Connection types

The network adapter PCI functions have modes for two connection types: IP-based connections and RDMA-based connections.

Both connection types can be used concurrently on the same PCI function. The connection type depends on the socket application and how it is started, see Using a PCI function for SMC-R connections.

The adapter port of an operational PCI function must be connected to an external LAN. Even so, for both connection types communication peers on the same hardware system that share a PCI network adapter port can communicate through the adapter without routing packets through the LAN.

TCP/IP connections

You can set up connections to any communication peer that you can reach with TCP/IP or other IP-based protocols. The network adapter of the communication peer can be a PCI network adapter, but it can also be a different adapter.

In particular, possible peers include z/OS®, z/VSE®, and Linux® instances that are reached through an OSA adapter.

SMC-R connections

Shared Memory Communications Remote (SMC-R) uses remote direct memory access (RDMA) over Converged Ethernet for high-throughput, low-latency connections.

Both communication peers must support SMC-R connections.

A regular TCP/IP connection between the peers must be in place to initiate an SMC-R connection. On both sides of the connection, the PCI function must be associated with the TCP/IP network device, see Using a PCI function for SMC-R connections.

Linux on IBM Z® can use SMC-R connections to instances of Linux or of z/OS. On Linux, the initial TCP/IP connection can use any available network adapter, including the PCI network adapter that is also used for the SMC-R connection.

Extra handshaking in the protocol incurs more initial latency for establishing an SMC-R connection than for opening a TCP/IP connection. Once established, the SMC-R connection can combine superior data throughput with low latency and reduced CPU consumption. Expect best results for multiple concurrent and long-lasting connections for transferring large amounts of data.