Socially-aware pub-sub system for human networks Yaxiong Zhao Jie Wu Department of Computer and Information Sciences Temple University Philadelphia 19122
Outline Background and motivation Pub-sub system design –Subscription representation and processing –Pub-sub routing Experiment results Conclusion
Outline Background and motivation Pub-sub system design –Subscription representation and processing –Pub-sub routing Experiment results Conclusion
Background: Why human networks? Mobile wireless networks have been a dream –A lot of research Ad hoc networks Central of the past 20 years' research –Hardly hear any successful stories People used to believe that mobile wireless networks should: –Support wireless internet –Be connected at all times –These are difficult and even impossible to realize
Background: Wireless networks that we did not build Twitter: send messages to your followers and receive from people you are following –Very popular on mobile devices Delay Tolerant networks –Intermittently connected mobile devices/hosts How about combine them together? –A network formed by human carried wireless devices –Running social network applications –Do not require Internet-like infrastructure
Pub-sub for human networks Pub-sub is a powerful paradigm –Publishers generate messages –Clients consume messages –Brokers forward messages according to their contents The benefits of Pub-sub –Anonymity –Loose coupling –Flexibility However, it requires complex processing on brokers and does not consider mobility –This paper tackles these problems
Outline Background and motivation Pub-sub system design –Subscription representation and processing –Pub-sub routing Experiment results Conclusion
Overview Two components –Content representation Subscriptions and events We use old classic methods in the literature –Pub-sub routing Social election Find socially-active users to forward messages
Outline Background and motivation Pub-sub system design –Subscription representation and processing –Pub-sub routing Experiment results Conclusion
Traditional content representation Subscriptions are represented as conjunctions of multiple attribute constraints –Each attribute has a constraint –Age = [10, 20], Height = [120, 190] –A subscription corresponds to a multi-dimensional region –An event is a multi-dimensional point Excellent expressiveness High processing and storage costs –Matching in multi-dimensional space is NP-hard in worst-case
Outline Background and motivation Pub-sub system design –Subscription representation and processing –Pub-sub routing Experiment results Conclusion
Pub-sub routing Brokers are responsible for forwarding messages Who should be brokers? –This is not a problem for traditional pub-sub systems –However, in Human networks, it is difficult to find such users
Does DTN routing work? The answer is, NO –It breaks the anonymity of pub-sub –Requires a lot of pre-processing Impractical in practice The obtained results do not hold for newly aquirred users in the network It is difficult to obtain such data in the first place
Social election Human networks are a social network –There will be active users moving around –How to find such users? –Election! Each user should be in contact with a certain number of brokers –An interval [lower_bound, upper_bound] –If a user meets brokers less than or lower_bound I may stay too far from the crowds –If the number is larger than upper_bound I do not need so many brokers
Social election cont'd Eventually, the most active users will become brokers –Since they move around in a larger area –They are more likely to become brokers
Social election cont'd A heuristic based on popularity –The popularity of a user is measured as the number of different users it met in a time window [now – T, now] –This time window is the same as the one used in the election –The user should always select those of a higher popularity to be brokers
Pub-sub forwarding based on utility A message's utility is defined as the division of the message's matching score and its age –An old message has less utility The messages in a brokers buffer are ranked according to their utilities
Pub-sub forwarding cont'd Forwarding happens only between brokers Always forward highest-ranked messages Buffer management –When the buffer is over-flowed –The lowest ranked messages will be purged from the buffer
Delegation forwarding A utility threshold for each message Forward it only when the next-hop has a better utility than its own threshold –The threshold raises after a successful forwarding –Reduce copy numbers
Outline Background and motivation Pub-sub system design –Subscription representation and processing –Pub-sub routing Experiment results Conclusion
Experiment setting Two mobility models –RWP and SLAW (mimic human mobility) Written in C users in a 1000*1000m 2 region Communication range 50m Compare with Random selection of brokers –A fraction of users are selected as brokers –The ratio is made to be the same as that obtained in our system
Delivery ratio RWP and delegation forwarding
Delivery ratio SLAW and delegation forwarding
Changing of brokers’ numbers with moving speed (RWP)
Outline Background and motivation Pub-sub system design –Subscription representation and processing –Pub-sub routing Experiment results Conclusion
Flooding in the entire network is too resource consuming Finding a small set of brokers is sufficient for efficient message delivery
Questions? Thanks for listening!