Distributed Ranked Data Dissemination in Social Networks

Zhang, Kaiwen; Sadoghi, Mohammad; Muthusamy, Vinod; Jacobsen, Hans‐Arno

doi:10.1109/icdcs.2013.19

Cited by 5 publications

(2 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They assume that the entire stream of matching publications is first collected at the subscribers' edge brokers before top-k filtering is applied. We are instead considering dissemination or distribution of the top-k processing [31]. Our approach seeks to apply the top-k filter as close to the sources as possible in order to reduce traffic within the pub/sub network (see Figure 5).…”

Section: Ranked Datamentioning

confidence: 99%

Language expressiveness and quality of service for publish/subscribe systems

Zhang

2012

Proceedings of the 9th Middleware Doctoral Symposium of the 13th ACM/IFIP/USENIX International Middleware Conference

Self Cite

View full text Add to dashboard Cite

The publish/subscribe paradigm is known for its loosely coupled interactions and event filtering capabilities. Traditional applications using pub/sub systems require large-scale deployment and high event throughput. Thus, pub/sub has always put the emphasis on scalability and performance, to the detriment of filtering expressiveness and quality of service. The matching language is usually limited to topic-based or content-based event filtering and does not allow complex stream-based subscriptions to be expressed. Messages are delivered on a best-effort basis without any ordering or reliability guarantees. Installing additional services and event processing systems at the endpoints can overcome the limitations of pub/sub systems. However, we argue that such solutions are inefficient and put a lot of strain on the pub/sub layer itself. Therefore, the focus of this thesis is to develop integrated solutions to extend pub/sub language expression and quality of service, and demonstrate that our approach results in better performance from a holistic perspective. We first describe multiple case studies for pub/sub and identify major features which need to be supported. We then extract those requirements from our use cases and develop general solutions within the pub/sub layer. Features we have supported so far includes total order and ranked data dissemination. Finally, we conduct experiments to compare the performance of our approach to baselines which rely on endto-end services and perform holistic evaluations to assess the impact of our work.

show abstract

Section: Ranked Datamentioning

confidence: 99%

Language expressiveness and quality of service for publish/subscribe systems

Zhang

2012

Proceedings of the 9th Middleware Doctoral Symposium of the 13th ACM/IFIP/USENIX International Middleware Conference

Self Cite

View full text Add to dashboard Cite

show abstract

“…This is a common and practical assumption for many systems; we leave it for future work to consider more general overlays. Moreover, there exist several extensions [24,23,9] for pub/sub on a tree overlay. With this overlay as a basis, we hope that these extensions can also be adapted to our system.…”

Section: Introductionmentioning

confidence: 99%

Towards Byzantine fault tolerant publish/subscribe

Jehl

Meling

2013

Proceedings of the 9th Workshop on Hot Topics in Dependable Systems

View full text Add to dashboard Cite

More than a decade of research has gone into techniques aimed at tolerating arbitrary failures in client/server interaction, using consensus based replication. These works made Byzantine fault tolerance possible [5], competitive [18], robust [7], and feasible to apply [6]. In this paper we establish a connection between the pub/sub interaction model and consensus based replication protocols, that makes the above results applicable to the design of large scale eventbased middleware. We propose a Byzantine fault tolerant pub/sub system, on a tree-based overlay, tolerating a configurable number of failures in any part of the system, with minimal divergence from traditional pub/sub specifications and forwarding schemes.

show abstract

Distributed Ranked Data Dissemination in Social Networks

Zhang

Sadoghi

Muthusamy

et al. 2013

2013 IEEE 33rd International Conference on Distributed Computing Systems

View full text Add to dashboard Cite

Abstract. The amount of content served on social networks can overwhelm users, who must sift through the data for relevant information. To facilitate users, we develop and implement dissemination of ranked data in social networks. Although top-k computation can be performed centrally at the user, the size of the event stream can constitute a significant bottleneck. Our approach distributes the top-k computation to reduce the number of events flowing through. The solution operates on an overlay network model, such as pub/subİnstead of forwarding all events, nodes closer to the sources can identify and forward top-k events over the collected stream. Nodes closer to the sinks can then merge the topk streams into the final result. We use a hybrid chunk-based approach which alternates between full forwarding of events and pre-computed top-k streams at upstream nodes. This hybrid approach allows us to maintain correctness by recreating a possible top-k interleaving of the entire event stream. Our results show that our buffered solution exhibits significant traffic reduction in the overlay while maintaining latency comparable to that of the centralized approach. We also offer a sensitivity analysis which demonstrates the impact of various parameters on the performance of the solution. Experiments performed using real Twitter and Facebook datasets with 5K and 30K query subscriptions demonstrate that social workloads exhibit properties that are advantageous for our solution.

show abstract

Distributed Ranked Data Dissemination in Social Networks

Cited by 5 publications

References 31 publications

Language expressiveness and quality of service for publish/subscribe systems

Language expressiveness and quality of service for publish/subscribe systems

Towards Byzantine fault tolerant publish/subscribe

Distributed Ranked Data Dissemination in Social Networks

Contact Info

Product

Resources

About