greencloud:
TRANSCRIPT
GreenCloud:A Packet-level Simulator of
Energy-aware Cloud Computing Data Centers
Dzmitry Kliazovich
ERCIM FellowUniversity of Luxembourg
Apr 16, 2010
Outline Data center architectures
Two-tier, three-tier, and three-tier high-speed
Structure of data center simulator Energy efficiency, simulator components
Case study data center simulations
April 16, 2010 2Dzmitry Kliazovich ([email protected])
Why energy is important? Increased computing demand
Data centers are rapidly growing Consume 10 to 100 times more energy per square foot than a
typical office building
Energy cost dynamics Energy accounts for 10% of data center operational expenses
(OPEX) and can rise to 50% in the next few years Accompanying cooling system costs $2-$5 million per year
April 16, 2010 3Dzmitry Kliazovich ([email protected])
Distribution of data center energy consumption
April 16, 2010 4Dzmitry Kliazovich ([email protected])
Data center architectures Two-tier data center architecture
Access and Core layers 1 GE and 10 GE links Full mesh core network Load balancing using ICMP
April 16, 2010 5Dzmitry Kliazovich ([email protected])
Data center architectures Three-tier data center architecture
Access, Aggregation, and Core layers Scales to over 10,000 servers 8-way ECMP load balancing
April 16, 2010
S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S
CoreNetwork
AggregationNetwork
AccessNetwork
S
Links10 GE 1 GE
Nodes
L3 Switch L2/L3 Rack Switch Computing Server
6Dzmitry Kliazovich ([email protected])
Data center architectures Three-tier High-Speed data center architecture
Increased core network bandwidth 2-way ECMP load balancing 100 GE standard (IEEE 802.3ba) still in works since Nov 2007
April 16, 2010
S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S
CoreNetwork
AggregationNetwork
AccessNetwork
S
Links
10 GE 1 GE
Nodes
L3 Switch L2/L3 Rack Switch Computing Server100 GE
7Dzmitry Kliazovich ([email protected])
Data center simulator Greencloud is an extension of NS-2 network
simulator for energy-aware cloud computing simulations
Provides packet-level simulation dynamics
Focused on workload distribution strategies and energy consumption models of simulator components (servers, switches, links, etc.)
Dzmitry Kliazovich ([email protected]) 8April 16, 2010
Data center simulator
Dzmitry Kliazovich ([email protected]) 9April 16, 2010
CoreNetwork
AggregationNetwork
AccessNetwork
Computing Server
1 RU Rack Switch
L3 Switch
TaskSchedulerTask
Scheduler
Links10 GE 1 GE
WorkloadGenerator
CloudUserCloud
User
Data Center
Data CenterCharacteristics
TaskScheduler
TaskComAgent
L3 Energymodel
L2 Energymodel
ServerCharacteristics
TaskComSink
Connect ()
SchedulerServerEnergy model
S S S S S S S S S
WorkloadTrace File
Simulator components Servers
Responsible for task execution Single-core nodes Preset processing limit in MIPS or FLOPS
Supported power management modes DVFS: Dynamic Voltage/Frequency Scaling DNS: Dynamic Shutdown Both: DNS if server is idle, DVFS otherwise
Dzmitry Kliazovich ([email protected]) 10April 16, 2010
Simulator components Servers’ Energy Model
Dzmitry Kliazovich ([email protected]) 11April 16, 2010
Fmin
Fmax
0
0.2
0.4
0.6
0.8
1
Pfixed
Ppeak
CPU FrequencyServer load
Pow
er c
onsu
mpt
ion
CPUmemory modules, disks, I/O resources
Idle server consumes about 66% of the peak load for all CPU frequencies
Simulator components Switches
Most common Top-of-Rack (ToR) switches typically operate at Layer-2 interconnecting gigabit links in the access network
Aggregation and core networks host Layer-3 switches operating at 10 GE (or 100 GE)
Links Transceivers’ power consumption depends on the quality of signal
transmission in cables and is proportional to their cost 1 GE links: 0.4W is consumed for 100 meter transmissions over twisted
pair 10 GE links: 1W is consumed for 300 meter transmission over optical fiber
Supported power management modes DVFS, DNS, or both
Dzmitry Kliazovich ([email protected]) 12April 16, 2010
Simulator components Switches’ Energy Model
Dzmitry Kliazovich ([email protected]) 13April 16, 2010
Chassis~ 36%
Linecards~ 53%
Port transceivers~ 11%
Simulator components Workloads
Model cloud user applications (social networking, instant messaging, content distribution, etc.)
Workload properties Computational: MIPS, duration Communicational: workload size, its internal and
external transfers
Generation Trace-driven Using random distribution (Exp, Pareto, etc.)
Dzmitry Kliazovich ([email protected]) 14April 16, 2010
Simulation Setup Data center architectures
Two-tier (2T), three-tier (3T), and three-tier high-speed (3Ths)
Simulation parameters Average data center load is 30% 1536 computing servers 1, 10, and 100 GE links with 10 ns delay 4500 bytes workloads (3 Ethernet packets) 60 minutes of simulation time
Dzmitry Kliazovich ([email protected]) 15April 16, 2010
Simulation Setup Energy-aware “green” scheduler
Dzmitry Kliazovich ([email protected]) 16April 16, 2010
0 200 400 600 800 1000 1200 1400 1600
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Server #
Ser
ver l
oad
Servers at the peak load
Under-loaded servers,DVFS can be applied
Idle servers,DNS can be applied
Evaluation results Distribution of energy consumption in data center
Dzmitry Kliazovich ([email protected]) 17April 16, 2010
Servers355kW·h (82%)
Core switches0.87kW·h (0.2%)
Aggregation switches1.74kW·h (0.4%)
Access switches75.6kW·h (17.4%)
Data center433kW·h
Chassis36%
Linecards53%
Port tranceivers
11%
Switches
CPU130W (43%)
Memory36W (12%)
Disks12W (4%)
Peripherial50W (17%)
Motherboard
25W (8%)
Other48W (16%)
Computing Servers301 W
Evaluation results Comparison of energy-efficiency schemes
Dzmitry Kliazovich ([email protected]) 18April 16, 2010
Conclusions Energy consumption is becoming a concern in cloud computing
data centers
Developed a packet-level simulator for energy-aware data centers
Obtained results compare the performance of dynamic voltage/frequency scaling (DVFS) and dynamic server/network shutdown (DNS) schemes
Future work will focus on adding storage area network as well as on the development of novel workload consolidation and traffic aggregation techniques
Dzmitry Kliazovich ([email protected]) 19April 16, 2010