Comprising more than 61,000 servers located across nearly 1,000 networks in 70 countries worldwide, the Akamai platform delivers hundreds of billions of Internet interactions daily, helping thousands of enterprises boost the performance and reliability of their Internet applications. In this paper, we give an overview of the components and capabilities of this large-scale distributed computing platform, and offer some insight into its architecture, design principles, operation, and management.

/pdf/the-akamai-network-a-platform-for-high-performance-internet-1f8jpgl9we.pdf

The Akamai network: a platform for high-performance internet applications

Data centers are critical, energy-hungry infrastructures that run large-scale Internet-based services. Energy consumption models are pivotal in designing and optimizing energy-efficient operations to curb excessive energy consumption in data centers. In this paper, we survey the state-of-the-art techniques used for energy consumption modeling and prediction for data centers and their components. We conduct an in-depth study of the existing literature on data center power modeling, covering more than 200 models. We organize these models in a hierarchical structure with two main branches focusing on hardware-centric and software-centric power models. Under hardware-centric approaches we start from the digital circuit level and move on to describe higher-level energy consumption models at the hardware component level, server level, data center level, and finally systems of systems level. Under the software-centric approaches we investigate power models developed for operating systems, virtual machines and software applications. This systematic approach allows us to identify multiple issues prevalent in power modeling of different levels of data center systems, including: i) few modeling efforts targeted at power consumption of the entire data center ii) many state-of-the-art power models are based on a few CPU or server metrics, and iii) the effectiveness and accuracy of these power models remain open questions. Based on these observations, we conclude the survey by describing key challenges for future research on constructing effective and accurate data center power models.

Data Center Energy Consumption Modeling: A Survey

Large-scale Internet applications, such as content distribution networks, are deployed across multiple datacenters and consume massive amounts of electricity. To provide uniformly low access latencies, these datacenters are geographically distributed and the deployment size at each location reflects the regional demand for the application. Consequently, an application's environmental impact can vary significantly depending on the geographical distribution of end-users, as electricity cost and carbon footprint per watt is location specific. In this paper, we describe FORTE: Flow Optimization based framework for request-Routing and Traffic Engineering. FORTE dynamically controls the fraction of user traffic directed to each datacenter in response to changes in both request workload and carbon footprint. It allows an operator to navigate the three-way tradeoff between access latency, carbon footprint, and electricity costs and to determine an optimal datacenter upgrade plan in response to increases in traffic load. We use FORTE to show that carbon taxes or credits are impractical in incentivizing carbon output reduction by providers of large-scale Internet applications. However, they can reduce carbon emissions by 10% without increasing the mean latency nor the electricity bill.

/pdf/it-s-not-easy-being-green-14hm420p5s.pdf

It's not easy being green

Timely and cost-effective analytics over “Big Data” is now a key ingredient for success in many businesses, scientific and engineering disciplines, and government endeavors. The Hadoop software stack—which consists of an extensible MapReduce execution engine, pluggable distributed storage engines, and a range of procedural to declarative interfaces—is a popular choice for big data analytics. Most practitioners of big data analytics—like computational scientists, systems researchers, and business analysts—lack the expertise to tune the system to get good performance. Unfortunately, Hadoop’s performance out of the box leaves much to be desired, leading to suboptimal use of resources, time, and money (in payas-you-go clouds). We introduce Starfish, a self-tuning system for big data analytics. Starfish builds on Hadoop while adapting to user needs and system workloads to provide good performance automatically, without any need for users to understand and manipulate the many tuning knobs in Hadoop. While Starfish’s system architecture is guided by work on self-tuning database systems, we discuss how new analysis practices over big data pose new challenges; leading us to different design choices in Starfish.

/pdf/starfish-a-self-tuning-system-for-big-data-analytics-1c4ab07m8g.pdf

Starfish: A Self-tuning System for Big Data Analytics.

Virtualization is often used in cloud computing platforms for its several advantages in efficiently managing resources. However, virtualization raises certain additional challenges, and one of them is lack of power metering for virtual machines (VMs). Power management requirements in modern data centers have led to most new servers providing power usage measurement in hardware and alternate solutions exist for older servers using circuit and outlet level measurements. However, VM power cannot be measured purely in hardware. We present a solution for VM power metering, named Joulemeter. We build power models to infer power consumption from resource usage at runtime and identify the challenges that arise when applying such models for VM power metering. We show how existing instrumentation in server hardware and hypervisors can be used to build the required power models on real platforms with low error. Our approach is designed to operate with extremely low runtime overhead while providing practically useful accuracy. We illustrate the use of the proposed metering capability for VM power capping, a technique to reduce power provisioning costs in data centers. Experiments are performed on server traces from several thousand production servers, hosting Microsoft's real-world applications such as Windows Live Messenger. The results show that not only does VM power metering allows virtualized data centers to achieve the same savings that non-virtualized data centers achieved through physical server power capping, but also that it enables further savings in provisioning costs with virtualization.

/pdf/virtual-machine-power-metering-and-provisioning-1yj7epvvd8.pdf

Virtual machine power metering and provisioning

Energy expenses are becoming an increasingly important fraction of data center operating costs. At the same time, the energy expense per unit of computation can vary significantly between two different locations. In this paper, we characterize the variation due to fluctuating electricity prices and argue that existing distributed systems should be able to exploit this variation for significant economic gains. Electricity prices exhibit both temporal and geographic variation, due to regional demand differences, transmission inefficiencies, and generation diversity. Starting with historical electricity prices, for twenty nine locations in the US, and network traffic data collected on Akamai's CDN, we use simulation to quantify the possible economic gains for a realistic workload. Our results imply that existing systems may be able to save millions of dollars a year in electricity costs, by being cognizant of locational computation cost differences.

/pdf/cutting-the-electric-bill-for-internet-scale-systems-4awcltuqwb.pdf

Cutting the electric bill for internet-scale systems

There is an increasing trend toward massive, geographically distributed systems. The largest Internet companies operate hundreds of thousands of servers in multiple geographic locations, and are growing at a fast clip. A single system's servers and data centers can consume many megawatts of electricity, as much as tens of thousands of US homes. Two important concerns have arisen: rising electric bills; and growing carbon footprints. Our work develops a new traffic engineering technique that can be used to address both these areas of concern. 
We introduce Power-Demand Routing (PDR), a technique that redistributes traffic between replicas with the express purpose of spatially redistributing the system's power consumption, in order to reduce operating costs. Cost can be described in monetary terms or in terms of pollution. Within existing Internet services, each client request requires a meaningful amount of marginal energy at the server. Thus, by rerouting requests from a server at one geographic location to another, we can spatially shift the systems marginal power consumption at Internet speeds. 
We show how PDR can be used to reduce electric bills. We describe how to couple request routing policy to real-time price signals from wholesale electricity markets. In response to price-differentials, PDR skews client load across a system's clusters and pushes server power-demand into the least expensive regions. Our analysis quantifies the potential reduction in energy costs. We use simulations driven by empirical data and models: we collected a real-world request traffic workload in collaboration with Akamai; constructed data center energy models; and compiled a database of historical electricity market prices. We conclude that existing systems can use PDR, to cut their annual electric bills by millions of dollars. 
We also show how PDR can be used to reduce carbon footprints. Not all joules are created equal and in power pools like the grid the environmental impact per joule varies geographically and in time. We show how to construct carbon cost functions that can be used with PDR to dynamically push a system's power-demand toward clean energy. (Copies available exclusively from MIT Libraries, Rm. 14-0551, Cambridge, MA 02139-4307. Ph. 617-253-5668; Fax 617-253-1690.)

Power-demand routing in massive geo-distributed systems

Inverse multiplexing, or network striping, allows the construction of a high-bandwidth virtual channel from a collection of multiple low-bandwidth network channels. Striping systems usually employ an immutable packet scheduling policy and allow applications to be oblivious of the way in which packets are routed to specific network channels. Though this is appropriate for many applications, other applications can benefit from an approach that explicitly involves the application in the dynamic determination of the striping policy.Horde is middleware that facilitates flexible striping in wireless environments for a diverse range of applications. Horde separates the striping policy from routing and scheduling. It allows applications to specify network quality-of-service objectives that the striping mechanism attempts to satisfy. Horde can be used by a set of application data streams, each with its own quality-of-service policy, to flexibly stripe data over a highly heterogeneous set of dynamically varying wireless network channels.We present the Horde architecture, describe an early implementation, and examine how different policies can be used to modulate the quality-of-service observed across different independent data streams.

/pdf/horde-separating-network-striping-policy-from-mechanism-2mvk1v5puu.pdf

Horde: separating network striping policy from mechanism

Tavarua is a multimedia streaming system that leverages network-striping to deliver relatively high bit rate video over present-day cellular wireless wide-area networks. The Tavarua system achieves this by building on our previously developed flexible network-striping middleware. This paper describes a motivating mobile telemedicine application, and the design of the Tavarua system. It also describes experiments in which our initial Tavarua implementation was used to stripe video over multiple 3G cellular-phones from different providers.

/pdf/tavarua-video-streaming-with-wwan-striping-3lbun6ng46.pdf

Tavarua: video streaming with WWAN striping

In North America, electricity prices exhibit both temporal and geographic variation—the later exists due to regional demand differences, transmission inefficiencies and generation diversity. Using historical market data, we characterize the variation and argue that existing distributed systems should be able to exploit it for significant economic gains. We consider pricing in cloud computing systems, and also use simulation to estimate the advantage of dynamically shuffling computation between different

/pdf/plugging-into-energy-market-diversity-4ldlszdnqh.pdf

Asfandyar Qureshi

Papers

Cutting the electric bill for internet-scale systems

Power-demand routing in massive geo-distributed systems

Horde: separating network striping policy from mechanism

Tavarua: video streaming with WWAN striping

Plugging Into Energy Market Diversity.