Private versus shared memory

Private and shared data are differentiated by their accessibility and by the physical memory classes in which they are stored.

thread_private data is stored in node-local memory. Access to thread_private is restricted to the declaring thread.

When porting multinode applications to the HP single-node machine, all legacy shared memory classes (such as near_shared, far_shared, and block_shared) are automatically mapped to the node_private memory class. This is the default memory class on the K-Class and V-Class servers.

`thread_private`

thread_private data is private to each thread of a process. Each thread_private data object has its own unique virtual address within a hypernode. This virtual address maps to unique physical addresses in hypernode-local physical memory.

Any sharing of thread_private data items between threads (regardless of whether they are running on the same node) must be done by synchronized copying of the item into a shared variable, or by message passing.




	NOTE: `thread_private` data cannot be initialized in C, C++, or in Fortran `DATA` statements.

`node_private`

node_private data is shared among the threads of a process running on a given node. It is the default memory class on the V-Class single-node server, and does not need to be explicitly specified. node_private data items have one virtual address, and any thread on a node can access that node’s node_private data using the same virtual address. This virtual address maps to a unique physical address in node-local memory.

Private versus shared memory

Technical documentation

» Table of Contents

» Glossary

`thread_private`

`node_private`