Next: Virtual Shared Memory Computers Up: Parallel Computers Previous: Shared Memory Computers

Distributed Memory Computers

A distributed memory computer consists of multiple processors, each having its own memory. Because the processors can access only their own memory, they have to use for exchange of data communication network. This network can be a special high-speed network (e.g. high performance switch in IBM SP2 machine) or general purpose network as fast Ethernet or even in near future the Internet (so called P2P architecture). The performance of the network communication is given by the latency and the bandwidth. The latency is the time necessary to start an interaction between two processors and the bandwidth is the number of bytes that can be transfered via the network within one second. Since a distributed memory computer has no shared resources like a bus, the number of processors is virtually unlimited. The today three largest computers have thousand processors This is why distributed memory computers are called Massively Parallel Processors (MPP).

Next: Virtual Shared Memory Computers Up: Parallel Computers Previous: Shared Memory Computers

Daniel Rypl
2005-12-03