Load management for publish/subscribe message oriented middleware

To provide time-critical early warnings in a Tsunami Warning System (TWS), the time delay for both the sensor data exchange process (upstream) and the warning message dissemination processes (downstream) should be minimal, maximising the time available for accurately analysing the situation and giving more time for people in the affected region to react to the warnings. Publish/Subscribe Message-oriented Middleware (PSMOM) in combination with a novel use of a federated broker (broker overlay) can be deployed in TWS to support both time-critical and resilient communication. PSMOM can better manage message bursts caused by a sudden increase in sensor data exchange frequency or by additional sensors coming online. PSMOM can better manage the decrease in available system resources (bottlenecks) caused by a disruption in the underlying network infrastructure (limited resource case). Otherwise, these burst and bottlenecks can cause some brokers to become overloaded, which may in turn degrade the overall system performance and delay decision-making. Existing PSMOM load management solutions have two key limitations when applied to TWS. First, existing work does not consider the message delay requirements for the redistribution and offloading phases of load management. Here, some data is only useful or valid for a short time-span (from tens of seconds to tens of minutes); hence, it needs to be exchanged within this maximum allowed end-to-end transmission delay. Time critical subscribers need to be de-prioritised from being offloaded, as the offloading processes take some time to complete, introducing unexpected delays to message exchange. Second, existing solutions assume that there are surplus system resources for offloading, i.e., less loaded brokers can accept loads from overloaded brokers. However, in a TWS, the underlying network infrastructure may be disrupted which in turn reduces the system capacity. It always takes time to recover from the limited resources situation and during that time, brokers may not have enough system resources to accept loads from overloaded brokers, which may result in total failure of the overloaded brokers. A novel load management framework called ePEER is proposed that extends an existing messaging system, Publish/Subscribe Efficient Event Routing (PEER), with the following main contributions. First, for the surplus resource case, the message delay requirements for different subscription services are considered in the load analysis process when offloading the load to different brokers. Second, for the limited resource case, a feedback driven congestion control mechanism can be used when the underlay network infrastructure is damaged, reducing the available bandwidth of PSMOM. This mechanism limits the publication rate of messages with less value, to better maintain the quality of experience (QoE) of subscribers for the more important messages ePEER is validated with emulation-based experiments. The results show that ePEER outperforms the state of the art load management solution used by PEER: through preventing unnecessary delays introduced to time critical services, and through ensuring important messages can be more efficiently exchanged to improve the QoE of subscribers.