Nielsen and Chuang are referring to a scheme known as measurement-based quantum computation, for which many good learning resources are a short google search away.
The idea is that you prepare a large, highly entangled state involving a large number of qubits at the start of the experiment. You then proceed to make measurements on the state, potentially changing which measurements you perform based on the results of previous measurements. The key strength of this scheme is that every measurement is local to a single qubit. There is then no need to perform entangling operations once the state has been prepared.
This is particularly important in photonic quantum computing, because it is very, very hard to entangle photons, and at the moment this can only be done probabilistically. For photons, you can implement the two-qubit entangling gates required for universal quantum computing, but they will only work a fraction of the time. (Nowadays they tend to be heralded, which means that you know whether they worked or not, but this does not really solve the problem.) This scales very badly, because you need all your gates to work, and even a 95% success rate per gate looks terrible if you have more than a few gates.
For MBQC, on the other hand, you can try repeatedly to get a resource state that is entangled enough, and once you have it you don't have to generate more entanglement - you just have to do local measurements on each of your photons (plus potentially some classical postprocessing and feedback).