data flows as well. Scalability Ensures Systems Can Be Extended and Federated With data-centric publish-subscribe, there is no need to maintain N-squared network connections like there is with message or connection centric protocols. As a result DDS based Figure 3 systems can scale to over 10 milWith DDS, automatic discovery matches publishers and subscribers of a data topic and Quality of Service parameters lion publish-subscribe pairs, and describing application data needs shape the resulting data flows. Specifying and relying on data-centric properties enables decoupled, resilient systems. application code can be reduced by a factor of 10. Since DDS does not assume allows nodes to subscribe to many, possibly thousands, of similar data a reliable underlying transport, it can easily take advantage of streams with a single subscription. When the data arrives, the midmulticasting. With multicasting, a single network packet can be dleware can sort it by the key and deliver it for efficient processing. sent simultaneously to many nodes, greatly increasing throughput DDS also provides a “state propagation” model. This model and scale. Most client-server designs, by contrast, cannot handle allows nodes to treat DDS-provided data structures like distributed a client sending to multiple potential servers simultaneously. In shared-memory objects, with local caches efficiently updated only large networks, multicasting greatly increases throughput and when the underlying data changes. There are facilities to ensure reduces latency. coherent and ordered state updates. Much like the security in a database where data is secured table DDS is fundamentally designed to work over unreliable transports, by table, DDS allows data to be secured topic by topic. The system such as UDP or wireless networks. No facilities require central servers integrator specifies which DDS applications have read or write or special nodes. Efficient, direct, peer-to-peer communications, or permission for which topics in the system. Only those data topics even multicasting, can implement every part of the model. that need to be made confidential should be encrypted. Topics DDS does not require a central server, so implementations can that do not need to be secured can be left unencrypted, allowing use direct peer-to-peer, event-driven transfer. This provides the higher performance, and signed with a message authentication shortest possible delivery latency. This is a significant advantage code (MAC), to verify authenticity. This fine-grained security over client-server or broker-based designs. Central servers or model helps mitigate malicious insider attacks by limiting the brokers impact latency in many ways. At a minimum, they add an access of a compromised application or a malicious user. intermediate network “hop,” nominally doubling the minimum The Industrial Internet Consortium’s Microgrid Testbed peer-to-peer latency. If the server is loaded, that hop can add very program applies the cross-industry Industrial Internet protocol significant latency. Client-server designs do not handle inter-client DDS, intelligent edge controllers and industrial network equiptransfers well; latency is especially poor if clients must “poll” for ment to implement a communication and control architecture for changes on the server. microgrid applications. Long term, the real power of the IndustriFine control over real-time QoS is perhaps the most importal Internet of Things (IIoT) is to connect sensor to cloud, power ant feature of DDS. Each publisher-subscriber pair can establish to factory, and road to hospital. To do that, we must change core independent QoS agreements. Thus, DDS designs can support infrastructure to use generic, capable networking technology that extremely complex, flexible data flow requirements. can span industries, field systems and the cloud. Applying the Periodic publishers can indicate the speed at which they can IIoT to microgrids is a key step to enable large-scale efficient use publish by offering guaranteed update deadlines. By setting a of green energy. deadline, a compliant publisher promises to send a new update at a minimum rate. Subscribers may then request data at that or any Real-Time Innovations slower rate. Publishers may offer levels of reliability, parameterSunnyvale, CA ized by the number of past issues they can store to retry trans(408) 990-7400 missions. Subscribers may then request differing levels of reliable www.rti.com delivery, ranging from fast-but-unreliable “best effort” to highly reliable in-order delivery. This provides per-data-stream reliability control. The DDS publish-subscribe model provides fast location transparency. That makes it well suited for systems with dynamic configuration changes. It quickly discovers new participants and data topics. The system cleanly flushes old or failed nodes and
RTC Magazine OCTOBER 2015 | 27