Performance of Synchronous and Asynchronous Schemes for VLSI Systems
IEEE Transactions on Computers
Computer architecture (2nd ed.): a quantitative approach
Computer architecture (2nd ed.): a quantitative approach
Advanced Computer Architecture: Parallelism,Scalability,Programmability
Advanced Computer Architecture: Parallelism,Scalability,Programmability
Computer Architecture and Parallel Processing
Computer Architecture and Parallel Processing
Hi-index | 0.01 |
When computing systems with a large number of processors are designed, the exclusive use of a common memory may lead to a bottleneck, slowing down not only the communication between processors, but also their independent execution. The NUMA (Non-Uniform Memory Access) concept is seen as a solution to that problem. In order to implement a NUMA architecture, it is necessary to create a protocol and an arbiter for an easy access to the global memory. The paper presents such a protocol and the automaton that implements the arbiter. It is also necessary to align to the standard protocols, if possible. That is the reason why the system is similar to the PCI protocol. All transfers are achieved on the same address/data bus, strobed by the command lines. The system consists of n cells - a cell consisting in a processor and its local memory - and the global memory, easily accessed by all cells for communication purposes.