A DDS Discovery Protocol based on Bloom filters

The DDS middleware (Data Distribution Service) has being used for addressing the data distribution and interoperability between different networks, architectures and operating systems. A essential point in the integration of these heterogeneous environments relies on the discovery process. A DDS discovery protocol matches publisher and subscriber entities (Endpoints) placed in different network nodes. The scalability of a DDS infrastructure is limited by the discovery protocol, hence the discovery process is one of the most important features of a DDS implementation. This master project address the design and evaluation of alternatives to the current DDS discovery protocol, the Simple Discovery Protocol (SDP). These alternatives should improve the scalability of the SDP protocol for allowing the use of DDS in larger scenarios. In SDP, the number of sent messages and the consumed network bandwidth grow with the number of Endpoints. Additionally, each node consumes memory for storing all the remote node’s discovery information. Basically, the total number of messages sent over the network is P ∗ E where P is the number of Participants and E the number of Endpoints with E P . We propose to improve the DDS discovery process by using a compact information representation technique called Bloom filter. We propose to modify the SDP for using Bloom filters. We call this proposal SDPBloom. By using Bloom filters the number of messages sent over the network is closer to P ∗ P instead of P ∗ E. Additionally, both the spent network bandwidth consumption and nodes memory usage decrease. The analytical results show the improvement of SDPBloom over SDP. In addition, we have implemented a test application for measuring the real advantages of using Bloom filters for representing discovery information. The analytical and experimental results show that the number of messages sent to the network for announcing Endpoints still constant with no dependency with the Endpoints number (E). As it will be show, under some circumstances the proposed approach can provide a discovery information compression ratio equal to 50 : 1.