Optimizing expansion strategies for ultrascale cloud computing data centers

https://doi.org/10.1016/j.simpat.2015.03.002Get rights and content

Abstract

With the increasing popularity gained by cloud computing systems over the past few years, cloud providers have built several ultrascale data centers at a variety of geographical locations, each including hundreds of thousands of computing servers. Since cloud providers are facing rapidly increasing traffic loads, they must have proper expansion strategies for their ultrascale data centers. The decision of expanding the capacities of existing data centers or building new ones over a certain period requires considering many factors, such as high power consumption, availability of resources, prices (of power, land, etc.), carbon tax, free cooling options, and availability of local renewable power generation. While a rich volume of recent research works focused on reducing the operational cost (OPEX) of the data centers, there exists no prior work, to the best of our knowledge, on investigating the trade-off between minimizing the OPEX of the data centers and maximizing their revenue from the services they offer while respecting the service level agreement (SLA) with their customers. In this study, we model this optimization problem using mixed integer-linear programming. Our proposed model is unique compared to the published works in many aspects such as its ability to handle realistic scenarios in which both data centers’ resources (servers) and user generated traffic are heterogeneous. To evaluate the proposed model and the impact of different parameters on it’s performance, several simulation experiments are conducted.

Introduction

One of the main concepts related to cloud computing is the migration of computations from the user-side to the Internet. With the cloud computing paradigm, companies no longer need to establish and run their own servers to provide on-line services to their customers. Instead, they can simply “rent” the required infrastructure from a specialized cloud provider under a pay-per-use model reducing the Total Cost of Ownership (TCO) and allowing the companies to focus on their own businesses especially in the case of startup companies. Such an option is becoming more appealing for an increasing number of companies, which creates more demand on cloud providers forcing them to optimize their expansion strategies. These expansion strategies should take into consideration both the quality of the service provided to the customers and the economical impact on the service provider [1], [2].

Cloud providers may own several data centers distributed across different locations to serve their clients. Such data centers are usually huge containing tens of thousands of servers and consuming more power than a medium-size town.1 Even with these huge data centers, a cloud provider might still be unable to provide a high quality of service (i.e., one where the service-level agreement (SLA) with the client is not violated) due to the high demand. Thus, expansion strategies must be devised. The cost of expanding a data center or building a new one can vary greatly depending on the land cost and the required computing capacity. In this paper, we address the problem of deciding the best expansion strategy for a given cloud provider by deciding whether it is beneficial for the cloud provider to build new data centers or to simply expand the data centers it currently has. To solve this problem, one needs to address several issues such as where to build the new data centers and in which capacities and how to distribute the current and future traffic loads among the new and existing data centers.

Data centers are a crucial part for governmental institutions, businesses, industries, and many others. They vary greatly in size from small in-house data centers to large scale data centers that provide their services publicly for millions of users. Data centers of one service provider may be distributed over a large geographical area which requires an extra overhead for managing them efficiently. Moreover, they consume large amounts of power that can reach up to tens of megawatts for running their hardware and cooling them. These facts are creating many problems on both the environment and energy resources. A 2010 study showed that large-scale data centers consumed about 2% of all electricity usage in the United States [3]. This percentage can be translated to be over 100 billion kW h with an approximate cost of $7.4 billions [4]. Power usage in data centers is divided into the power consumed by the IT components and the power consumed by non-IT components such as ventilation and cooling systems, and lighting.

Being environmentally responsible is definitely a concern in the cloud computing society. Researchers from both Academia and the industry are collaborating to address environment grand challenges and to accelerate the research in this field [5]. Managing carbon footprint and power consumption [6] are examples of such efforts. From a monetary perspective, the increasing prices of power offer more reasons to reduce the power consumption of data center and to increase the usage efficiency of the available power. The new laws for carbon tax are also pushing forward the optimization of power usage. The adoption of renewable energy usage to cover data centers power requirements is showing a momentum between data centers owners. Also, building data centers in locations that provide free air cooling is a good choice for data centers owners (e.g., Facebook data center in Prineville, Oregon). Moreover, management overhead of today’s data centers requires a lot of manpower to handle the extended traffic loads. The shortage of such skills is a very serious issue especially in case of constructing many distributed data centers. Another important issue with having many distributed data centers is the load balancing between the data centers. This can be impacted by the availability and cost of high network bandwidth connecting data centers.

The contributions of our work are as follows. First, the objective of our proposed model is to decide the best expansion strategy for a given cloud provider by deciding whether it is beneficial for the cloud provider to build new data centers or to simply expand the data centers it currently has. To the best of our knowledge, no prior work has addressed this problem explicitly. Second, our proposed model addresses the problem of heterogeneity of resources (like servers) and traffic types (with their varying delay constraints). This is another aspect that has not been addressed explicitly before, to the best of our knowledge. Third, our proposed model aims to satisfy the conflicting goals of maximizing the revenue while minimizing the operational cost (OPEX) for the provider. Moreover, it has to perform well for varying conditions at the different geographical regions, varying prices of electricity, different kinds of renewable power sources and their availabilities and different traffic types throughout the day/year.

The rest of this paper is organized as follows. Section 2 discusses the system model. Section 3 explains the simulation results and shows the optimization results. Section 4 includes a literature review for some of the optimization techniques. Finally, the conclusion and future work are presented in Section 5.

Section snippets

System model

In this section, an optimization problem is formulated using mixed integer-linear programming to address the problem of determining the best expansion strategy a cloud provider can take to face the increasing demands and to increase its revenue. The computed strategy may include expanding current data centers by increasing the number of servers they contain or building new data centers (which involves determining how many data centers to build, where to build them and in which capacities). As

Simulation results

In this section we present the simulation results of the proposed optimization models discussed in Sections 2.1 The expansion strategy optimization model, 2.2 The heterogeneity of resources and traffic model. The simulation experiments are conducted on a virtual machine running Windows 7 (64-bit) with 16 GB of RAM and 4 processors.

The optimization problems in the two models are solved using CPLEX2 with Microsoft Visual Studio. CPLEX is a mixed integer-linear

Literature review

The main problem addressed in this work is the expansion strategies of cloud providers to meet the increasing user demands. The body of work on this problem is limited since most of the current works focus on optimizing the currently available data centers by improving power consumption, cooling, request routing, etc. For a broader coverage of such issues, the interested readers are referred to recent surveys such as [22]. We start our discussion of the related work by discussing these issues

Conclusion and future work

With the growth of ultrascale data centers around the world, research on reducing operational cost (OPEX) in the data center is still in its infancy. In this work, we addressed the problem of deciding the best expansion strategy for a given cloud provider by deciding whether it is beneficial for the cloud provider to build new data centers or to simply expand the data centers it currently has. Choosing future sites for constructing new data centers requires careful consideration on several

References (43)

  • W. Van Heddeghem et al.

    Distributed computing for carbon footprint reduction by exploiting low-footprint energy availability

    Future Gener. Comput. Syst.

    (2012)
  • U. Brenner

    A faster polynomial algorithm for the unbalanced hitchcock transportation problem

    Oper. Res. Lett.

    (2008)
  • J. Shi et al.

    Resource planning for parallel processing in the cloud

  • J. Shi et al.

    Sustainable GPU computing at scale

  • J. Koomey

    Growth in Data Center Electricity Use 2005 to 2010

    (2011)
  • A. Rahman et al.

    A survey on geographic load balancing based data center power management in the smart grid environment

    Commun. Surv. Tutorials, IEEE

    (2014)
  • Y. Jararweh, A. Hary, Y. Al-Nashif, S. Hariri, A. Akoglu, D. Jenerette, Accelerated discovery through integration of...
  • Y. Jararweh et al.

    Power and performance management of GPUs based cluster

    Int. J. Cloud Appl. Comput.

    (2012)
  • M. Wardat et al.

    To build or not to build? Addressing the expansion strategies of cloud providers

  • A.-H. Mohsenian-Rad et al.

    Energy-information transmission tradeoff in green cloud computing

    Carbon

    (2010)
  • R. Brown et al., Report to Congress on Server and Data Center Energy Efficiency: Public Law 109-431,...
  • X. Fan et al.

    Power provisioning for a warehouse-sized computer

    SIGARCH Comput. Archit. News

    (2007)
  • J.L. Berral et al.

    Building green cloud services at low cost

  • S. Baldwin

    Carbon footprint of electricity generation

    London: Parliament. Office Sci. Technol.

    (2006)
  • M. Ghamkhari et al.

    Energy and performance management of green data centers: a profit maximization approach

    IEEE Trans. Smart Grid

    (2013)
  • D.P. Bertsekas

    Nonlinear Programming

    (1999)
  • H.R. Lourenço et al.

    Iterated Local Search

    (2003)
  • Energy Information Administration, State Electricity Profiles – Summer Capacity,...
  • Í. Goiri et al.

    Intelligent placement of datacenters for internet services

  • A. Qureshi et al.

    Cutting the electric bill for internet-scale systems

    ACM SIGCOMM Comput. Commun. Rev.

    (2009)
  • K. Papagiannaki et al.

    Analysis of measured single-hop delay from an operational backbone network

  • Cited by (14)

    • Resilient service provisioning in cloud based data centers

      2018, Future Generation Computer Systems
      Citation Excerpt :

      The authors solved it using the fast polynomial algorithm proposed by Brenner [29]. Another line of work using MILP formulation is [30,31]. Given the workload of different geographical locations and the current capabilities of existing data centers, the devised model’s goal is to maximize revenue by either expanding the existing data centers and/or building new ones.

    • Sustainable Cloud Data Centers: A survey of enabling techniques and technologies

      2016, Renewable and Sustainable Energy Reviews
      Citation Excerpt :

      As MDC designs are based on shipping containers, they enable relocation of CDC nodes to places where sustainable computing opportunities are abundant. Hence, the opportunistic relocation of CDCs nodes is based on two factors: (a) on-site availability of renewable energy resource and (b) proximity to free cooling resources and waste heat recovery opportunities [12,13]. Sustainable CDCs are supported by and dependent on geo-dispersed MDC designs and virtualization based workload migration techniques.

    • Green machine learning approaches for cloud-based communications

      2023, Green Machine Learning Protocols for Future Communication Networks
    View all citing articles on Scopus
    View full text