Self-Evolving Subscriptions for Content-based Publish/Subscribe Systems

CÚsar Ca˝as, Kaiwen Zhang, Bettina Kemme, J÷rg Kienzle, and Hans-Arno Jacobsen.

MSRG, McGill University, 2016.


Traditional pub/sub techniques such as subscription covering are inadequate to fully handle workloads of applications with high subscription churn where new subscriptions follow a predictable pattern, such as some social location-based notification systems, predictive stock trading, and online games. In the context of those applications, the main unaddressed factor of traditional pub/sub techniques is that they have to wait for subscriber input before replacing older subscriptions with updated ones, creating overhead and timing issues.

In this paper we present a new type of subscription, called evolving subscription, that is able to autonomously adapt to the predicted needs of a subscriber. We propose a general framework for evolving subscriptions and discuss their advantages, limitations, and the type of applications they are fit to support. For this end, we develop and implement three different variations of evolving subscriptions, which are then evaluated and compared to the traditional re-subscription approach in the context of two use cases: online games and high-frequency trading. Our evaluation shows that our solutions reduce subscription traffic by 96.8% and improves delivery accuracy by 95.3% compared to the baseline unsubscription and re-subscription mechanism.


Tags: content-based publish/subscribe, online games, distributed systems

Readers who enjoyed the above work, may also like the following: