Remote direct memory access: Difference between revisions

Revision as of 04:32, 29 May 2005

Revision as of 16:00, 16 June 2005

Line 1:

'''Remote Direct Memory Access''' ('''RDMA''') is a concept whereby two or more [[computer]]s communicate via [[Direct Memory Access]] directly from the [[main memory]] of one system to the main memory of another. As there is no CPU, cache, or context switching overhead needed to perform the transfer, and transfers can continue in parallel with other system operations, this is particularly useful in applications where high thoughputthroughput, low latency networking is needed such as in massively parallel [[Linux]] clusters. The most common RDMA implementation is over [[InfiniBand]]. Although RDMA over InfiniBand is technologically superior to most alternatives, it faces an uncertain commercial future.

== RDMA over TCP/IP ==

An alternate proposal is RDMA over [[TCP/IP]], in which the TCP/IP protocol is used to move the data over a commodity data networking technology such as [[Gigabit Ethernet]]. Unlike conventional TCP/IP implementations, the RDMA implementation would have its TCP/IP stack implemented on the network adapter card, which would thus act as aan [[I/O processor]], taking up the load of RDMA processing.

This also has the advantage that software-based RDMA emulation will be possible, allowing interoperation between systems with dedicated RDMA hardware and those without. One example of this might be the use of a server with a hardware-equipped RDMA to serve a large number of clients with software-emulated RDMA implementations.