Download Performance of some Local Area Network Technologies and more Study notes Local Area Network (LAN) in PDF only on Docsity!
------Performance^ - -of^ some^ Local^ Area^ Network
Technologies by Samuel T. Chanson Arun Kumar
Ashok v. Nadkarni
Technical Report 82- August 1982 Dept. of Computer Sci•nce, University of British Columbia, Vancouver, B.C., Canada V6T 1W ABSTRACT This paper classifies local area network (LAN) technologies according to their topology and access method. The characteristics of the popular LAN technologies (namely Ring/Token passing, Ring/Message slots and Bus/Contention) are discussed. Analytic models are developed to estimate the mean packet delay time of each technology as a function of the network loading for various packet sizes and number of active stations. It is found that in the case of slotted rings (but not the other two technologies) an optimal value of the number of active stations exists which minimizes the mean delay time at all load levels given a packet arrival rate. The LAN technologies are compared with regard to their performance, reliability, availability, maintainability, extensibility, fairness and complexity. It -is hoped that potential users may be able to select the appropriate technology for their intended applications based on their specific performance requirements and operation environment. As well, LAN designers may benefit from the insight provided with the analysis. ACKNOWLEDGEMENT The authors would like to thank Prem Sinha for useful discussions. This work was supported in part by the Canadian Natural Sciences and Engineering Research Council under grant No.3554.
A local area network is generally considered to be one which covers a "limited" geographical area such as a building or a group of buildings within a few kilometers of one another. LANs typically also exhibit certain attributes such as high data rates, a high degree of interconnection between devices on the network with each station having the potential of communicating with every other station. As well, each station generally listens to every transmission, whether addressed to it or not [5). Because of the characteristics of LANs, the technologies used are quite different from those of long-haul networks [6) Various LAN technologies have been developed. It is increasingly clear that no single technology is superior to the others in all respects. This is evidenced by the IEEE 802 Committee (which is considering the lower level protocol standards for LANs) proposing two incompatible systems to be standardized contentions (CSMA/CD busses) and token passing (both for rings and busses) [7) • This paper compares the characteristics and performance of the popular LAN technologies. It is hoped that potential users may be able to select the appropriate technology for their intended applications based on their specific performance requirements and operation environment. As well, network designers may benefit from the insight provided with the analysis. An understanding of the basic principles of LAN (such as those contained in [3] ) is assumed.
r
- Classification of LAN technologies LAN technologies consist of at least the following aspects:
a) Physical data transport mechanism: E.g., digital signalling using baseband (single channel) or broadband (modulated, multi-channel) techniques, over twisted pair, coaxial cable, fiber optics or radio.
b) Topology (the connectivity characteristics of the network nodes): E.g., star, ring, bus which may be fully connected, hierarchical, cross-connected or irregular.
c) Sharing technique (the bandwidth allocation scheme for multiple users):
E.g., dedicated (non-shared), time division multiplexing, frequency division multiplexing, contention.
d) User services and protocols: Each LAN invariably has its own terminal support functions. Despite th~ efforts of the IEEE 802 Committee, the American National Standards Institute (ANSI), the International Standards Organization (ISO) and other organizations, standards for higher-level protocols are not expected in the near future.
Given the physical environment the network is to operate in and the financial considerations, the choice of the physical data transport mechanism is usually straightforward. As well,
Reliability A station or a $ection of the channel failure should not affect the rest of the network. As well, hardware and software failure rate should be as low as possible. The IEEE 802 Committee, for example, recommends that the number of undetected error should not exceed once per year[7].
Availability Rapid fault isolation and short mean-time-to-repair enhance the availability of the network. A guaranteed upper-bound on the waiting time to access the network further increases its availability.
Maintainability Errors should be easy to detect and correct. It should be possible to remove a station from the network for repair without disturbing the rest of the network. A network which is easy to maintain usually is more available than one which is not.
Extensibility It should be possible to add (or remove) stations from a network without disturbing it. The maximum length of the network should be reasonably long. As well, the network technology should allow easy adaptation to new transportation media (which can be orders of magnitude faster than is currently available).
Fairness By fairness we mean no request from any station to access the channel should be discriminated against. In other words, the variance of the delay time to gain access to the channel should be low. In addition, the mean delay time should be nearly equal for all stations.
Complexity The cost of the network as well as its reliability and maintainability is usually directly related to its complexity. Thus the simpler the technology the better.
Before the different LAN technologies are compared we first give a brief description of the salient features of each technology. The reader is referred to [3,8,10,12,17] for details.
3.1 Contention Busses There are several variations of this technology. The nodes or stations are all attached to a passive transmission line (the bus). In the simplest scheme, the stations may transmit at any time. A collision occurs if more than one station transmit simultaneously, in which case the stations involved will retransmit after some random time (backoff). This unrestrained contention of the use of the channel results in very low useful throughput rate , with the maximum value at approximately 0. of the transmission rate. This is the classical Aloha scheme [ 9 ] •
a packet on the channel may only be picked up by one station at a time if the network is operating correctly. Also a packet is sent by a station on only one link connected to a single station (the one down stream).
..---5tation
~.....--active repeater and
interface
Figure 1. Basic Ring Topology
3.2.1 Slotted Rings
This access method requires one of the stations on the ring to be designated the "monitor station". It initially transmits a number of empty packets (slots), which usually are of fixed size. If any station wishes to send data, it marks a passing
packet which is empty and puts the data and the destination's address into the packet. The packet circulates around the ring and is checked by each station. If the destination's address matches with the station address, the data is read from the packet and the packet is marked as empty. If the packet completes a full circuit of the ring and is detected by the sending station without having been emptied, the sending station empties the packet and returns an error code to its associated machine's operating system. The monitor station also checks each packet for errors and clears a. marked packet if it has completed two circuits. Note that this scheme allows more than one station to transmit and receive packets simultaneously if the number of slots exceeds one. The Cambridge Ring [12,17] is an example of the slotted ring.
3.2.2 Token Rings There is no "monitor" station in this scheme. A special bit pattern known as the token circulates around the ring. There is only one token. A station may only transmit a packet onto the ring if it is in possession of the token. It does this by replacing the token by a special bit pattern often called the "connector", append the data after it and the token after the data. Unlike the slotted rings, the packet can be of variable size.
Every station on the ring "sees" the packet and may verify its integrity and its address. However, only the Sending station
Appendix for the analytic models). ·The normalized load is used so as to remove the channel transmission rate as an explicit independent parameter in the relationship. It also has more intuitive meaning than the actual load. The normalized load varies between zero (no transmission) and one (network is saturated and no more bandwidth is available). Since t depends on the packet size and increases as the packet size increases (everything else remains constant), the absolute delay time may be misleading in some cases. It is customary to use the delay time normalized by a base packet size B. Thus the normalized delay time with respect to packet size xis simply its absolute delay time multiplied by B and divided by x. The normalized delay time represents the time required to transmit B bits of information through the network. The absolute as well as the normalized delay is plotted against the normalized load for various values of packet size (Figures 1,2,4,6). To study how well the technologies respond to increase in the number of stations, the absolute delay is also plotted against the normalized load for various values of the number of active stations in the network (Figures 3,4,7). For the slotted ring technology, there are two additional system parameters which affect performance - slot size and the number of slots in the ring. For simplicity, we shall assume the slot size to be the same as the packet size. This eliminates a parameter and simplifies the model. Figure 5 shows the relationship of t as the normalized load changes for different number of slots, keeping the slot size constant at 64 bits. In Figure 7, as the number of slots varies, the slot size is also
changed to maintain the same number of bit delay in the ring.
In all the plots, the channel has a transmission rate of 3 Mbits/sec. and is 1 km long.
a) Contention Bus (Figures 1-3)
The model is actually that of the Ethernet (Appendix I), which is the most popular contention bus technology. It is a modification of the one by Almes and Lazowska [22). Instead of an infinite population model, our model allows the mean delay time to be expressed in terms of the number of active stations in the network. To better utilize the bandwidth of the channel, Ethernet recommends the minimum packet size be 512 bits for a 10 Mbits/ sec. line ?8? (or 154 bits for a 3 Mbits/sec. line). The 64-bit packet size in the figures are there to show that performance will degrade drastically when smaller packet sizes are used. Observe that the mean delay time remains very low when the packet size exceeds the recommended minimum value until the load is close to the maximum channel capacity. Also, the larger the packet size, the less time is required to transmit each bit of information and the channel can handle a larger loading before reaching saturation (Figure 2) • (In this paper, saturation is loosely (^) defined as the (^) state when (^) mean delay time approaches infinity.)^ Because the tm component of the delay time is linearly proportional to the packet size, the absolute delay time per packet decreases as the packet size
load levels. This is because an additional station adds to the walk time delay (i.e., the time required for a bit to go once around the idle ring) thus increasing the propagation delay tp. Contention bus technologies do not suffer from this disadvantage.
c) Slotted Ring (Figures 5-7) Very few work exist on the analytic modelling of the slotted ring technology. Our model (Appendix Ill) makes use of several results of queuing theory to obtain the mean steady state value oft. For simplicity, it is assumed that the packet size coincides with the slot size and that the gaps between slots are negligible. ln Figure 5, the slot size is kept constant while the number of slots n is varied. Thus, the more the number of slots, the longer the walk time. This explains the increase in absolute delay time as n increases when the load is light. However, for a given load, the probability of finding an empty slot increases as n is increased. At higher loads, this wait time tw dominates the delay time and the mean delay time tactually decreases as n goes up. As well, more load can be accommodated before the system saturates. ln Figure 6, the walk time is kept constant as the number of slots is varied. Thus the slot size decreases as n increases. The normalized delay (with respect to slot size of 256 bits) is better for larger slot (and thus packet) sizes.
Figure 7 shows that for a given slot size and number of slots in the ring there is an optimal number of active stations N which minimizes the mean delay time and maximizes the saturation load. For a given load, the larger the value of N the less the mean number of packets waiting at each station. Thus the mean queue wait time monotonically decreases as N increases. However, because the number of stations ready to transmit has increased, the mean wait time to acquire an empty slot is monotonically up. Furthermore, the walk time also increases monotonically with N. The opposing effects on the mean delay time as N varies imply that there is a value of N which minimizes the mean delay time for a given load. Figure 7 shows that this value remains constant for all load levels. In our case, the number lies between 2 and 128. It may be possible to compute the number-mathematically. Knowledge of this information allows us to estimate how close an existing ring is operating from its theoretical optimum. It also allows an estimate of how many stations should be attached to the ring.
4.2 Reliability
a) Rings
The basic ring topology has often been criticized as being unreliable on the ground that an open circuit anywhere or the failure of any repeater will aisrupt the entire network. This is certainly a problem with a large number of repeaters strung together. Current ring designs, however, often use redundant paths or fail-safe bypassing to avoid this problem.
synchronous operation. At high data rates (such as those in excess of 10 Mb/sec.) synchronous operation provides much more reliable performance than the burst mode operation of the contention bus systems. We shall assume this or similar ring design in subsequent discussions.
a.1 Token ring vs slotted ring The decentralized control scheme of the token ring technology means that there is no single station failure which may bring the entire network down. The monitor station of the slotted ring, however, is a critical component whose failure may disrupt the whole system.
b) Busses The bus is essentially a passive device thus its reliability is much higher than that of the basic ring design. The wire centre concept of the star-shaped ring, however, has reduced the advantage offered by this aspect of the bus. While there are no central control points in the bus technology, a short circuit or a transceiver which fails to stop transmitting will disrupt the bus. Thus we feel that the reliability attributable to topology is comparable for the bus and the star-shaped ring. The reliability attributable to the access method has much to do with the corresponding complexity and will be discussed under that heading.
4.3 Availability
4.3.1 Rings
The wire centre misbehaving station.
concept allows rapid isolation of a As well, the centralized location facilitates maintenance and reconfiguration of the network, thus enhancing its availability.
Ring technologies, particularly the token-passing scheme and· also the slot message design with appropriate protocols, provide a guaranteed maximum in the queue wait time to access the channel. This is primarily due to the synchronous control operation and the unidirectional data flow in ring technologies.
4.3.2 Contention Busses A basic weakness of the contention scheme is that an unlucky station may have to wait a long time to gain control of the bus, though it seldom happens in practice. This is especially so when the network loading is heavy [15].
4.4 Maintainability
4.4.1 Rings Again, the wire centre concept allows the system to fail gracefully. This reduces the maintenance burden, since not all failures require instant attention. Stations may be removed for repair without disrupting the network. The centralized location also allows failures to be rapidly detected.
