GreenCloud: A Packet-level Simulator of Energy-aware Cloud Computing Data Centers Dzmitry Kliazovich ERCIM Fellow University 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, Dzmitry Kliazovich

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, Dzmitry Kliazovich

Distribution of data center energy consumption April 16, Dzmitry Kliazovich

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, Dzmitry Kliazovich

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, Dzmitry Kliazovich

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, Dzmitry Kliazovich

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 8 April 16, 2010

Data center simulator Dzmitry Kliazovich 9 April 16, 2010

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 10 April 16, 2010

Simulator components Servers’ Energy Model Dzmitry Kliazovich 11 April 16, 2010 CPU memory 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 12 April 16, 2010

Simulator components Switches’ Energy Model Dzmitry Kliazovich 13 April 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 14 April 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 15 April 16, 2010

Simulation Setup Energy-aware “green” scheduler Dzmitry Kliazovich 16 April 16, 2010 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 17 April 16, 2010

Evaluation results Comparison of energy-efficiency schemes Dzmitry Kliazovich 18 April 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 19 April 16, 2010

Thank you! April 16, Dzmitry Kliazovich