Scalable and Efficient Content Distribution

Funded by: Center for Industrial Information Technology (CENIIT) (Start date: Jan. 2011) Project leader: Niklas Carlsson, Associate Professor (docent; universitetslektor) Department of Computer and Information Science (IDA), Linköping University Email: niklas.carlsson@liu.se

Overview: Content-based services (such as video streaming) play a central role in society and our everyday lives. Today, content delivery applications (such as YouTube, Netflix and Spotify, for example) consume a majority of the Internet bandwidth and the demand is continually increasing. This research project provides efficient and improved ways to disseminate content and information. We explore the best protocols and architectures to serve large numbers of users/clients. Of particular interest is design, modeling, and performance evaluation that provides solid insights towards the best possible performance, scalability, efficiency, and/or quality of service.
content

Background and Industrial Perspective

With tremendous improvements in network bandwidth and computer capabilities, many new high-bandwidth content distribution services have emerged in the entertainment, business, and scientific communities. In contrast to traditional content distribution systems such as TV and radio broadcasts, many of these new services operate in an on-demand basis and only serve clients when explicit requests for service are made. Further, many of these new services allow the delivered content to be personalized.

Today, content delivery applications (such as YouTube, Netflix and Spotify, for example) consume a majority of the Internet bandwidth. With continued growth in demand for such applications anticipated, the problem of cost-efficient and/or sustainable content delivery becomes increasingly important. For efficient delivery, protocols and architectures must scale well with the request loads; i.e., it is important that protocols are designed such that the marginal delivery costs reduce with increasing demands. Using scalable techniques can allow a content distributor to handle higher demands more efficiently, and/or to offer its existing customers better service while reducing its resource requirements and/or delivery costs.

A variety of techniques have been studied to improve the scalability and efficiency of content delivery, including replication, service aggregation, and peer-to-peer techniques. With replication, multiple servers (possibly geographically distributed as in a CDN) share the load of processing client requests and may enable delivery from a nearby server. With aggregation, multiple client requests are served together in a manner that is more efficient than individual service. Finally, with peer-to-peer techniques, clients may contribute to the total service capacity of the system by providing service to other clients.

While much work have considered various scalable solutions, there is a lack of literature considering the problem of cost-efficient content delivery, in which the application incurs both a network delivery cost (e.g., from cross ISP traffic or, more generally, operation/energy costs at Internet routers) and costs at the servers (e.g., due to cost of ownership, energy, or disk bandwidth). While the problem is complicated by the fact that the cost objective and the absolute cost tradeoff may be different from case to case, contributions that help identify the most efficient delivery architectures and protocols are becoming increasingly important for organizations distributing content over the Internet. This importance may be further augmented by the potential of increasing energy costs and carbon taxes, for example.

There are also many other trends that make content and service delivery a topic with increasing commercial impact. For example, increased bandwidth capacities, improved wireless access, and new high-capacity devices, such as the Apple iPhone, have enabled the emergence of many new, popular, and exciting applications, which themselves have the potential to attract new customer audiences. While this step towards greater mobility may be seen as the actualization of anything-anywhere-anytime services (that ideally allow users access to anything they want, whenever and wherever they are), the trend has also had great performance implications, as much more bandwidth is being used for data traffic generated by smartphones and other mobile devices. These and other trends have gathered significant interest from the industry, which currently is investing in research of new technologies that help disseminate content efficiently. In summary, there is much potential for new services and technologies of content delivery.

Research Vision and Project Goals

We will build a highly recognized research group that specializes in distributed systems and networks. For this goal to be achieved it is important that the group remains aligned with the most important problems, helps educate the next generation of top-scientists, and consistently produces high-quality research, shared through publications in premier conferences/journals or more targeted high-profile venues. For greater impact we also plan to release open source implementations and datasets. Overall, we want to be known for doing great research! This will involve cooperation with both academic and industrial partners.

At a technical level, the primary long-term goal of this research project is to contribute towards providing simple and resource effective ways to disseminate content and information. Ultimately, content/information should be easily accessible whenever and wherever you are, at a low environmental and monetary cost.

Our research will primarily focus on performance and resource usage/cost aspects of protocols, services, and systems. Our efforts will be focused towards the problems that we believe have the potential to yield the most significant contributions, and we should always try to use the best tools for the problem at hand. Following the direction of the applicant, the research will include a combination of analytic modeling, simulations, measurements, system implementation, as well as real-world experiments.

Project Description

While the research group has a relatively broad interest, we are very much interested in exploring the best protocols and architectures to serve large numbers of users/clients. Of particular interest is design, modeling, and performance evaluation that provides solid insights towards the best possible performance, scalability, efficiency, and/or quality of service.

The long-term goal of this research project is to contribute towards providing simple and effective ways to disseminate content and information.

We will design, model, and evaluate candidate solutions, such as to provide solid insights towards the best possible performance, scalability, efficiency, and/or quality of service. Our work will use a combination of measurements, mathematical modeling, system implementation, and real-world experiments. When possible, analytic lower bounds will be developed for the purpose of rigorous protocol and system evaluation.

To better understand the underlying conditions under which protocols and systems operate, we perform measurement-based studies in which network dynamics, workloads, and/or various system performance aspects are characterized. These results help answer questions such as what are the current limitations and bottlenecks; what makes some content, services, and systems more popular and/or successful than others; etc. However, these studies also gives insight towards more fundamental questions regarding access patterns, content selection, etc. Insights, datasets, and models created during this part of the project is also leveraged when we design and evaluate new distributed systems and protocols.
We design and evaluate the performance of various new scalable protocols and architectures that help improve the quality of service and resource usage. Ongoing work include the design of hybrid system solutions that minimize the total service cost associated with large-scale systems, in which information and content are replicated at many locations and with the help of different technologies (e.g., peer-to-peer, cloud, proxies, and in datacenters). We are particularly interested in solutions that scale with both the number clients and with the diversity of service that these clients are provided (e.g., a large catalogue of contents), not only one or the other.

Project Environment and Relevance

Industrial Relevance

Content-based services (such as video streaming) is a very big part in society and our everyday lives. In addition to our initial observations that content delivery contributes a majority of current Internet traffic, and that the portion is continually increasing, we note that more and more users expect to watch whatever content they want to watch, online, whenever and wherever they are. This trend towards anything-anytime-anywhere service is likely to continue, and companies are trying to take advantage of these trends. Simultaneously, companies are becoming increasingly aware of their energy consumption and overall delivery costs. As carbon taxes and energy-price differentiation (due to differences in energy source, for example) becomes common, energy efficiency and sustainable content delivery approaches are likely to have a very direct impact of the bottom line of many companies. Finally, we note that there is a very wide range of companies that deliver (or plan to deliver) content over the Internet. Some companies are using peer-assisted solutions, while others are selecting server-based solutions. We expect both approaches generating commercial interest and revenue; our research will not be restricted to either approach.

Research Environment

Our new networks and systems group has begun as a subgroup under ADIT. During the building of the group, this will ensure access to some local resources. The group is lead by Niklas Carlsson, Associate Professor (docent; universitetslektor). Niklas is supervisor (or co-supervisor) for four local PhD students (Vengatanathan Krishnamoorti, Rahul Hiran, Anna Vapen, and Mats Gustafsson), one researcher (Cyriac James), and a number of thesis students. Through thesis opportunities, for example, we are actively trying to identify good candidate members for the group.

As of today we have also hosted a visiting PhD student for six months (from NICTA, Australia, that the applicant has assisted in the supervision for the last two years of her thesis), hired a postdoc for three months, two researchers (one of which has been upgraded to PhD student in the group). The work by these individuals has contributed greatly to the project. These works include the design of throughput optimal second-spectrum auctions to better utilize the wireless spectrum, models and statistical analysis to gain insights into what makes some content more popular than others, an analytic framework that allow for the design and evaluation of optimal replica selection policies for content delivery systems with replicated service, and the evaluation framework and performance evaluation of proxy-assisted HTTP-based adaptive streaming protocols. Within this framework, we have also hosted senior researchers from NICTA (Australia) and University of Calgary (Canada), hosted a visiting undergraduate internship student from France, and helped graduate two PhD thesis projects (Youmna Boghol, NICTA, and Aniket Mahanti, University of Calgary).

Cooperation and Industrial Partnership

Much of the proposed research is done in cooperation with researchers in both academia and industry. Collaboration and partners include both Swedish and international companies and institutions.

We are actively sharing findings, datasets, and discussing various research projects with our Swedish industry partners: Ericsson (Kista and Linkoping), Peerialism (Kista), and Spotify (Stockholm).
We are actively writing reserach papers, organizing workshops, and have regular research visits with our primary international industry partners and their employees: National ICT Australia (NICTA) and HP Labs (Palo Alto, CA).
We are actively writing research papers and have regular research visits/meetings with our (primary) academic collaborators: Derek Eager (University of Saskatchewan, Canada), Carey Williamson (University of Calgary, Canada), Aniket Mahanti (University of Auckland, New Zeeland), and György Dan (Royal Institute of Technology, Sweden).

While much of the existing work has done with international partners, we are actively increasing our collaboration and discussions with Swedish industry. This includes master theses projects and discussions about future PhD student internships. However, we are also interested in other forms of collaboration with existing and future industry partners.

Relationship to Other CENIIT Projects

We believe that computer networks research is a very important research area that the Department and the University must strengthen its research in. For this reason, the primary applicant was hired (in Sept 2010) and is currently in the process of building a new research group. We further believe that the direction of the project align very well with CENIIT's goals.

Recruitment and Outreach

Research opportunities: If you are interested in working on this project (or related projects), we are looking for hardworking and ambitious people that are interested in research and/or problem solving. If you fit this description, please send an email to Niklas Carlsson (niklas.carlsson@liu.se). More information about the research, thesis projects, and positions can be found here.

Companies and organizations: If you are a company or organization that is interested in any of the topics included within this project (and/or the general research interest of the group), we would be interested to discuss potential mutual interests. Please send an email to Niklas Carlsson (niklas.carlsson@liu.se).

Publications

This project started in January 2011. Since then we have had two successful and productive years, with many publications published in premier journals (e.g., IEEE/ACM ToN, ACM TWEB, ACM TOIT, and Performance Evaluation), magazine (IEEE Network), and selective conferences (including including ACM SIGKDD and IFIP Performance). For an up-to-date list of publications related to the project we refer to Niklas Carlsson's publication list.