Binding Arrays:

many of the considerations made for stack-copying are valid also in the case of Binding Arrays. The execution model is quite similar, in the fact that the scheduler will be called upon backtracking in the public part of the tree (part that is shared between different agents), and it will eventually send a request to an agent for taking work from a new choice point. In the Binding Arrays case the operation of taking work is considerably faster, since the only activity required is updating the content of the binding arrays. On the other hand, the model has an additional overhead due to the fact that every conditional variable is now stored in binding arrays, and every access to it requires one step of indirection.

Additionally, the use of binding arrays requires that, during trailing, both the address of the variable and its current value are saved in the trail stack (which doubles the cost of trailing/untrailing).