plane Wave with power P0
Transmitted Wave Cluster Leader RF Powering Source depth d
System Model
• Consider only one cluster • 2D Model• FCC
Regulation
IEEE Requirement (1g Tissue)
E induced Electric Field Ρ = tissue density
σ = electric conductivity of tissue Reflected Wave Control Volume and
a cluster of biosensors
• 2D Model • Rotate cluster
head - dist.
energy consump.
d h ti Temperature Rise: Pennes
Heat by metabolism SAR = 0.4W/Kg Whole Body Average SAR = 8W/Kg
Peak Local reduce heating Temperature Rise: Pennes
Bio-heat Equation
Heat
accumulated bHeat transfer d ti
Heat by
radiation Heat transferb ti
Heat by power dissipation SAR = .08W/Kg Whole Body Average SAR = 1.6W/Kg Peak Local CE UCE metabolism accumulated by conduction radiation by convection dissipation
Solution
• Random selection may lead to higher
1 2 5 1 2 5 1 2 5 Results FDTD + enumeration FDTD + Genetic Algorithm Optimal Near Optimal 720960 hrs (est.) 100 hrs (est.) temperature rise • Similar to Traveling
salesman problem but with dynamic metric
3
4 4 3 4 3
(a) Ideal Rotation (b) Nearest Rotation (c) Farthest Rotation
Four Approaches
• FDTD + enumeration Ge e c go TSP + Genetic Algorithm TSP +enumeration Optimal Near Optimal Near Optimal 100 hrs (est.) 7.6 hrs 5 min ureTemp rise in sensor surroundings 0.11 Temperature Comparative Result • Heuristic: Leader selection based on sensor location, rotation history • FDTD + Genetic Algorithm • TSP + enumeration • TSP +Genetic T emperat u surroundings 0.07 0.08 0.09 0.1 Temperature Rise ° C Temperature Mean ± Deviation Mean TSP +Genetic Algorithm
Q. Tang, N. Tummala, S. K. S. Gupta, and L. Schwiebert, Communication scheduling to minimize
thermal effects of implanted biosensor networks in homogeneous tissue, Proc of IEEE
Transactions of Biomedical Engineering
Worst Dynamic Manual Genetic Optimal 0.06