The Multi-Service Network Architecture (MSNA) was developed to address both the naming and signalling required for the establishment of end-to-end ATM cell streams, and the integration of the in-band processing of cells with the systems present at the end points of communication links [16].
This protocol architecture takes an internetworking approach as we believe that different varieties of ATM networks will appear, depending on the functionality and data rate required, and the geographic area covered. An operational MSNA `internet' exists in Cambridge UK and spans the sites of Olivetti Research Ltd., the University Computer Laboratory and University Engineering Department. The physical network components consist of Fairisle ATM switches [14], slotted rings [11] and [6], and, in the near future, radio based ATM systems.
Figure 1: Functional mapping of MSNA to the OSI reference model and CCITT
B-ISDN reference model.
The Multi-Service Network Layer (MSNL) is the basic ATM service, giving an end-to-end stream of cells. It defines the MSNL address and circuit establishment techniques, including the mechanism for requesting various qualities of service (we are currently working on call acceptance and control algorithms to enforce the QoS guarantees).
Different ATM networks must present a common interface (MSDL) to the MSNL layer. During MSNL circuit establishment, routing is performed at each MSNL entity to select the appropriate MSDL instance for the next hop. The MSDL implementation for each network type then maps the MSNL circuit establishment mechanisms to those appropriate for its underlying network type; this can involve both address mapping and, in some cases, protocol mapping. For example, the MSDL instance for B-ISDN would map the MSNL mechanisms to Q.93B to establish that portion of the virtual circuit which traverses the B-ISDN network.
Once the circuit is established, the routing of cells from one MSDL instance to another is normally performed by hardware. Where the two MSDL instances are identical, this is the normal VCI mapping function seen in ATM switches. Where they are different, the hardware may also be required to translate the other fields of the header.
A particular design goal in MSNA is to eradicate the requirement to perform layered multiplexing whcih is often found in communications systems. For reasons of performance and quality of service, we often wish to identify the end-system application entity with the minimum of overhead and invoke the normal resource allocation mechanisms of the end system (e.g. processor scheduling and buffer management) before engaging in higher-layer protocol processing. Hence the MSNL address contains a component which identifies the end-system and a component which identifies an entity within the end-system, even to the extent of identifying an application level entity; this is similar to the common IP/UDP, or IP/TCP, host/port pair.
Where possible, and certainly when application specific quality of service is required, we aim to map a single application association to an MSNL virtual circuit so that when communication is establishedq, it is possible to identify the application entity within the end system directly from the VCI of a received cell. Then, although we may build adaptation, error control, transport, session and presentation layers on top of MSNL for particular services, these provide no further multiplexing. This also fits naturally with some services, such as video, which implement the majority of the in-band processing in hardware.
Such an approach of course does not insist that the identified entities be applications, and where appropriate (examples in use include IP over ATM and Ethernet bridging), the entity can be a particular protocol stack which implements further in-band multiplexing.