Self-stabilizing Publish/Subscribe Systems: Algorithms and Evaluation

Abstract

Most research in the area of publish/subscribe systems has not considered fault-tolerance as a central design issues. However, faults do obviously occur and masking all faults is at least expensive if not impossible. A potential alternative (or sensible supplementation) to fault masking is self-stabilization which allows a system to recover from arbitrary transient faults such as memory perturbations, communication errors, and process crashes with subsequent recoveries. In this paper we discuss how publish/subscribe systems can be made selfstabilizing by using self-stabilizing content-based routing. When the time between consecutive faults is long enough, corrupted parts of the routing tables are removed, while correct parts are refreshed in time, and missing parts are inserted. To judge the efficiency of self-stabilizing content-based routing, we compare it to flooding, which is the naïve implementation of a self-stabilizing publish/subscribe system. We show that our approach is superior to flooding for a large range of practical settings.

On adopting Content-Based Routing in service-oriented architectures

“…In both cases the CBR network must be able to adapt to such situations. Again, some proposals already exist to solve this issue, like those on self-stabilizing CBR [50,64,75], those that suggest using a peer-to-peer overlay substrate to exploit its self-organizing capabilities [6,39,70,68,59], and our own works in the area [26,23,21,24]. Unfortunately, all these approaches still present some limitations in terms of performance.…”

Reconfiguring Self-stabilizing Publish/Subscribe Systems

“…As starting basis, we build on a model for self-stabilizing pub/sub systems developed in previous work [8]. Basically, we assume a hierarchical routing algorithm based on an acyclic broker topology with bidirectional FIFO links connecting individual nodes.…”

“…Such a pub/sub system works correctly, if it meets the following two requirements [8]: (i) every client receives only the published notifications it has subscribed for (without duplicates) and (ii) every subscription becomes active after finite time, from which on the client receives every published notification matching its subscription until it unsubscribes.…”

“…Therefore, we rely on the leasing mechanism described earlier [8]: clients regularly refresh their subscriptions and brokers use a second chance algorithm to remove stale entries from their routing tables. To incorporate reconfigurations into this mechanism, we demand that control messages are colored with c new , while notifications are colored with c cur .…”

“…Self-stabilization is an elegant mechanism for gaining fault tolerance. However, the solutions presented for pub/sub routing do not yet explicitly deal with reconfiguration [8,10], although managing these systems has to include reconfiguration of the broker overlay topology. In many self-stabilizing systems, reconfigurations are treated as faults and the system tries to "recover" from them.…”