It seems unlikely that there will ever be a one-size-fits-all solution for MCAs. Many solutions for network control exist today (e.g. Q.2931, P-NNI, IP Switching, etc.) each serving its purpose, but it is not realistic to expect any of these to evolve into The MCA that will cater to all our needs, present and future.
Instead, we would like to enable users to control their networks with the MCA that suits their environment best. For this purpose, previous work in the Computer Laboratory has allowed us to partition physical networks into virtual networks, each of which can be controlled by its own MCA [#!Merwe:98b!#]. This is illustrated in Figure , where a switch divider process partitions the resources on a switch into switchlets and offers the same switch interface to the MCAs as found on the switch itself. It appears to the MCAs as if they are controlling a real (albeit smaller) switch. Clients C1, C2 and C3 each then request its own MCA to exercise management and control, e.g. to set up connections.
As mentioned before, we think that the infeasibility of a one-size-fits-all solution applies also to the MCA itself. Therefore, we would like to extend the idea of switchlets into the MCA by enabling individual applications to specify their own policies for reserving and allocating resources. In this way, we can really speak about open control: flexible control that is not dictated by any one standard, organisation or network operator. Even so, the MCA proposed here is not intended to replace any existing MCAs. Instead it is expected to run alongside them as illustrated in Figure .
In the current implementation of the Sandman, almost all the control is moved out of the switch into a general purpose workstation, facilitating development and deployment of new services and allowing us to control very dumb devices.