Low-Carbon Routing Algorithms
For Cloud Computing Services
in IP-over-WDM Networks
TD3.3: Energy-efficient service provisioning and content distribution
Contributors: CNIT-PoliMi & FW (presented at ICC 2012)
Speaker: Francesco Musumeci (CNIT-PoliMi)
Summary
n
Introduction
n
Power consumption models
n
Low-carbon routing algorithms
n
Case study
Summary
n
Introduction
q
Scope
q
Renewable energy sources
n
Power consumption models
n
Low-carbon routing algorithms
n
Case study
Scope
n
ICT responsible for 8% of global energy consumption
(2% of total CO
2
emissions)
n
Great benefits in reduction of greenhouse gas
emissions using renewable energy sources (RES)
e.g., sun and wind
q
RES usually in remote locations, hard to connect to electrical
grid à DCs delocalization
n
Scope: dynamic low-CO
2
emissions routing
q
Difference between CO
2
reduction and energy efficiency
Renewable energy sources
§
Solar
powered data centers
§
Power profile during 24 hours of the day
§
Shifted according to the time zone
§
Variations mainly due to diurnal cycle of
the sun àpower from 6 to 22
q
Wind
powered data centers
q
Power profile during 24 hours of the
day
q
Shifted according to the time zone
q
Variations due to the waxing and
Summary
n
Introduction
n
Power consumption models
q
Data center power supply
q
Processing power
n
Low-carbon routing algorithms
n
Case study
Renewable energy production
n
Power needed in each DC to support the whole
traffic request:
q
300 MW for USA24 topology
q
150 MW for ItalyNet topology
n
Solar Energy
generated from CSP (Concentrated
Solar Power) system with parabolic troughs
q
Reference: Nevada Solar One
q
Our case: area of 6 km
2
for USA24 and 3 km
2
for ItalyNet
n
Wind Energy generated from Wind farm, group of
wind turbines joined together
q
Reference: Wild Horse Wind Farm, Washington
q
Our case: 166 turbines (1.8 MW each) covering an area of
Processing power
n
Analysis on energy consumption of Cloud
Computing*, describing 3 possible types of service:
q
Storage as a Service (S
t
aaS)
q
Software as a Service (SaaS)
q
Processing as a Service (PaaS)
n
We estimate the energy for a single connection
during the processing phase
n
Model for the
energy consumption per unity of Gbit
(Wh/Gbit) of a
single user connection:
E
proc=1.5 * T
proc* P
serverData center consumption and design
Server
consumption
Cooling
factor + OH
Processing
Data center consumption and design
n
Model for the
energy consumption per unity of Gbit
(Wh/Gbit) of a
single user connection:
E
proc=1.5 * T
proc* P
servern
We assume a 8.54 Gbyte (DVD) video file conversion from MPEG-2
to MPEG-4 which requires on the computation server (HP DL380
G5) 1.25 hours à 0.0183 hours/Gbit
n
Consider a 10 Gb/s lightpath which lasts t seconds (10*t Gb of data):
1.5 * 10*t (Gbit) * 0.0183 (h/Gbit) * 355 (W) = 97.4*t Wh ~ 100*t Wh
Summary
n
Introduction
n
Power consumption models
n
Low-carbon routing algorithms
q
Anycast routing
q
Energy-aware routing algorithms
n
Case study
Routing problem
0
Dummy
Node
Anycast
link
3
4
7
1
5
6
2
CO2Is it better to route
towards a remote
DC with renewable
Anycast Routing
Problem
Trade-off between transport
and processing power
DC
1DC
2Algorithms
n
Routing algorithms: aim to minimize the total
non-renewable (brown) energy of the network (emissions
à 1 kwh=228 gCO2)
n
Sun&Wind Energy-Aware Routing
(SWEAR):
chooses for each connection between two paths
q
1 - Maximum usage of renewable energy
q
2 - Lowest transport energy
n
Green Energy-Aware Routing
(GEAR): finds for
each connection the path with
q
Lowest non-renewable (brown) energy
n
Consider trasport power (brown)
q
Trade-off between transport power and (green) processing
SWEAR - A simple example
3
4
7
1
5
6
2
CO2Routing towards
Green Sources
DC
1DC
2DC
3l
sl
gP
lg-P
ls< P
p?
The excess in
transport power is
compensated by
usage of RES?
YESàchoose l
gSWEAR - A simple example
3
4
7
1
5
6
2
CO2Load Balancing when the
traffic on a link overcome
the fixed threshold
(higher cost for that link)
DC
1DC
2DC
3l
sl
gP
lg-P
ls< P
p?
The excess in
transport power is
compensated by
usage of RES?
NOàchoose l
sGEAR - A simple example
3
4
7
1
5
6
2
CO2100
Transport links weight equal
to transport power
Anycast links weight equal to
brown processing power
Homogeneous weights
assignment (brown power)
for all network links
DC
1DC
2DC
3l
2l
1l
3Summary
n
Introduction
n
Power consumption models
n
Low-carbon routing algorithms
n
Case study
Case Study
§
Poisson interarrival traffic
§
Holding time with negative
exponential distribution
•
Average duration 1 s
§
Each connection requires an entire
10 Gbit/s lighpath
§
Traffic uniformly distributed
§
24 nodes, 43 links
§
16 wavelenghts/link
§
6 Data centers:
•
1 with
Wind
energy (1)
•
2 with Solar energy (14,15)
Case Study
§
21 nodes, 36 links
§
16 wavelenghts/link
§
6 Data centers:
•
1 with
Wind
energy (15)
Case Study
n
GEAR and SWEAR compared with two reference
algorithms:
q
Shortest Path
(SP)
q
