Minimizing Electricity Cost: Optimization of Distributed Internet Data Centers in a Multi-Electricity-Market Environment

胡秩瑋

Overview

INTRODUCTION RELATED WORK FORMULATION AND MODELING SOLUTION METHOD DESIGN ELECTRICITY PRICE AT CERTAIN

LOCATIONS FOR GOOGLE INTERNET DATA CENTERS

PERFORMANCE EVALUATION CONCLUSION AND FUTURE WORK

INTRODUCTION

As the developing of cloud computing, there are more and more Internet Data Centers(IDCs).

IDC demands a lot of power, it means that it costs much money.

The goal of this paper is to minimize the cost of IDCs while guaranteeing quality of service.

INTRODUCTION

Authors achieve the goal with linear programming and polynomial-time method.

The evaluations are based on real-life electricity price data.

FORMULATION AND MODELINGTotal Electricity Cost Modeling for IDC

Two assumption:1. The power consumption profile is

constant for every server.2. Each server at the same location

receives the same traffic rate in steady-state.

FORMULATION AND MODELINGTotal Electricity Cost Modeling for IDC

Thus, the total electricity cost for N data center location is:

i:IDC location :the number of turned on servers at i

:Spot price of electricity at i :power consumption for one server at i

FORMULATION AND MODELINGWorkload Constraint Modeling

We model the workload constraints in order to capture the requests allocations among all the data center locations for each front-end Web portal.

We have:

:the requests demand at each front-end Web portal server j (j=1,…,C)

C :total number of front-end Web portal servers :the request arrival rate from front-end Web portal

server j to location i

FORMULATION AND MODELINGDelay Constraint Modeling

At location i with servers, when each server has the service rate and the total arrival rate is , the average delay is given as

= : service rate

FORMULATION AND MODELINGFormulation of Total Electricity Cost Minimization Problem

The objective function(problem one):

SOLUTION METHOD DESIGN

We are going to present a solution method for problem one in this section.

We approximate the problem by linear programming and solve it with polynomial-time algorithm.

SOLUTION METHOD DESIGNMixed Integer Linear Programming Formulation

First of all, we rewrite the problem one as:

subject to:

.

SOLUTION METHOD DESIGNMixed Integer Linear Programming Formulation

As must be integer, we can transform Problem1 to Problem2:

Subject to , = ,

SOLUTION METHOD DESIGNPolynomial-Time Solution for Approximated Total Electricity Cost Minimization Problem

In this subsection, we show that Problem Two can be converted to a minimum cost flow problem.

we consider a simple case that N =3 and C = 5.

SOLUTION METHOD DESIGNPolynomial-Time Solution for Approximated Total Electricity Cost Minimization Problem

Subject toλ11 + λ12 + λ13 = L1,λ21 + λ22 + λ23 = L2, λ31 + λ32 + λ33 = L3, λ41 + λ42 + λ43 = L4, λ51 + λ52 + λ53 = L5, λ11 + λ21 + λ31 + λ41 + λ51 λ12 + λ22 + λ32 + λ42 + λ52 ≤ λ13 + λ23 + λ33 + λ43 + λ53 ≤

Total workload from 5 web serversMaximum workload that all locations can afford

ELECTRICITY PRICE AT CERTAIN LOCATIONS FOR GOOGLE INTERNET DATA CENTERS

California : hour-ahead marketTexas : 15-min ahead marketAtlanta : fixed electricity rate

ELECTRICITY PRICE AT CERTAIN LOCATIONS FOR GOOGLE INTERNET DATA CENTERS

PERFORMANCE EVALUATION

PERFORMANCE EVALUATION

PERFORMANCE EVALUATION

PERFORMANCE EVALUATION

Average electricity cost of request processing: :

Average electricity cost of upper limitation number of servers: :

CONCLUSION AND FUTURE WORK