You are on page 1of 1

Shared Memory

Mainframes and supercomputers deploy both shared and distributed memory architectures.
These computers vary widely where multiple processors may function separately but share a
common memory resource. Modifications on a memory location due to one processor are
available to all other processors. Shared memory devices are classified as UMA and NUMA,
depending on memory access times.
Uniform Memory Access
Has Identical processors
There is equal access to memory resources
Non-Uniform Memory Access
Occurs by physically connecting two or multiple SMPs
All processors do not necessarily have the same memory access time
Slow Memory access across connections
Advantages:
 Has user-friendly programming aspect to memory
 Fast and uniform data sharing between tasks
Disadvantages:
 lacks scalability between memory and CPUs.
 Programmer responsibility for synchronization constructs that ensure "correct" access of
global memory.
Distributed Memory
General Characteristics:
They vary widely and require a connection network to link inter-processor memory.
Processors have their dedicated memory available locally implying that they operate separately.
Modifications on the local memory have no effect on the memory modules of other processors.
They are devoid of cache coherency
Advantages:
 Highly scalable memory
 Processors access their local memory without interference and overhead
 It is cost effectiveness
Disadvantages:
 High involvement of the programmer in data communication among processors.
 Hard to map data structures
 Lacks a uniform memory access times

You might also like