Transport protocol for a high availability manager

The high availability manager network components can be built on either a channel framework, unicast or multicast transport protocol. Channel framework is the default protocol, but there are options and features available with multicast that might be a better match for your application server environment. Multicast emulation can be used with any of these protocols.

The following network protocol configurations can be used with WebSphere Application Server:

Channel framework

Channel framework is the default network protocol configuration for the high availability manager. It provides a common model for connection management, thread usage, channel management, and message access within WebSphere Application Server. It extends the concept of a networking protocol stack, or transport chain, to the WebSphere run time. Each transport chain consists of one or more types of channels, and each channel supports a different type of I/O protocol, such as TCP, DCS or HTTP.

The transport chain configuration settings determine which I/O protocols are supported for that chain. Custom channels that support requirements unique to a particular customer or environment can also be added to a transport chain.

Network ports can be shared among all of the channels within a chain. The channel framework function automatically distributes a request arriving on that port to the correct I/O protocol channel for processing. However, this flexibility comes at a cost because the channel framework is slower than either multicast or unicast in terms of CPU overhead and maximum bandwidth.

Channel Framework must be used whenever SSL or HTTP tunneling is required for the high availability manager message transport.

Unicast protocol

Unicast protocol uses only native TCP/IP to send and receive high availability messages. Its performance is similar to that of the multicast protocol. However, it is almost twice as fast as using the channel framework protocol. This performance difference should only be a factor if you are doing memory to memory session replication.

Both channel framework and unicast protocols can be used when a core group is not located on a single subnet or when multicast is not appropriate to use. Most messages sent over the high availability manager messaging system have a low fan out. Therefore, under normal circumstances, unicast protocol is the preferred transport when the flexibility provided with the channel framework protocol is not required and performance is important.

Multicast protocol

Multicast protocol usually requires that all members of a core group are located on a single subnet. The same level of performance is normally achieved using either the multicast or the unicast protocol.

Host names specified in the DCS_UNICAST_PORT field

For both channel framework and unicast protocols, a single IP address must be specified for the host name. Host names are not acceptable unless they resolve to a single IP address on all of the servers that are members of the core group.

The high availability manager can not listen on all IP addresses for a host name. If a host name is specufued that can resolve to multiple IP addresses, it is possible for a core group to become partitioned into two or more groups of servers. If this partitioning occurs, the number of servers that message HMGR0218I indicates are running will be less then the number of servers that were started.

A single IP address should also be specified for the multicast framework protocol. This IP address identifies the network card the high availability manager uses to listen to the multicast core groups.