The JOURNEY Active Network Model Maximilian Ott et al. IEEE Journal on Selected Areas in Communications, vol.19, no. 3, March 2001
Introduction Processing at the terminal end Processing at the server end The goal is to provide processing as an additional network service. The request-response processing model is transferred to continuous transformations on the date streams.
JOURNEY Network Model The transformation of a media unit (MU) is considered as a independent processing job. A sequence of jobs pass through a JOURNEY node, which consists of multiple stages: Classification stage Admission stage Routing stage
Computation as a Network Service Streams of MUs are injected into the network for routing and customizing. An MU is independently processed anywhere along the path guided by routing. A computing router utilize local condition of resource availability for deciding whether to process an MU.
Computation as a Network Service Similar to IP networks, the best-effort processing collocates with error recovery at higher layers. Customization information can be originated at any point of the stream path, such as a client node or a resource manager. Specific path routing is required for dealing with fragmentation of MUs. (MPLS, IP source routing)
Computing Router Architecture The cluster-based active router architecture (CLARA) Routing element Computing element(s) System area network (SAN) Cluster manager
CLARA Architecture
CLARA Functional Overview Ingress Engine DispatchCollect Egress Engine Admit
Router Programming Framework The CLARA software framework is also designed to support: Accounting of the resource utilization of a packet or stream; Division and vending of portions of the computational resources available on a router; Dynamic addition of customization functionality Functionality repositories
Active Media Packet Format
Packet Programming Interface
Admission Control for Soft-QoS Guarantees Unprocessed packet rate (UPR) Packet Admission Control (PAC) :observed delay for a packet :estimator function :current processing backlog :upper bound processing :packet’s delay budget
Cascade Transformations with Multiple Nodes The performance goal of the active network is to bring the UPR of flows below some acceptable value.
Scenic Routing
Experience and Evaluation Media gateway Dropping frames, removing color, stronger uniform compression Meta-information MPEG4, MPEG7 The trend toward thin and mobile clients The scalability problem at the gateway
MPEG Transcoding Service
Performance Measurements
Analysis The larger the input/output bit-rate ratio, the less time it takes to transcode. Frame drop and/or spatial resolution adjustments DCT requantization The total store-processing forward-service is double the processing time. User-space routing engine IP over Myrinet
Scalability in JOURNEY Manageability Self-configuration and self-healing Availability Performance Number of computing routers
Conclusions The JOURNEY network model provides computation as a scalable network services. The computation model trades off hard guarantees for computation in favor of architectural simplicity. The CLARA architecture collocates computing and routing functionality.
Future Works Studying the performance of the admission control and routing mechanisms at different traffic loads Development of a management framework for the discovery and on-demand deployment of transcoding services Development algorithms for admission control and load distribution within a CLARA computing router
Possible Directions Handing computation from proxies into the network Mobile computing, WAP Improving efficiency of multicast routing with heterogeneous receivers Pre-customization of data streams Active flow and congestion control Re-transcoding and/or re-routing of data streams Layered multimedia multicast tree