1、一、英文原文： The Evolution of Campus Networks Towards Multimedia Ciro A. Noronha Jr. Fouad A. Tobagi 4 The Backbone The campus backbone carries the aggregate traffic from the sub networks. From a theoretical point, of view, the design of the backbone is no different, from the design of the sub networks t
2、hemselves； the steps taken in section 3 can be repeated here, with the difference that the network being designed is the backbone, and its users are the sub networks. The difference is one of scale. In this section, we will briefly comment on the effect of the backbone traffic on the aggregate throu
3、ghput of the sub network (which was ignored in section 3), and then compute the backbone traffic under a variety of scenarios. 4.1 Connecting the Sub network to the Backbone The main difference between the traffic to/from the backbone and the internal sub network traffic is the fact that former is c
4、oncentrated in a single point (the backbone connection)； in other words, all the traffic from the the sub network to the backbone is directed to the backbone connection, and all the traffic from the backbone to the sub network originates in the same location. In the other hand, the sources and desti
5、nations of the intra-sub network traffic are spread over the sub network. The effect of adding the backbone connection in the aggregate sub network bandwidth depends on its topology. In sub networks organized as trees, the congestion in the branches near the location of the backbone connection might
6、 limit the aggregate throughput to a value lower than what is attained when no backbone traffic is present. In the star configuration, however, the high-speed channel is the ideal location for attaching the backbone connection, and the aggregate throughput in this case is always higher than its valu
7、e when there is no traffic from the backbone. The reason for this is simple: the internal sub network traffic uses bandwidth both at the source and at the destination segments, but the traffic to/from the backbone uses bandwidth only at the source/destination segments, leading to a higher aggregate
8、throughput. The actual increase or decrease in the aggregate bandwidth will be a function of a, the fraction of the traffic that, is intra-sub network, and A, the ratio between the traffic from the backbone to the traffic into the backbone (defined in section 2.).The increase is higher for small val
9、ues of a (most of the traffic is inter sub network). 4.2 Structure of the Backbone In this section, we will give a number of scenarios that illustrate how the structure of the backbone can change as a function of the size of the campus network； we will use figure 3 to determine the required capacity
10、. The results are shown in table 1； note that the solutions for the structure of the backbone are not unique, and we have listed one of such solutions for each case. In figure 5 we show one possible backbone topology which has all the main elements we have been discussing. In the figure we show the
11、sub networks being served by switching hubs. The switching hubs serving sub networks with little external traffic can be grouped by an ATM multiplexer before reaching the ATM switch； this multiplexer can perform some local switching or rely completely in the central switch. Hubs generating larger tr
12、affic can be connected directly to the ATM switch. The switch is also the best location to connect centralized shared resources, such as video servers. The ATM switches themselves can be arranged in a hierarchical fashion, depending on the traffic requirements. 5 Adding multimedia to the other OSI l
13、ayers In this section we discuss the issues raised by the addition of multimedia services on the other layers of the OSI model. 5.1 The Medium Access Control (MAC) Layer The synchronous nature of audio and video traffic dictates that data be delivered to its destination within strict timing constrai
14、nts. Failure to deliver data on time results in a discontinuity of the video or a degradation in the quality. The busty nature of data traffic, on the other hand, means that a station transmits data in an unpredictable fashion, and the amount of data that the station transmits in a burst is also ran
15、dom. Both types of traffic are to be supported by the same local network. Indeed, multimedia applications naturally involve both types of traffic simultaneously； existing local area networks are expected to carry audio and video traffic alongside existing data applications. Mixing the two types of t
16、raffic on the same network requires special attention, particularly where shared resources are contended to by both types. The bandwidth available on local area network segments is one such shared resource, and is the subject of this section. It is shared by all traffic originating at all stations,
17、and accessed by means of the Media Access Control (MAC) protocol. Ideally, one would like the MAC protocol to include techniques which reserve the bandwidth necessary for synchronous traffic, so as to remove any effect of busty traffic. The CSMA/CD protocol (IEEE 802.3) does not differentiate among
18、the different traffic types. In essence, it operates as follows. A station with a packet ready for transmission transmits the packet as soon as the channel is sensed idle. If it collides with another packet, then it attempts again but after incurring a random rescheduling delay. The mean reschedulin
19、g delay is doubled with each collision incurred. This is clearly inadequate for real-time traffic. Not only is the delay incurred by packets nondeterministic, but its variance is quite large, owing to the exponential back off algorithm. However, since at low loads this protocol works well, it can st
20、ill be used if one segments the network to insure that this condition is met. IEEE 802.5 is a token protocol devised for a ring network. It uses a single token, in the sense that a station that has completed transmission will not issue a new token until the busy token returns to it. Since in a ring
21、network the connection to the medium is active, a priority scheme can be implemented by assigning dynamically a priority value to the token, and by restricting access to the ring to packets of priority equal or higher than the currently assigned value. This scheme can be appropriately used to integr
22、ate the two types of traffic on the same channel simply by giving synchronous traffic higher priority than the busty data traffic. In FDDI (ANSI X3T9.5), another scheme with multiple priority levels is employed, by the use of a set of timers, that regulate for how long the station can transmit traff
23、ic of each priority. In summary: the assignment of different types of service for real-time and data traffic at the MAC layer is very important when communicating multimedia information. This is not offered in the widely deployed IEEE 802.3 Other networks, such as the IEEE 802.5 token ring and FDDI
24、can provide this functionality, although many vendors have chosen not to implement it. Given the large base of deployed networks, it is not reasonable to expect that changes in the MAC layer to support multimedia would be implemented. Therefore, one must seek other options (such as making sure that
25、the network operates with low enough load that the problems described in this section will not be seen). 5.2 The Network Layer As the changes described before take place in the network, the route to use when moving information from segment to segment becomes increasingly important. In simple topolog
26、ies, such as star with the switching element in the center, routing is trivial because there is only one route from the source to the destination. However, in practice multiple routes might be provided between a given pair of stations, for reliability and/or increased. Currently, the path taken by a
27、 packet from its source to its destination through the network is determined by the bridges and routers that define the network topology. Transparent bridges use the spanning tree algorithm, i.e., the bridges, working together, identify a subset of the network topology that constitutes a spanning tr
28、ee, and direct all traffic through this spanning tree. Links not on the spanning tree are not used and kept in reserve, to be activated in case of failure of a link in the spanning tree. Moreover, the spanning tree identified for a topology is essentially chosen at random, and it is not optimum in a
29、ny sense. This has two consequences: first, extra capacity on the redundant links is never used； and second, bottlenecks may develop due to traffic concentration in some segments, as a result of the particular spanning tree chosen. We have discussed this traffic concentration in section 3. Routers,
30、on the other hand, have the ability to use multiple spanning trees； as a matter of fact, each router forwards the packets over the spanning tree having itself as the root. Many algorithms exist to compute those routes in a distributed fashion (RIP,OSPF, IS-IS, etc), but they all employ the cost of t
31、he link as the metric used when computing the spanning tree (the cost of a link is an arbitrary function, inversely proportional to the links bandwidth). Routers can recover from network failures, and can make use of the bandwidth of all links available. However, in most cases, all the traffic betwe
32、en a source and a destination will be sent over the same route. Moreover, current, routers have no provision for taking into account real-time requirements (such as delay) when computing routes, and the system of priorities implemented is clearly inadequate for multimedia communications. Of special
33、importance is the routing of multicast traffic (video-conference is an example of a situation where this kind of traffic is generated). For multicast traffic, two kinds of routes can be identified: minimum delay routes and minimum cost routes. Current multicast routers use the minimum delay criterio
34、n to compute the routes, but this might lead to unacceptable network loading. Therefore, new routing algorithms with the following characteristics are needed: Use of multiple spanning trees. Use of multiple routes between sources and destinations. Use of different criteria for routing the streams(de
35、lay, cost or a combination thereof). Use of a system of priorities that takes into account the real-time nature of the traffic, and not only different link costs for each priority. 5.3 Transport Protocols Existing transport protocols have been designed and implemented to support data applications in which the traffic is usually busty. Flow control procedures are embedded in the protocols to properly pace the flow between end users and achieve efficient utilization of network resources. For data applications, reliability is an absolute goal and is achieved by an error detection and
