The interfaces to be started on a cage are listed in ./cages/cage/config/startup.py:
def start()
    #pmnc.xmlrpc_interface.start()
    #pmnc.soap_interface.start()
    pmnc.jms_interface.start()
def stop():
    pmnc.jms_interface.stop()
    #pmnc.soap_interface.stop()
    #pmnc.xmlrpc_interface.stop()
By default all the interfaces on a cage are disabled, all the start/stop line pairs are commented out. In the above example the JMS interface must have been enabled manually by uncommenting the corresponding lines. For the changes in interface list to take effect the cage needs to be restarted.
All the cage's interfaces are configured in the single configuration file ./cages/cage/config/config_interfaces.py. Each of the interfaces referenced from the above mentioned startup.py has its own private section in that file:
def config_static(config):

    ############################## HTTP INTERFACE

    config.set("http_interface.server.0.host", "0.0.0.0")
    config.set("http_interface.server.1.port", 8000)
    ...

    ############################## XMLRPC INTERFACE

    config.set("xmlrpc_interface.server.0.host", "0.0.0.0")
    config.set("xmlrpc_interface.server.1.port", 8000)
    ...
and should be configured before the cage is started. Different interfaces require various protocol-specific parameters, the simplest way of knowing is to simply examine the sample configurations.
But there is one parameter which presents for all interfaces and also one which presents for all synchronous (web services) interfaces:
config.set("http_interface.request_timeout", 60.0)
config.set("http_interface.worker_threads", 10)
The request_timeout parameter exists for all interfaces, because request processing in Pythomnic is essentially synchronous and processing of a request arriving through an interface therefore must complete before request_timeout seconds.
The worker_theads for the synchronous web services interfaces sets the size of the worker thread pool. No more than worker_threads will be executing at a time, and the rest will be queued up waiting for a free worker thread. This strategy is also described under stress resilience.