1	Subscriber Mobility in Pub/Sub Systems: Pro-Active vs. Reactive Handoffs Thomas Kunz Systems and Computer Engineering Carleton University
2	Publish/Subscribe Systems: Background The pub/sub service is a logically centralized service intermediates the communication between publishers and subscribers in a distributed setting. Basic primitives: sub, unsub, pub Various routing topologies and semantics The pub/sub communication paradigm provides: space, flow, and timing decoupling between producers and consumers of information content-based filtering and communication Multi-way delivery A frequently used communication paradigm Proposed for mobile computing
3	Pub/Sub System on Top of Wireless Network
4	Pub/Sub & Mobility: Problem Definition How to cope with mobile subscribers? Disconnected operations: Buffering & queue management Handover protocol relocating subscriptions and updating the topology How to balance subscriber’s load in the event-service? Load balancing strategy Few studies focused on extending pub/sub Goals of research Develop a mobility support service for pub/sub systems Validate its performance in a testbed
5	Java Message Service (JMS): Overview API specification to implement MOM services Two communication modes specified: Point-to-point (one-to-one) Publish/subscribe (event-based service) Two subscription schemes adapted: Durable scheme Non-durable scheme Two consumption modes supported: Asynchronous mode Synchronous mode Other features
6	Characterization of JMS Primitives Wired/wireless networks JMS primitives Compare primitives with high & low reliability Detailed results appears in “Performance of Pub/Sub Systems in Wired/Wireless Networks”, in Proceedings of 64th IEEE VTC, September 2006 Key observations Non-durable subscriptions outperform durable ones Durable subscriptions show comparable throughput results to non-durable ones in wireless domain Reliability cost of durable subscriptions is relatively low in wireless environment
7	Possible Mobility Solutions Reactive approach: state-transfer occurs only after the mobile subscribers reconnect to the new broker Commonly used: JEDI, REBECA, SIENA, CEA, .. Durable subscription-based approach: every broker buffers all the published messages irrespective of its current active subscriptions Recently used: JMS implementations (OpenJMS, FiornaMQ, JBossMQ, JavaSMQ), Ivana et al., .. Pro-active Approach: context transfer/caching occurs prior to the actual movement of the mobile subscriber (our proposal) Passive subscriptions are propagated to neighbors of current broker Subscriptions get activated once subscriber disconnects, start buffering messages Once subscriber reconnects to a broker, will with high probability find active subscriptions and locally buffered messages, subscriptions in old and new neighborhood get updated accordingly
8	Neighbor Graph (NG) Is an undirected graph with a number of edges that represent the mobility paths between the vertices Provides the abstractions to identify the candidate subset of next potential brokers Forms the basis for pre-loading the subscriber context one hop ahead of its current broker Automatically generated and adaptively changes according to the mobility graph A timestamp based LRU method can be used to ensure the freshness of NG and remove the outlier edges Is stored in a distributed manner
9	Implementing Pro-active Approach
10	Experimental Setup & Mobility State Diagram
11	Performance Results and Comparison I Mobility extension overhead In terms of message processing time and message throughput
12	Performance Results and Comparisons II Cumulative Distribution of Handoff Time
13	Performance Results and Comparisons III Message Loss, Duplication, Throughput As a function of the publication rate
14	Summary and Future Work Proactive approach to support mobile subscribers Evaluated through small testbed Superior performance to reactive approach: Higher throughput Lower message latency Lower handoff latency No message duplication Low message loss rate Future Work: Larger testbed (some initial results) Impact of mobility patterns/neighborhood graph approximation Compare to durable approach (some initial results) Analytical performance bounds/prediction models