Currently, in a multi-server topology where servers are connected to each other, each server can only send one job at a time to each of its peers. This has an impact on scalability.

It is possible to "trick" each server into connecting multiple times to the same peer, but this only works with manual peer configuration, for example:

jppf.peers = driver2a driver2b
jppf.peer.driver2a.server.host = localhost
jppf.peer.driver2a.server.port = 11111
jppf.peer.driver2b.server.host = localhost
jppf.peer.driver2b.server.port = 11111

However, this is quite cumbersome and is not possible with auto discovery of peer drivers.

We propose to enable the definition of connection pools instead, with a configurable pool size:

jppf.peers = driver2
# five connections to driver2
jppf.peer.driver2.pool.size = 5
jppf.peer.driver2.server.host = localhost
jppf.peer.driver1a.server.port = 11111

or, with peer discovery enabled:

jppf.peer.discovery.enabled = true
# five connections to each discovered peer
jppf.peer.pool.size = 5