| 1 |
|
|---|
| 2 | /** \page group The Group Library
|
|---|
| 3 |
|
|---|
| 4 | The group library provides three mechanisms for parallel execution and
|
|---|
| 5 | communication. These are:
|
|---|
| 6 |
|
|---|
| 7 | <ul>
|
|---|
| 8 | <li> \ref messagepassing
|
|---|
| 9 | <li> \ref distmemory
|
|---|
| 10 | <li> \ref multithreading
|
|---|
| 11 | </ul>
|
|---|
| 12 |
|
|---|
| 13 | \section messagepassing Message Passing
|
|---|
| 14 |
|
|---|
| 15 | The MessageGrp class is the ancestor for classes that provide the passing
|
|---|
| 16 | of messages between processes. There are three important specializations
|
|---|
| 17 | of this class:
|
|---|
| 18 |
|
|---|
| 19 | <dl>
|
|---|
| 20 |
|
|---|
| 21 | <dt>ProcMessageGrp<dd>Provides a dummy MessageGrp for single processor
|
|---|
| 22 | case.
|
|---|
| 23 |
|
|---|
| 24 | <dt>MPIMessageGrp<dd>Implements MessageGrp using the Message Passing
|
|---|
| 25 | Interface (MPI) library.
|
|---|
| 26 |
|
|---|
| 27 | <dt>ShmMessageGrp<dd>Implements MessageGrp using SysV Interprocess
|
|---|
| 28 | communication.
|
|---|
| 29 |
|
|---|
| 30 | </dl>
|
|---|
| 31 |
|
|---|
| 32 | Due to the widespread acceptance of MPI and the large number of features
|
|---|
| 33 | that it supports, the MessageGrp specialization may one day be eliminated.
|
|---|
| 34 |
|
|---|
| 35 | \section distmemory Distributed Shared Memory
|
|---|
| 36 |
|
|---|
| 37 | The MemoryGrp class is the ancestor for classes that permit access to the
|
|---|
| 38 | memory in different processes, possible on difference machines. There are
|
|---|
| 39 | two important specializations of this class:
|
|---|
| 40 |
|
|---|
| 41 | <dl>
|
|---|
| 42 |
|
|---|
| 43 | <dt>ProcMemoryGrp<dd>Provides a dummy MemoryGrp for single processor
|
|---|
| 44 | case.
|
|---|
| 45 |
|
|---|
| 46 | <dt>MTMPIMemoryGrp<dd> This works reliably and efficiently, however,
|
|---|
| 47 | requires POSIX threads and a thread-safe MPI
|
|---|
| 48 | implementation.
|
|---|
| 49 |
|
|---|
| 50 | <dt>ARMCIMemoryGrp<dd> This message group uses the Aggregate Remote Copy
|
|---|
| 51 | Interface (ARMCI). This exploits the RDMA
|
|---|
| 52 | capabilities of interconnects such as Myrinet and
|
|---|
| 53 | InfiniBand.
|
|---|
| 54 |
|
|---|
| 55 |
|
|---|
| 56 | </dl>
|
|---|
| 57 |
|
|---|
| 58 | \section multithreading Multi-Threading
|
|---|
| 59 |
|
|---|
| 60 | The ThreadGrp class is the ancestor of classes that provide
|
|---|
| 61 | multiple-threads within a single address space. There are
|
|---|
| 62 | two important specializations of this class:
|
|---|
| 63 |
|
|---|
| 64 | <dl>
|
|---|
| 65 |
|
|---|
| 66 | <dt>ProcThreadGrp<dd>Provides a dummy ThreadGrp for the single thread
|
|---|
| 67 | case.
|
|---|
| 68 |
|
|---|
| 69 | <dt>PthreadThreadGrp<dd> This is a ThreadGrp that uses POSIX threads.
|
|---|
| 70 |
|
|---|
| 71 | </dl>
|
|---|
| 72 |
|
|---|
| 73 |
|
|---|
| 74 | */
|
|---|
