1. Towards an integrated multimedia service hosting overlay Dongyan Xu Xuxian Jiang Proceedings of the 12th annual ACM international conference on Multimedia

2. Author Dongyan Xu Department of Computer Sciences, Purdue University Xuxian Jiang Department of Computer Sciences, Purdue University

3. Outline Introduction MSODA Overview Three-Layer MSODA Middleware Service Composition Layer Service Configuration Layer Service Mapping Layer Performance Evaluation

4. Introduction Digital multimedia is permeating and enriching. Many different platform are ready to process and display multimedia data of varying format and quality. Multimedia data become more manipulatable and semantics. So, multiple media services may be composed.

5. MSODA A service-oriented platform that host multimedia services for on-demand user access and composition. MSODA federates a large variety of media services and forms a rich collaborative service hosting overlay network.

6. Overview From the user’s point of view, MSODA is shared integrated “market place” for media service access and composition. From the media service provider’s point of view, MSODA provide an integrated platform for individual providers to deployed their service without concerning the global networking.

7. The Design of MSODA A virtualization approach is taken to decouple the provisioning of MSODA functions from individual media services. MSODA enables flexible delivery of media services by suggesting alternative service configurations suitable for different resource conditions. MSODA performs cost-effective monitoring of end-to-end network conditions between MSODA nodes.

8. Virtualization-Based Service Hosting The data plane of MSODA. Each MSODA node hosts a number of media services. The same media service can be dynamically replicated or migrated within MSODA overlay network. In a MSODA node, each media service instance runs inside a virtual machine. The virtual machine has it own IP address, customized operating system, and networking capability.

9. Advantages of Virtualization-Based Approach Decoupling of multimedia service management and MSODA platform management. Service installation isolation. Easy service relocation and replication. Isolation of fault and attack impact.

10. Three-Layer MSODA Middleware The control plane of MSODA.

11. The service composition layer facilitates the specification of new composite media service. The service configuration layer accepts a service specification, resolves the basic services involved, and customizes the composite service according to client capability and user preferences/needs. The service mapping layer is responsible for mapping service configurations to MSODA overlay. Three-Layer MSODA Middleware (cont.)

12. Service Configuration Layer

13. Overlay Monitoring for Service Mapping Service mapping involves two interrelated tasks: To map the basic media services to MSODA virtual machines. To route the media streams in the MSODA overlay network. The main challenge in finding service delivery overlays in MSODA is the monitoring of dynamic capacity of network and MSODA nodes.

14. Basic Mesh Maintenance

15. Basic Mesh Maintenance (cont.)

16. Mesh Augmentation via Cold Connection Jumpstart The basic mesh maintenance method fits for some application-level multicast, but it is not sufficient for service mapping in MSODA. d max (P) is a small value or even only one. Such a small d max (P) will make the following problems: Even if the overlay network has enough resource for a service request, but the system could not find out the path. The route path will be too long to make some nodes that relay rather than process data.

17. Mesh Augmentation via Cold Connection Jumpstart (cont.) The cold connections are the connections not currently in the mesh. Modify the mesh adjustment as follow:

18. Performance Evaluation The simulations are conducted in a virtual machine-based emulation test-bed called vBET. The simulated MSODA overlay and underlying IP network are show as the figure.

19. Simulation Setup To set different d max in each simulation run. The total capacity of each node is randomly selected between 2500 units and 10000 units. The total bandwidth of each underlying network link is randomly chosen between 400 units and 1600 units. Service mapping requests are generated according to a Poisson process. T p = 5mins, T q = 30mins.

20. Simulation Results In the simulation, these algorithm are compared: CG(-): The service path finding algorithm with complete graph mesh. SPF(w/): The service path finding algorithm with mesh augmentation method. SPF(w/o): The service path finding algorithm with the basic method only and without mesh augmentation. SW(w/o): The shortest-widest-path algorithm. The algorithm computes the shortest widest path and determines the service-to-virtual machine mapping.

21. Simulation Results (cont.)

24. Analysis of Mesh Augmentation Method N: the total number of MSODA nodes. For a given d max, the number of non-neighbors of each nodes is n = N – d max -1. Assume the MSODA node probes a non-neighbor every fixed interval of t q =T q /n. The arrival of service path mapping requests follows Poisson distribution with an average arrival rate λ req. The duration of each service session is an exponentially distributed random variable with a mean of 1/μ. Let n(t) be the number of warm connection at time t.

25. Analysis of Mesh Augmentation Method (cont.)