Big Network Simulation

BigNetSim takes the POSE timestamp correction simulation to the next level. Instead of using some preset latency value to determine message transit time, we actually model the message as it passes through a detailed contention-based network model. The power of this approach is that we could model any type of network we wish and plug it into the original timestamp correction simulation and get new results. This enables us to run the application emulation once, and reuse the trace logs generated by the emulation to repeatedly analyze the application in a variety of network configurations. Figure 4 shows how these components interact.

Figure 4: Interaction between BigSim, POSE timestamp correction and BigNetSim

For our network simulation, the nodes are configured with dimensions as originally given to the emulator. Each node has a switch which routes message packets through the network via six channels. Communication is achieved by first trying to lock a path of channels between switches from the source to the destination. Once a path is obtained, the switch sends out the packets on the appropriate outgoing channel. After the last packet is transmitted, the path is unlocked one channel at a time, following the last packet.

BigNetSim is currently parameterized by bandwidth, wraparound cost (for channels from edge and corner nodes to other edge/corner nodes), and maximum packetsize.