Distributed Load Management e-Energy 2010 Kiril Schrder C. v. O. - - PowerPoint PPT Presentation

▶

Jun 15, 2023 207 likes •304 views

Power and Cost Aware Distributed Load Management e-Energy 2010 Kiril Schrder C. v. O. University of Oldenburg schroeder@informatik.uni-oldenburg.de Daniel Schlitt Marko Hoyer Wolfgang Nebel OFFIS Institute for Information Technology

SLIDE 1

e-Energy 2010

Kiril Schröder

C. v. O. University of Oldenburg

schroeder@informatik.uni-oldenburg.de

Daniel Schlitt

OFFIS – Institute for Information Technology daniel.schlitt@offis.de

Marko Hoyer

OFFIS – Institute for Information Technology marko.hoyer@offis.de

Wolfgang Nebel

C. v. O. University of Oldenburg

nebel@informatik.uni-oldenburg.de

Power and Cost Aware Distributed Load Management

SLIDE 2

High energy consumption in data centers
Rising performance, slower rising energy efficiency of servers
Forecast 2020, Greenpeace: 623 TWh (DCs+Network)
Saving energy through better capacity utilization
Server consolidation using virtualization
Higher efficiency at runtime with dynamic load management
Through live migration
Now: reactive,

Better: proactive using load prediction

14.04.2010 2 Power and Cost Aware Distributed Load Management

Motivation 1/2

SLIDE 3

Enterprises maintaining several data centers
Different data center efficiency and location characteristics
Operation
Temperature, humidity
Energy price
Energy supply
Sun and wind power
Application
Data center network
Marketplace of computing power

14.04.2010 3 Power and Cost Aware Distributed Load Management

Motivation 2/2

7°C

+5°C

SLIDE 4

Data Center (Server, Cooling, UPS etc.)
Database Replication
Network
Live Migration (migration costs)
Problems
Data Security and Privacy,
Control,
Responsibility etc.
like in Cloud Computing

14.04.2010 4 Power and Cost Aware Distributed Load Management

Vision of Distributed Load Management 1/3

SLIDE 5

Simple example for migrations:

Constant work load
Three data centers with

different energy prices

Lowest prices rotating

Reality, much more difficult:

Varying loads
Unsure forecasts
Different migration costs
Need dynamic (partial) replication
Good locations can change

14.04.2010 5 Power and Cost Aware Distributed Load Management

Vision of Distributed Load Management 2/3

SLIDE 6

Related Work

IBM: Utilize Sun Power
Asfandyar Qureshi et al., MIT: “Cutting the Electric Bill for Internet-Scale

Systems”

Query Distribution
Full Replication
Traces for
Electricity Prices
Load & Traffic

14.04.2010 6 Power and Cost Aware Distributed Load Management

Vision of Distributed Load Management 3/3

SLIDE 7

Simulation Setup
Simple data center model
(Semi-)Homogeneous environment
Typical parameter values in Germany
Savings
Energy (~Temperature): less than 5 %
Present energy prices: up to 10 %
Dynamic energy prices: up to 40 %
Electricity Regulation
Reducing peak demand
Enable even demand (50 %)

 Increasing power station efficiency

14.04.2010 7 Power and Cost Aware Distributed Load Management

Potential Analyses

0 % 50 % 100% Pstandby Pmin Pmax

standby active server work load power consumption load dependent load independent cooling unit

humidifier hot air cold air free air cooling chiller

cold accumulator/ free air cooling control always active active in full climate control and hybrid mode

free air cooling

SLIDE 8

Detailing simulation setup
Modeling:
Data Center Abstraction (in progress)
Network (in planning)
Load Prediction (proceeded)
Load Management (in progress)
Adaptive Replication (in progress)
Embedding in Smart Grids (in planning)

14.04.2010 8 Power and Cost Aware Distributed Load Management

Ongoing and required future work

renewable energy with weather-dependent feed-in energy supply with time varying peak loads data center as an adaptive regulating bulk consumer

utsourcing
f services

internet access video on demand cloud computing S ME s private households cooperating data centers regional data center energy provider dynamic electricity pricing energy demand and price aware load management

SLIDE 9

Data center comprehensive load management
f virtual machines
High saving potentials:

Costs > Energy

Several application areas
Energy and cost aware cloud service
Large enterprises, network of cooperating enterprises
Marketplace of performance (open cloud)
Embedding in smart power grids
Demand regulation
Increasing efficiency
Many challenges
Technical: assurance of service level agreements, network

coordination …

Political: (legal) obligation, different national laws …

14.04.2010 9 Power and Cost Aware Distributed Load Management