Extended Slotted Ring Architecture for a Fully Shared and Integrated Net.
The recent proliferation of microprocessor-based workstations in offices, homes and even automobiles has finally focused our attention on the need for an efficient communication network architecture that is independent of distance and protocol constraints. In the absence of a coherent constraints. In the absence of a coherent network architecture, most existing network systems are specialized for either local or wide-area applications and consist of diverse
types of nodes performing task such as packet switching, packet assembly and disassembly (PAD), multiplexing/demultiplexing, data regeneration, network diagnostics and network control. Frequently, such nodes in a network system are manufactured by different vendors.
Complexities Are Uneconomical
All these facts make existing network systems uneconomical, due to complex system interfaces and lack of the flexibility in topological design needed for minimization of transmission costs.
There are several proven architectures good for either local-are networks (LANs) or wide-area networks (WANs). Ethernet and the Advanced Research Project Agency (ARPA), Department of Defense, are well-known examples. Most data switching networks are not very responsive do not provide priorities, as a consequence have difficulty integrating various kinds of data, and have no potential for future integration of voice.
Furthermore, network diagnostic, management and control functions are generally achieved through a superimposed network specially designed to perform these functions. Thus the existing data networks are not fully shared and integrated.
The purpose of this report is to present an overview of the evolution toward a fully shared and integrated intelligent network, define the architectural requirements for such a network, and present a network architecture based on distributed processing and VLSI (very-large-scale integration) technology. The extended slotted ring architecture (ESRA) described in this report satisfies the stated architectural requirements.
Easrliest Net Used Star Topology
The earliest information-processing network appeared in the form of a teleprocessing system. The user devices were directly connected to a host computer according to the star topology. As the geographical domain of the network increased, the transmission line cost also increased. To reduce the transmission cost, the host software was extended to achieve polling and calling, resulting in the well-known single-center multidrop (SCMD) topology. Although SCMD resulted in a "shared" transmission line, the overhead and delays due to polling and calling could not be ignored.
As the number of data terminals and the geographical domain of the network increased, the communication network functions such as message queing/dequeuing, multiplexing/demultiplexing for the host channel, polling and calling, began overloading the host. A clever solution was to off-load the communication functions into another network node called a "front-end."
This development probably represented the first step in the distributed-processing evolution. However, one cannot ignore the increased interface complexity. The earlier remarks made regarding the star and multidrop topologies are still valid. The original front-end network systems suffered from low reliability. The system would die when the host or the FE malfunctioned. Many innovations, such as redundant FEs and/or disk-based FEs, thus were employed to achieve high availability and improved data protection.
As the size of the network system increased even further, a network of hosts and FEs, as shown in Figure 1, was developed. Terminals attached to one FE could access hots attached to other FEs and host-to-host communication was supported. However, all devices had to be supplied by the same vendor or conform to interfaces defined by that vendor. The above architectures encourage the use of devices that are all manufactured by the same vendor, thus resulting in the concept of closed networks.
Open Nets Need More Intelligence
Open networks owe their existence to the need for sharing of the network by heterogeneous hosts, and terminals. Open networks created a need for additional intelligence in the network and resulted in the network assuming and identity of its own. The concept of a host-centered network is replaced by the concept of a network-centered management information system.
The ARPA network, based on the Datagram-type packet-switching concept, was the first implementation of a shared open network. One could then attach hosts and data terminals from different vendors to the shared network.
Although the first packet-switching networks used distributed-processing techniques and shared physical channels on a statistical basis, there were disadvantages, including:
* They were based on old minicomputer technology.
* Had no network capability for distribution of traffic from the major nodes.
* Were slow in terms of response time.
* There was a lack of priorities.
Many existing corporate and government networks can be classified as assembled networks. These networks frequently incorporate multiple architectures and diverse types of nodes generally manufactured by different vendors. A large assembled network may consists of a packet-switched backbone network, a distribution network based on statistical multiplexer components, and local-area networks employing contention-bus or token-ring technologies.
Assembled networks present severe problems in efficiency and network management resulting from the complex interfaces between heterogeneous nodes. The on-goind system support can be difficult and costly. As a consequence, system availability can only suffer.
It is believed that networks will evolve toward a fully shared and integrated form that can handle not only asynchronous and synchronous data terminals, but also voice terminals and the emerging integrated voice/data workstations (Figure 2).
The fully shared and integrated network will be based on packet technology and on a homogeneous architecture that is suitable for both wide-area and local-area environments. Network management and control functions will be integral to the network.
Requirements for the Network
In addition, the fully shared and integrated network must be cost-effective and must provide both high responsiveness and availability. These qualities will require extensive use of VLSI and judicious choice and design parameters (packet size, dynamic-buffer sizes, flow-control and ARQ scheme, and so on), and freedom of topological design to achieve minimum transmission line costs. One can enumerate the requirements for a fully shared and integrated network as:
* Integrated network management and control.
* Freedom in choice of topology.
* High responsiveness.
* User-defined priorities.
* High availability.
* Flexibility in link rates.
* Ability to disable error control selectivity by virtual circuit.
* Transparent virtual circuits.
* Protocol translation as a shared network resource.
Hemogeneity implies that the same network architecture spans all distances and geographical boundaries. Moving from a building to a campus and finally to an entire country doesn't require any changes in the architecture.
Integrated network management and control implies full sharing of nodal logic and transmission facilities between communication functions and network management and control functions.
Need Not Obey a Fixed Topology
Freedom in choice of topology implies the ability to employ a variety of network interconnection schemes as required to achieve high performance characteristics and low transmission costs. The intelligent network is not constrained to obey a fixed topology such as star, ring or multidrop.
High responsiveness implies a short response time, defined (for data) as the interval between the moment the user presses the send key and the moment the first character of the response appears on the screen or the printer.
User-defined priorities imply the ability to allocate network resources based on user need for responsiveness. High-priority messages must be treated differently messages must be treated differently from low-priority transactions. Some transactions, such as voice packets, may require higher responsiveness than other transactions, such as data packets.
High availability implies high service reliability achieved through automatic alternate routing and redundant design of critical nodes.
Flexibility in link rates implies the ability to choose suitable transmission media interconnecting network nodes. The bandwidth of any physical link will be determined by the traffic to be supported. The architecture should not be tied to any given link rate or specific set of link rates.
The ability to disable error control selectively by virtual circuit is required to achieve full sharing of transmission facilities for asynchronous data, synchronous data and voice.
Transparent virtual circuits are required for full resource sharing.
Protocol translation as a shared network resource eliminates the inefficiencies associated with dedicating protocol translators to virtual circuit ends.
A traditional slotted ring architecture invariably implies a local-area network. An ingenious scheme to extend a slotted ring for both local and wide-area use has been developed. Digital transmission techniques employing either analog or digital modems over leased lines are permitted. See Figure 3 for an illustration of an extended slotted ring (ESR) organized to provide a simple tree topology.
According to Figure 3, the two-wire connection wraps around all the Elite One nodes on the physical ring. The Elite One nodes on the ring can be interconnected according to a physical ring, star, tree or multidrop topology, with flexibity to mix such topologies to achieve reduced transmission line costs.
A packet-switching technique is employed to fully share all the transmission link segments connecting the Elite One nodes. Data flows in a bit serial fashion on the ring. The nodal hardware achieves frame synchronization via the empty slots, each of which is a "one" followed by 95 "zeros". The nodal logic is interrupted and decisions regarding packet reception/transmission are made at every two-byte boundary. Packets can thus flow through the node with either partial-packet or full-packet store-and-forward (bypass or no-bypass) conditions.
The user can define up to four priorities assignable to each switched virtual circuit (VC) at the time of VC set up. Two bits of overhead in each packet represent the VC priority.
This feature, when combined with the use of very small packets and the capability of turning off the end-to-end error control, allows voice/data integration. Packets flow on switched virtual circuits. Packets are inserted into slots. EAch slot on the ring either carries an incoming data packet or it is an empty slot. In case the arriving packet carries the highest priority, it is allowed to pass on to the next node. However, if it has a lower priority than that of a packet waiting in the node, the arriving packet is queued in the node and the higher-priority packet in the node ia dequeued and sent on to the next node.
Packets of 12-Byte Length
ESRA employs very short packets of 12-byte length to obtain high responsiveness. There are four types of 12-byte blockettes (packets): empty (E) blockettes; First/Last (F-L), consisting of five overhead and 10 bytes of data; and Last (L), consisting of three bytes of overhead and up to nine bytes of data. The combination of F and L types of blockettes gives rise to a 24-byte-long block.
The packets travel on virtual circuits or permanent virtual circuits (PVCs) established at the time of service request. Signaling (or control) packets are required for setting up a VC or PVC. Some control packets, such as all-call packets, are also employed for collecting configuration, traffic and diagnostics data, and for automatic-retransmit-request (ARQ) purposes to maintain end-to-end error control. No node-to-node (only end-to-end) ARQ is employed within the ESR. Furthermore, the end-to-end ARQ feature on any VC can be inhibited for any user-defined application.
The extended slotted ring architecture allows the implementation of any network topology. In a building, a twisted-pair can connect all Elite One nodes in the form of a star or a ring. On a campus, all the Elite One nodes in different buildings can be connected to an end Elite One node in a star, a ring or a tree topology, whichever is most economical.
Assembling a Tree Topology
In a wide-area environment, four-wire common carrier channels can be used to assemble trees suitable for economical multipoint distribution of traffic. Figure 3 illustrates how a tree topology is implemented using an ESR. A single ESR may extend over a city or state or country.
Only end-to-end error control is provided over a single ESR. An interesting feature of the ARQ employed within an ESR is the provision of a rotary group of four logical VCs between the two end ports, supporting a continuous transmission of blocksets (each blockset equals 16 blocks) on the four logical VCs on a rotary basis. Transmission is delayed only when the end-to-end acknowledgement (ACK) is not received on a given logical VC within a time-out period.
Since each transmission segment (satellite link or not) connecting two Elite One nodes is shared by many concurrent VCs, highly efficient use of that transmission segment is possible. This scheme is different from other ARQ schemes that are based on the physical links and not an logical VCs. The rotary group of four logical VCs can be made even larger if and when necessary. The only cost is increased buffer space at the external windows that are situated at the ends of the virtual circuit. There are no compromises in the architecture.
ESRA accepts data from asynchronous or Bisync or SDLC-type devices in the original envelope, repackages the data into an internal envelope, moves the data packets on the switched VCs in a transparent mode without protocol conversion, and finally hands the data in the original envelope to the receiving device. Therefore, the transparent virtual circuit requires two similar devices at each end of the circuit.
Protocol translation is accomplished by inserting a shared protocol-conversion resource into the transparent virtual circuit. Dedicated protocol-conversion resources at the two ends of the virtual circuit are not provided.
EAch ESR has the capability of isolating a failed transmission line segment or a malfunctioning Elite One node and then restructuring the ring by using either a permanently assigned or dial-up standby line segment. Thus, when a break occurs on the transmission segment, a standby line segment is then employed to achieve self-healing. The process of self-healing is fully automatic and takes only a frew minutes. This capability results in high system availability.
Esprit One Node Can Be Used
A higher-level switching node called the Esprit One can be employed to realize a two-level extended multi-ring topology, as shown in Figure 4. An optimum network design can define the placement of Esprit One and Elite One nodes and their particular interconnections to minimize overall network system cost (consisting of transmission line and switching hardware costs). Local host ports, trunk ports and ISRs can be terminated in an Esprit One node to achieve inter-ESR communication. Each trunk acts as an ESR with two ends and identical network architecture (ESRA) is maintained throughout.
Each Esprit One node consists of a triple redundant processor and dual bus system. The switch employs majority processing logic. Each external interface (to local port groups, trunks and ESRs) is dual-terminated on the two buses for redundancy. Such redundancy in the Esprit One node results in a high (central-office type) availability. Each Esprit One node has the capacity to accommodate up to 67 external interfaces. Up to 99 Esprit One nodes can be employed in an ESRA-based network system.
Data Integrity of the ESRA
Data integrity over the two-level ESRA domain is still based on the end-to-end ARQ scheme (with a rotary group of four logical VCs per end-to-end path). End-to-end ARQ is also provided between the internal windows situated at the ESR/Esprit One and Esprit One/trunk interfaces. Large (128 24-byte blocks) dynamic-buffer areas are provided at each internal window for each VC. The flow-control scheme employed allows full resource sharing while rarely inhibiting new data inputs. Only when the return ACK doesn't come from the destination window within a certain time-out period is the transmission of a new blockset (16 blocks) delayed. Only when all internal windows (on a given VC) are blocked is data input prevented at the source. Up to 238 end-to-end virtual circuits are able to be maintained concurrently on each trunk bundle.
Self-healing in the two-level ESRA environment can be extended over two ESRs, so that either the dial-up scheme or redundant standby line segment may be employed to fold a section of any ESR into another ESR.
Network diagnostics and operational data are gathered by distributed network nodes and sent to network control processer node(s). This capability allows economical control of a very complex and distributed network. This would have been impossible if the network diagnostics and operational capabilities had not been built into the Elite One and Esprit One nodes.
It can be said that the nervous system (network intelligence required to diagnose and test the system continuously) and the brain (network control processors) are integral to ESRA (one-level or two-level). All storage elements can be read or modified by the network manager. ESRA provides a continuous self-learning capability, and thus provides an integral network management and control capability. Such a capability would be impossible without the homogeneity of ESRA architecture.
One can now test the various characteristics of ESRA against the requirements for a fully shared and integrated communication network. While an ESRA-based network system satisfies the architectural requirements of a fully shared and integrated network, it must be stressed that ESRA is a particular solution designed to meet these requirements.
|Printer friendly Cite/link Email Feedback|
|Author:||Doelz, Mel; Sharma, Roshan|
|Date:||Feb 1, 1986|
|Previous Article:||PC-LAN Combinations Should Be Helpful to MIS Departments of the Fortune 1000.|
|Next Article:||New ICA Survey Offers Updated Telecom Professional Profile.|