High-performance routing (HPR) is the next evolution of Advanced Peer-to-Peer Networking® (APPN). HPR is an extension of APPN and has many functional aspects common with APPN. Configuring adjacent stations, search processing, and route computation are the same in APPN and HPR. HPR differs from APPN in the areas of transport, intermediate session routing, congestion control, and error recovery.
The following are the HPR protocol operational characteristics:
HPR supports a key availability enhancement that is called non-disruptive path switching. This function provides the ability to recover from link or node outages without having session failures. This makes the outage transparent to the application. The application may experience a response time delay while data traffic is being rerouted. On iSeries™, the amount of time the system takes to establish a new path, or re-establish the original failed route path is configurable. This error recovery feature is the key difference between APPN and HPR.
HPR can support the non-disruptive path switching feature because of an enhanced data transport mechanism that is called Rapid Transport Protocol (RTP). RTP is the data transport protocol that is used between a pair of systems that support the HPR RTP tower. This pair of systems establishes an RTP connection which carries out APPN sessions (multiple APPN sessions can be multiplexed over a single RTP connection). In order to establish an RTP connection between a pair of HPR RTP tower systems, the following must be true:
This routing is known as Automatic Network Routing (ANR).
When an RTP node sends packets of data, it must keep those buffers until the RTP node receives acknowledgment that its RTP partner has successfully received the data. Maintaining detailed knowledge of data sent and received is necessary in order to provide the additional value provided by HPR, the non-disruptive path switching function. HPR does not rely on the data link layer to provide data retransmission functions. HPR supports a function that is called selective retransmission. With selective retransmission, only the data which has not been acknowledged gets retransmitted. For example, if an RTP node sends eight packets and all but the fourth packet is successfully acknowledged, then only the fourth would retransmit. This differs from other retransmission algorithms in which the first unsuccessful packet and all the subsequent packets would transmit.
Nodes performing intermediate routing of HPR traffic or ANR, have no session awareness. HPR uses source routing. The nodes performing ANR examine packets as they receive them and determine the next hop of the route. The next hop is based on something that is called the ANR label. All HPR packets contain the ANR label. Any ANR that a network node is performing does not count as an APPN intermediate session. The maximum intermediate sessions parameter that is configured by means of the Change Network Attribute (CHGNETA) command has no effect on the ANR capacity of a system. Controlling the amount of ANR that different systems will perform in a network is completely dependent on the route selection phase of APPN session establishment.
When sessions are carried over RTP connections, any segmentation, or reassembly is performed within the iSeries central processing unit (CPU). The communications IOPs do not have the information required to perform the segmentation and reassembly. The IOPs can not maintain the knowledge that is required by HPR to perform the data retransmission and non-disruptive path switching function.
HPR uses a function called Adaptive Rated Based (ARB) congestion control. ARB regulates the flow of traffic by predicting congestion in the network and reducing the sending rate of a node into the network. ARB then attempts to prevent congestion rather than reacting to it after it has occurred. If all of the traffic occurring over a network was HPR, then ARB would be a fair way of sharing the bandwidth of a network. ARB also allows high utilization of the networking resources. When HPR traffic mixes with straight APPN or TCP/IP traffic, the HPR throughput may suffer because the other protocols do not practice similar congestion control techniques.
For more information about configuring HPR, see Configure APPC, APPN, and HPR.