Research On High Performance Remote Routing Strategy In Large-Scale And Parallel Network Simulation

Network simulation has become one of the main methods in researching and studying network. Its great value in science research and application has been gradually reflected. However, with the increasing scale, complexity and heterogeneous attribute of topology, the simulation is getting more and more complicated. This is mainly reflected in the computing recourses of the large-scaled, complicated and heterogeneous attribute topology, which makes it hard to simulate on the original single simulator. So, for solving this problem, parallel network simulation was presented and widely used. The main aspects which affect simulation performance are topology partition, routing records and lookups, synchronous performance and scheduler algorithm. This paper mainly focused on the restoring and computing of remote routing.There are two key aspects in remote routing: the scale of routing tables and forwarding lookups. The problem of records is the high amount and redundancy which occupies lots of memories and makes it unavailable to simulate large-scale topologies. Problems in forwarding lookup are high time complexity, unnecessary computing and invalid lookup for a destination with a non-existent IP address.Focused on these problems, this paper presents a new remote-routing scheme called OERT, which is the abbreviation for Optimization-Edge Based Remote Routing Strategy. It converts the original method of using destination IP address to the method of using the edge router in the subnet that the destination belongs to, which reduces the redundancy and lows down time complexity in lookup per forwarding packet. Meanwhile, it adopts three effective ways to reduce memory footprint, which are Tree-Nodes Contraction, Duplicates Removing of Successor Nodes and Duplicates Removing of Edge Routers. In this way, the routing table tables decreases a lot and the memory footprint lows down apparently.Comparing with Border-Based method, the evaluation results in PDNS using OERT shows that this scheme behaves better in terms of memory usage and computation time with 75% less time cost and 85% less memory used.