• No se han encontrado resultados

Tipos de errores y significaci´ on

In document PARA ESTUDIANTES DE CIENCIAS (página 152-156)

8. Distribuciones continuas de probabilidad

12.2. Tipos de errores y significaci´ on

· The team decided first to evaluate a variety of suitable media access methods. Each alternative and the conclusions reached by the team are summarized below.

Carrier Sense Multiple Access with Collision Detection (CSMAjCDj 2

This was the alternative most familiar to the team members, since Digital was currently building products utilizing CSMAfCD for both baseband and broadband media. The performance of CSMA/CD does not degrade rapidly as a functio of the number of connected nodes (see Figure

1). However, its extent (maximum signal propa­

gation path length), transmission rate, and mini­ mum packet size are not independent because of fi nite propagation delays . 3 ·4 Therefore , to

increase the physical extent of CSMA/CD LAN,

the minimum packet size or the transmission rate or both must be decreased to ensure that there are no undetected collisions.

56 w 102 ...J <( (.) (/) (!) 0 ::::!. (/) 1 0 Cl z 0 e � 1 6 (.) w (/) e � a :J ...J � � >- <( ...J w Cl z <( w 1 a·' � 0 25 50 75 100

PERCENT OF OFFERED LOAD

Figure 1 Local Area Network Performance Carrier Sense Multiple Access (CSMA)

By eliminating the collision detection capability

of CSMA/CD, one can build a LAN whose trans­

mission rate scales well with distance . However, the obvious benefits of collision detection will be lost. Without collision detection, the delay variance experienced by applications tends to increase because CSMA relies on more-frequent higher-layer protocol time-outs. To compensate for this problem, the transmission rate could be increased sufficiently to reduce the probability of collision. However, this action would impose significant cost penalties on the end stations, which would need faster logic and would experi­ ence more difficult transmission problems. Carrier Sense Multiple Access with Partial Collision Detection (CSMA ± CD)

Either the physical extent or the transmission rate of a CSMA/CD network could be extended so that collisions would be detected only if the col­ liding stations were sufficiently close or if the packets were sufficiently large . Such a scheme would have good throughput-delay characteris­ tics if the physical extent were small; however, degraded performance would result if the physi­ cal extent were large. Unfortunately, this scheme yields a significant delay variance because the relative locations of the colliding stations and the size of the colliding packets now affect the layer, either the media access control (MAC) or transport, at which collision recovery is per­ formed.

Digital TecbnicaiJournal

Token Passing Bus5

The characteristics of the token-bus access method scale reasonably well with physical extent (see Figure 1 ) , but poorly with the num­ ber of nodes.3 This situation is complementary to that of CSMAjCD. The sensitivity of a token bus to the number of nodes makes it unsuitable for a single LAN with many nodes. The token bus, like CSMAjCD, is well suited to implementation on a CA1V-like cabie plant.

Token Ring6

The performance characteristics of the IEEE 802.5 token ring are somewhat similar to those of the token bus. However, an IEEE 802.5 token ring station will not reissue a token until the pre­ viously transmitted frame has circulated com­ pletely around the ring. This characteristic makes the ring more sensitive than a token bus to increasing physical extent. Moreover, a token ring cannot be applied directly to a branching­ tree physical topology, such as the one in a CA1V-like cable plant.

Slotted Ring

The design tradeoffs made in most slotted rings result in small slots, usually less than 20 bytes. Therefore, it is important to minimize the slot overhead, such as source and destination addresses and error detection fields. Such opera­ tions are usually associated with connection­ oriented services, such as voice transmission. In

slotted ring networks, mechanisms are often present to impose a measure of "fairness" in the network. Those mechanisms make it difficult for an individual station to acquire a significant frac­ tion of the instantaneous transmission rate. Such networks are often inadequate for handling the bursty traffic expected in the environments of interest.

Time Division Multiplexed (TDM) Bus

The principal disadvantages of using a TDM structure are that the number of time slots is fixed, and each time slot is assigned to only one station. Thus, with a large number of stations, even with low network utilization, the mean waiting time is large. Furthermore, since the bus is allocated in tum to each station, the maximum throughput of any station is limited to the data transmitted in that station's slot. The TDM bus is well suited to isochronous traffic, such as voice

or video.

Digital TecbnlcalJournal No. 3 September 1986

Frequency Division Multiplexed (FDM) Bus

The characteristics of an FDM bus are somewhat similar to those of the TDM bus. The FDM bus has an additional degree of freedom in that it could have slots of different bandwidths. The problem with the FDM bus, however, is logical connectiv­

ity. To have full connectivity, each node must monitor each frequency band for messages des­ tined for that node. In practice, this monitoring is prohibitively expensive.1As an alternative, one could apply a reservation' system to either the TDM or FDM buses. The dharacteristics of such an approach, however, arej much better suited to a connection-oriented ser\rice, such as voice or I video, rather than one with bursts of data. . I

I.

Hybrid of FDM and CSM

1

fCD

A hybrid scheme utilizing multiple slow-speed (approximately 1 million bits per second, or Mbps) CSMAjCD channels, each in its own 6-MHz band, is another specific alternative that was considered. Without increasing the mini­ mum packet size used in an Ethernet, each CSMA/CD channel can span an extent of approx­ imately 30 kilometers. Multiple CSMAjCD chan­ nels could be used to increase the aggregate capacity of the network. Unfortunately, logical connectivity cannot be achieved without some mechanism for switching packets between these channels. Furthermore, th� bandwidth available to any station is limited to a rate of 1 Mbps. Since there is no industry standard for a 1 -Mbps, 6-MHz CSMAjCD LAN, selecting;this approach would make necessary an attempt to standardize it.

These evaluations con

}r

inced the team that none of these access methods sufficed for build­ ing a single LAN capable bf successfully operat­ ing in all dimensions of interest to the project. Not one of these alternatives was capable of directly employing all the types of media that the customers wished to utilize. Furthermore, any choice was constrained , by the desire for an access method with a defined standard having the appropriate parameters. The project team would have to find a way to interconnect at least a subset of the standard LANs if the project were to be successful.

LAN Interconnection Alternatives

The team next investigated a variety of intercon­ nection methods, each of which had certain advantages and drawbacks.

57

The Extended Local Area Network Architecture and LANBridge 1 00

DECnet Router

The architecture for DECnet Phase IV+ could be

used to create a DECnet network for the inter­ connection. Such a network would be fully capa­ ble of handling the number of nodes needed in any of the three environments of interest. In fact this was quite an attractive alternative . One could choose the data links in such a network to optimize the cost and performance. For exam­ ple, Ethemets placed in local areas could offer good response at low and moderate network loads. Then a token bus, with its capability to handle high utilizations and large extents, could be used as a backbone to i nterconnect the routers. Those would in tum connect to the Eth­ emets. The sensitivity of the token bus to large numbers of nodes would be minimized since the only nodes on the token bus would be the routers. Unfortunately, not all customer nodes use the DECnet routing protocol, making this alternative useful for only a subset of the nodes in a network.

Central Switch

To complete the logical connectivity of a net­ work composed of multiple LANs, the team con­ sidered an architecture organized around a cen­ tral switch element. Conceptually, the switch could be connected to all LANs in the network and then selectively forward packets to the LAN with the destination end station. This alternative has most of the advantages of the DECnet router solution discussed above. Normally, however, all end stations need the same routing protocol. To avoid this problem the switch must either sup­ port a variety of routing protocols (and translate among them) or somehow perform its switching task in a way transparent to the end stations. A single switch of sufficient capacity and reliabil­ ity to do either task was likely to be fairly com­ plex to design and manufacture_ It would also need to scale in a cost-effective manner for a wide range of networks.

Bridge 7 •8•9

A bridge, or MAC layer relay, is a device connect­ ing two or more LANs so that a node on one LAN may communicate with a node on another' just as if they were on the same IAN. In operation, a bridge is a store-and-forward switch that isolates traffic to only those IANs on which the traffic must appear. For example, in Figure 2 , traffic between nodes X and Y would not appear on the

58

LAN to which node Q is connected. Bridges make use of data link layer addresses to make forward­ ing decisions. A bridge receives all frames from a particular LAN and then decides, based upon the destination address in the MAC header, whether to forward each frame.

A collection of LANs interconnected by bridges is referred to as an extended IAN. In general, bridges may be used to connect LANs of different types, as shown in Figure 2. Therefore, this alter­ native can successfully utill.ze diverse LAN tech­

nologies, if appropriate, to optimize some func­ tion (e .g., low cost, high performance, or a

combination of these) . Furthermore, a bridge appears to be merely another station on each LAN to which the bridge is connected. Therefore, multiple IANs, each fully configured, can be connected to eliminate their practical con­ straints on distance, number of nodes, media, and utilization.

Based on the reasoning above , the team selected the bridge alternative as the one best suited to realize the expanded goals of the pro­ ject. Thus the team began to develop an architec­ tural specification for extended LANs and bridges. The team also began to develop a work­ ing breadboard model that eventually led to the IANBridge 1 00 product. The architecture that was evolved for extended IANs and bridges is described in the next section.

lOCAL AREA NETWORKS

Figure 2 Bridged Network Configuration

Digital Tecbnical]ournal No. 3 September 1986

Advantages of Bridges

Bridges used to connect IANs have several useful properties:

• Traffic Filtering - Bridges isolate each IAN

from traffic that does not have to traverse that IAN. For example, in Figure 2 , traffic between nodes A and B is not sent on the IANs to which P and Q are connected. Because of this filter­ ing, the load on a given IAN can be reduced, thus decreasing the delays experienced by all users on the extended IAN.

• Increased Physical Extent - IANs are limited

in physical extent (at least in a practical sense) by either propagation delay or signal attenuation and distortion. Being a store­ and-forward device, a bridge forwards frames after having gained access to the appropriate IAN via the normal access method. In this way the extended LAN can cover a larger extent than an individual IAN. The penalty for this coverage is a small store-and-forward delay.

• Increased Maximum Number of Stations -

Because of either physical layer limitations or stability and delay considerations, most IAN architectures have a practical limit on the number of stations on a single IAN. Since the bridge contends for access to the IAN as a sin­

gle station, one bridge may "represent" many nodes on another IAN or an extended IAN.

• Use of Different Physical Layers - Some IAN

architectures support a variety of physical media (baseband and broadband coaxial cables, and optical fiber cable) that cannot be directly connected at the physical layer. Bridges allow these media to coexist in the same extended IAN.

• Interconnection of Dissimilar IANs - IANs of

different architectures are typically intercon­ nected with routers or gateways. Often these devices are complex with only moderate throughput, not an appropriate situation for a LAN environment. It is possible to build a bridge connecting dissimilar LANs (within constraints discussed in the section ."Perfor­ mance Considerations") . For example, such a bridge would allow stations on an IEEE 802 .3 (CSMAjCD) IAN to send frames to stations on IEEE 802.4 (token bus) or IEEE 802 . 5 (token ring) LANs. 2•5•6

Dtgital Tecbnical]ournal No. 3 September 1986

In document PARA ESTUDIANTES DE CIENCIAS (página 152-156)