A demand is a request for bandwidth between two nodes in the network.  In optical networks, we restrict these requests to fixed bandwidth of a limited selection of sizes.   No routing assumptions are made, although in many applications the normal path for a demand should be equal or close to the least-latency path through the network. 

Once a demand is assigned a working (normal) route through the network, it is called a routed-demand.  Each routed-demand is subject to a number of failures.  The goal of mesh restoration network survivability is to determine the spare resources needed to restore 100% of the affected demands after any one of the failure scenarios under consideration.  

We need to determine the basic unit of failure.  We want one failure to be independent of another, so upper-layer (IP) links are not always the best choice.  In long-haul networks, fiber spans are usually chosen, so a fiber cut in one span usually does not affect other spans.   The set of all single span failures describes the failure scenarios survived by most mesh restoration systems, although some also recover (as much as possible) from certain node failures.  

We've covered routed-demands and their failures.  When these are combined into a set of routed-demand-failures (RDFs), we next need to define the paths that can be used to recover a routed-demand once one of its failures occurs.