Recently, we studied the feasibility of a GLORI implementation. A single-pass, static optical geometry composed of elementary guided and free-space holographic optical components was analyzed. A number of issues determine the feasibility of the GLORI system. Some issues indirectly impact feasibility, such as those relating to the holographic beam steering element. Other issues, such as those relating to optical geometry, directly impact how the system size can be feasibly implemented due to fan-out and fan-in limitations.
A multiprocessor system
of 
processor-memory nodes can be feasibly implemented with a
static optical shared bus-hypercube GLORI network.
By improving aspects of the optical system which most
limit scalability, systems on the order of thousands of nodes
can be built.
The volume of GLORI's shared bus-hypercube optical routing unit
was found to be about one cubic centimeter,
which is far less than comparable electronic
hard-wired implementations.
The tradeoffs in implementing various GLORI strategy communication models
in terms of resource count, implementation complexity,
and latency reduction were also assessed.
Somewhat of an unexpected finding is that GLORI's block-frequent reconfiguration policy is not a general solution for improving performance. Block-frequent reconfiguration is shown to prove useful when it reduces global reference activity to levels that do not saturate the global network. In other words, if there is not sufficient cluster locality, reconfiguration is unlikely to reduce network latency.