Time synchronization in the DHT-based P2P network Kad for real-time automation scenarios

Abstract

In this paper, an approach to synchronize the P2P network Kad to be applied in automation scenarios is presented. The approach bases on a deterministic algorithm to synchronize the network, which is required for hard real-time applications. Todays Industrial Ethernet solutions include and support machine to machine communication in automation scenarios. However, instead of allowing direct communication, the communication relies on centralized structures, which are deficient in resilience and scalability However, instead of allowing direct communication, the communication relies on centralized structures. The presented decentralized approach benefits from nodes helping to synchronize the network. However, the higher the number of helping nodes the higher is the time deviation on the nodes of the network, which contrary results in a higher time error. Therefore, a trade-off between synchronization performance and time error has to be determined to meet predefined constraints depending on the application scenario. Moreover, the individual clock drift of every device is considered to define necessary re-synchronization intervals of the network. Additionally, the optimum number of nodes to synchronize the Kad based network has been identified and the resulting synchronization performance and generated traffic are determined. Furthermore, an approach is presented to handle the dynamic churn of nodes.

Dynamic search tolerance at runtime for lookup determinism in the DHT-based P2P network Kad

“…Therefore, the algorithms for ST determination can be integrated in the Kad-based network services proposed in [22], [19], [23], [12].…”

“…The shorter the periods are chosen, the more computation and communication effort is required to determine all nodes, calculate the ST, and disseminate it. However, in a stable Kad network consisting of, e.g., automation devices [12] the period can be set to a few hours without risking a ST being out of date.…”

“…Optimized DST Dissemination: This time complexity can even be reduced to O(log 2 N ) by determining nodes assisting in the ST dissemination (as done for the time synchronization in [12]). Thereby, all nodes receive the ST value shortly after its calculation.…”

“…As depicted in Figure 5, with 50,000 nodes we are still below 30 hops, which is about 7.5 ms if we assume that the duration of a hop, i.e., the transmission of a message in the Kad network and the processing time on a node amounts to a total of 250 µs. This value has been derived from an experimental setup comprising prototypes running on ZedBoards, which are connected by a 1 Gbit/s Ethernet switch [20], [12]. The number of hops needed to collect all nodes and to send the information back to the origin node is depicted together with the time needed for marking the nodes.…”