TCE is well suited to support parallel programming. The applied host-node style allows supporting data parallel an functional parallel programming paradigms. |
The higher level communication masks differences between message passing systems on different distributed architectures. Thus, programs written for one collection of distributed nodes should be ran without any changes on a different collection of nodes. |
Shared memory, parallel machines can be naturally adopted through the multithreading interface. |
Multithreaded support allows an efficient parallel execution, by masking latencies and communication delays. |
A user definable scheduling policy as well as different communication styles on channels provide a way to optimally map an application onto the underlying architecture. |