Distributed global memory

Distributed Global Memory A hardware and/or software mechanism for treating physically distributed memory as a global shared memory in the context of application development. 

Parallel programming for shared memory computers is relatively straightforward, as all processors can access any location in the common memory. Processor access to the global memory is coordinated by a switch. The complexity of this switch increases rapidly with the number of processors, resulting in longer memory access times, and shared memory machines therefore usually consist of a small number of processors, typically less than 100. In distributed memory computers, there is, in principle, no such bottleneck, as communication networks are constructed in such a way that communication between one pair of processors is typically independent of communication between another pair. 

J. Nieplocha, R.J. Harrison and R.J. Littlefield, Global arrays A portable "shared-memory" programming model for distributed memory computers, in Supercomputing 94 (Washington D.C., 1994). 

Among the current limitations of Linda, those important to computational chemistry applications are failure to provide information on where or how tuples are stored or accessed lack of structure within tuple space, making it hard to maintain modularity a requirement to match general tuples, leading to inefficiencies in memory usage and communication even for simple data structures (e.g., a distributed array) lack of primitives for efficient global operations (e.g., reduction and broadcast) and requiring the compiler to detect and optimize these constructs. There are many current directions of related research. 

Das and Saltz developed a fully distributed data parallel version of CHARMM. 1 8 jhe implementation used Saltz and co-workers PARTI (Parallel Automated Runtime Toolkit at ICASE) primitives, These primitives allow the distribution of arrays across the local memories of multiple nodes and provide global addressing of these arrays even on truly distributed memory machines. Whereas the replicated-data version of Brooks and Hodosceki ... 

MIMD parallel computers are usually divided into shared and distributed memory types. In the shared memory case, multiple processors are connected to memory by a switch so that any processor can access any memory location, and all processors have access to the same global name space. Thus, when each processor refers to the variable x they are all refering to the same location in the shared memory. The shared memory approach to parallelism is attractive because in porting an application from a sequential to a shared memory parallel computer usually only a few changes are required in the sequential... 

---