Modeling Node Failures¶
Simulation is often used to study the effects of unexpected events like a router crash on the network. In order to accommodate such scenarios, INET supports lifecycle modeling of network nodes. The up/down status of a node is changed via lifecycle operations.
INET supports the following lifecycle operations:
Startup represents the process of booting up or starting a network node after a shutdown or crash operation.
Shutdown represents the process of orderly shutting down a network node.
Crash represents the process of crashing a network node. The difference between crash and shutdown is that for a crash, the network node will not do a graceful shutdown (e.g. routing protocols will not have a chance of notifying peers about broken routes).
In a real-life router or other network node, a crash or shutdown and subsequent restart affects all parts of the system. All non-persistent information is lost. Protocol states are reset, various tables are cleared, connections are broken or torn down, applications restart, and so on.
Mimicking this behavior in simulation does not come for free, it needs to be explicitly programmed into each affected component. Here are some examples how INET components react to a crash lifecycle event:
Tcp forgets all open connections and sockets
Ipv4 clears the fragmentation reassembly buffers and pending packets
Ipv4RoutingTable clears the route table
EtherMac and other MAC protocols clear their queues and reset their state associated with the current transmission(s)
Ospfv2 clears its full state
While down, network interfaces, and components in general, ignore (discard) messages sent to them.
Lifecycle operations are currently instanteneous, i.e. they complete in zero simulation time. The underlying framework would allow for modeling them as processes that take place in some finite (nonzero) simulation time, but this possibility is currently not in use.
It also is possible to simulate a crash or shutdown of part of a node (certain protocols or interfaces only). Such scenarios would correspond to e.g. the crash of an OSPF daemon on a real OS.
Some energy-related INET components trigger node shutdown or crash under certain conditions. For example, a node will crash when it runs out of power (e.g. its battery depletes); see the chapter on power consumption modeling Modeling Power Consumption for details.
In the following sections we outline the INET components that participate in lifecycle modeling, and show a usage example.
Node models contain a NodeStatus module that keeps track of the status of the node (up, down, etc.) for other modules, and also displays it in the GUI as a small overlay icon.
The NodeStatus module is declared conditionally (so that it is only created in simulations that need it), like this:
status: NodeStatus if hasStatus;
If lifecycle modeling is required, the following line must be added to the ini file to ensure that nodes have status modules:
**.hasStatus = true
Lifecycle operations can be triggered from C++ code, or from scripts. INET supports scripting via the ScenarioManager NED type, described in chapter Scenario Scripting. Here is an example script that shuts down a router at simulation time 2s, and starts it up a again at time 8s:
<scenario> <initiate t="2s" module="Router2" operation="shutdown"/> <initiate t="8s" module="Router2" operation="startup"/> </scenario>
module attribute should point to the module (host, router,
network interface, protocol, etc.) to be operated on. The
operation attribute should contain the operation to perform:
t is the
simulation time the operation should be initiated at.
An alternative, shorter form is to use
<startup> elements instead of the
<scenario> <shutdown t="2s" module="Router2"/> <startup t="8s" module="Router2"/> </scenario>