1. INFORMATION AND COMMUNICATION SYSTEMS MERIT 2008 Research Symposium Melbourne Engineering Graduates Look to the Future System Architecture An internetworking of CDNs is formed by a set of autonomous CDNs, which cooperate through a mechanism that provides facilities and infrastructure for cooperation in order to virtualize multiple providers. Architecture of a system to assist the creation of peering CDNs is shown in Figure 1. Such a constellation permits flexible resource sharing and dynamic collaboration between autonomous CDNs in the form of a peering arrangement. The ‘resource sharing’ approach in the peering CDNs model endeavours to balance a CDN’s service requirements against the high costs of deploying customer-dedicated and therefore over-provisioned resources. It is anticipated that proper management and cooperation will enable a CDN to avoid violating SLAs even when the service demands could not have been predicted ahead of time. Mukaddim Pathan and Rajkumar Buyya GRIDS Laboratory, Department of Computer Science and Software Engineering Email: raj@csse.unimelb.edu.au; Website: http://www.gridbus.org/cdnraj@csse.unimelb.edu.auhttp://www.gridbus.org/cdn iCDN – Internetworking of Content Delivery Networks Figure 1: Abstract architecture for the creation of peering CDNs. Results Key Reference Pathan, M., Vecchiola, C., and Buyya, R. Load and proximity aware request-redirection for dynamic load distribution in peering CDNs. In Proc. of CoopIS’08, Monterrey, Mexico, 2008. Peering CDNs Formation The process of peering negotiation is triggered on traffic surges under degenerated load conditions (e.g. flash crowds). Figure 2 illustrates the typical steps to create a peering arrangement between CDNs. Load and Proximity Aware Request-Redirection A dominant factor for the success of peering between CDNs is to perform load distribution to handle highly skewed loads. Our approach for dynamic load distribution adopts a request- redirection mechanism by taking traffic load and network proximity into account. In our approach, load indices are obtained through an asynchronous feedback mechanism and network proximity is measured using a pinger logic with low messaging overhead. Overview Content Delivery Networks (CDNs) emerged to provide fast and reliable Web access services by distributing content to edge servers located close to end-users. To operate effectively a CDN is required to either over- provision its capacity or to harness external resources on demand. Cooperation between CDNs can reduce costs with over-provisioning and provide users with high quality services in a global scale. This collaboration, termed as peering between CDNs, can be short-term wherein CDNs operate to handle flash crowds, or long-term in which they explore the delivery of specialized services. Aim The proprietary nature of existing CDNs means that they are closed and do not naturally cooperate. Finding ways for distinct CDNs to coordinate and cooperate with other CDNs is necessary to achieve better overall service, as perceived by end-users, at lower cost. This research aims to provide a means for distinct CDNs to coordinate and cooperate with other CDNs, by investigating and developing  an architecture for an open and decentralized system to support effective internetworking between CDNs, which is achieved through a peering arrangement;  protocols for service delivery in a cooperative environment of CDNs;  economic models for an effective content replication policy; and  policies for autonomic management of service level through resource negotiation in an on-demand basis. Figure 2: Typical steps for creating a peering arrangement. PropertiesParametersDescription ActivationActivation trigger (when) Asynchronous (on CDN server request) Activation decision (where) Distributed (Gateway redirection upon requests from distributed servers) ImplementationStatus informationTraffic load (correlated with server response load = utilization * capacity) Alarm (Asynchronous feedback) Redirection policy Server selection (how) Minimize redirection cost from available server list (mapping of overloaded and underloaded server lists) Redirected entities (what) User requests Table 1: Significant properties of the request-redirection scheme Figure 4: Average utilization of the primary CDN in each scheme. Figure 5: Comparison of the request-redirection schemes. Figure 3: Server utilization in different request-redirection schemes.