STS supports three ways to generate dataplane traffic.
Random Traffic Generation
By default STS generates ICMP pings between random pairs of hosts in the network. The ICMP ping is injected at the access link of the source, and sent through the switches in the network.
In Fuzzer mode, the rate of packet generation is determined by the
traffic_generation_rate parameter in config/fuzzer_params.py. A value of 0.3 says that each host in the network has a 30% probability of injecting an IMCP ping to another host every fuzzer round.
In Interactive mode, the command
dp_ping with no arguments will generate a new packet, and
dp_ping <hid1> <hid2> will send a ping between host 1 and host 2.
See sts/traffic_generator.py for the details of how packets are generated.
Programmatic Traffic Generation
It is also possible to craft dataplane traces programmatically. sts/dataplane_trace/trace_generator.py contains several algorithms for generating synthetic ARP/ICMP exchanges between hosts. To invoke trace_generator from the command line, see tools/generate_dataplane_trace.py.
After a trace is generated with
trace_generator.py, you can cause STS to use that trace by specify the path to it in SimulationConfig. For example:
Note that dataplane traces are specific to the topology they were generated for; if you attempt to run a dataplane trace on a different topology, STS will raise an exception.
In Fuzzer mode, the rate of packet injection is determined by the
traffic_inject_interval, an argument to Fuzzer’s initializer. For example,
will cause Fuzzer to inject the next element in the trace every 5 rounds. Note that random traffic generation in Fuzzer mode is mutually exclusive with programmatic traffic generation; that is, if you specify a dataplane trace, no other packets will be injected.
In Interactive mode, the
dp_inject command will inject the next element in the given trace.
STS comes packaged with several dataplane traces, found in the dataplane_trace/ subdirectory.
The final and most powerful way to generate dataplane traffic is to make use of Linux’s network namespaces. Network namespaces provide us with a full networking stack that can respond to ARP/ICMP and run real networked applications such as Chrome or Apache Web Server. To use network namespace hosts in our simulated network, we just set a flag in the topology objection in our configuration file:
When we run
simulator.py with this configuration, we will be presented with one xterm window per host. From there, we can execute arbitrary bash commands, and the network namespaces will automatically respond to packets such as ICMP pings, just as a real host would.
Internally, the network namespace hosts are assigned a virtual interface each. STS reads and writes to these virtual interfaces to transfer packets to and from the simulated network.
Note that network namespaces require us to run
simulator.py as root.
It is worth noting that traffic generation is only necessary during the bug finding phase, since packets are logged for the purposes of replay.
Send questions or feedback to: email@example.com