摘要 |
The present invention, generally speaking, provides an efficient method of sending a long message from a first compute node to a second compute node across an interconnection network. In the first compute node, a message header field is set to a predetermined value and the message is sent. In the second compute node, the message header is received and processed, and a memory location is read in accordance with the contents of a base address register and an index register. Using Direct Memory Access, the message is then stored in memory at a storage address determined in accordance with the contents of the memory location. Preferably, the storage address is aligned on a memory page boundary.
|