Upon certain network events, such as node or link failures, IP routers need to update their affected routing table entries. During the period between the failure occurrence and the installation of the updated entries (on the line cards), the network traffic is lost when forwarded by routers that are still using old entries. Indeed, current IP routers do not involve network traffic information during this unordered update process. The consequence is more packet losses compared to a process that would order these entries based on local traffic information. In this paper, we model and predict network traffic passing through an IP router and define two dynamic heuristics in order to reduce the packet loss resulting from routing table updates. AutoRegressive Integrated Moving Average (ARIMA)-Generalized AutoRegressive Conditional Heteroskedasticity (GARCH) traffic models are used in combination with heuristics that dynamically sort the routing entries and improve the low-level routing table update process. In a realistic simulation environment, we show that this setup can result into a clear decrease of packet loss, depending on i) the network traffic model, ii) the applied heuristic, and iii) the network traffic aggregation level.
展开▼
