Distributed Memory Allocation Problem

This work attempts to study and offer a solution to an important resource allocation problem for explicitly parallel computing architectures. The problem of buffer space allocation in distributed on-chip memories is the key to lowering configuration (re-programming) overhead for systems such as SCORE [2] and enabling efficient virtualization of hardware resources essential for exploitation of such embedded platforms. The following section will provide some background into the problem and define the metrics necessary to evaluate its solution. The section will finish by reducing the buffer allocation problem into a more precise Multi-way Constrained Partitioning Problem that will be solved. Section 3 will show that the problem in question is NP-hard and then will describe four algorithms to solve it with different accuracy. The subsequent Section 4 will describe some implementation results.