A new multichannel Ethernet protocol for passive optical star local area networks using coherent transmission D. Rodellar✦, C. Bungarzeanu✦, H. Garcia✦, C. Brisson✧, A. Küng✧, Ph. Robert✧ ✦Telecommunications
Laboratory(1), ✧Metrology Laboratory Swiss Federal Institute of Technology - EPFL ABSTRACT
A new optical local area network that combines dense Wavelength Division Multiplexing (WDM) techniques and coherent transmission4,5 benefits is presented. A new multichannel protocol is specified based on Ethernet1, using WDM to carry multiple channels on a single fiber. The protocol is simulated to evaluate the performance under congestion conditions and to compare it with other protocols having the same total capacity. Some suggestions are made for congestion control in high-capacity networks and different channel-jumping strategies are proposed. A network architecture is proposed with tunable transmitters and multichannel receivers, and with a common local oscillator. Keywords: dense Wavelength Division Multiplexing, optical coherent transmission, Ethernet, multichannel protocol.
1. INTRODUCTION Today’s computers can exchange data at Mbit/s and those in the future may be capable of surpassing this bit rate. Networking such machines together opens up a new field of multimedia applications and also of new network protocols adapted to high-speed network architectures. A new architecture of an optical multichannel network using coherent transmission3,4,5, a passive star and a common local oscillator is presented. The advantage of distributing the total capacity over several wavelengths with an appropriate protocol is shown. The performance of a multichannel protocol is compared with the performance of other protocols (Fast-Ethernet) with the same total capacity over a single channel. The following section will outline an architecture proposal to effectively establish communication between hundreds of stations of a LAN. Then the protocol and its parameters will be described. Finally, performance measures will be given to compare our proposal with other protocols having the same total capacity.
2. ARCHITECTURE PROPOSAL We analyze new optical local area network (LAN) that combines dense wavelength division multiplexing (WDM) techniques and the benefits of coherent transmission6. With over 10 THz of intrinsic bandwidth, fiber-optic networks offer the possibility of meeting future communication challenges and provide more effectiveness than copper wire based networks. A new network protocol based on a Wavelength Division Multiplexing (WDM) scheme that makes use of the benefits of the multichannel capabilities is proposed. The geographic extent of such networks can reach up to 2.5km. Whereas other high capacity network architectures are forced to reduce their influence area due to their high bit rate. A logical bus Local Area Network (LAN) is implemented over a passive star physical topology (Fig. 2-1). Each station has a tunable transmitter, that chooses the channel to transmit its packets. It is also equipped with a coherent receiver that captures packets from all the different wavelengths. This reception is achieved using coherent technology: the incoming optical signal is mixed with that of an optical local oscillator, which is shared by all the hosts, and a group of electrical filters selects each different data channel at different intermediate frequencies.
(1)
For further information - D.R: email:
[email protected]; Telephone: +41.21.693 39 87 Fax: +41.21.693 26 83; TCOM - Department of Electrical Engineering CH-1015 LAUSANNE; http://tcomwww.epfl.ch/
The studied LAN can fulfill the communication demand of a large number of stations with a high total bit rate. There is a double motivation in using coherent lightwave techniques. Firstly, the receiver sensitivity is increased (up to 20 dB) compared to Intensity Modulation/Direct Detection (IM/DD) systems7,6. This is a key feature since one of the principal issues is to address as many stations as possible without requiring optical amplifiers. Secondly, channel spacing in a wavelength division multiplexing (WDM) network can be considerably reduced such that it allows the simultaneous detection of all the channels by using a simple receiver scheme.
r
λj Shared local oscillator laser 1
.
Σ λk+ λLO
λLO . . .
Tunable laser transmitter
λi
Coherent receiver
Σ λk+ λLO
Passive Star Coupler
Tunable laser transmitter Coherent receiver
. . .
(N+1) x N
1 < i, j, k < M channels
Fig. 2-1: Multichannel LAN physical star topology; consisting of a shared local oscillator, a passive star coupler, and a tunable laser transmitter and a coherent receiver for each of the N stations.
The network configuration is based on a passive star coupler with a shared local oscillator. The star topology minimizes the network loss and, due to the shared local oscillator, only one laser per station is needed, which is a great improvement as far as coherent transmissions are concerned. A detailed description of a station is shown in fig. 2-2.
TRANSMITTER
Incoming DATA
STAR coupler Tunable laser
External modulator
Polarization controller
COHERENT RECEIVER DATA recovery
Envelope detectors
Bandpass filters
E/O
Fig. 2-2: Detailed description of a station with both transmitting and receiving schemes.
The transmitter uses a tunable laser to switch to any channel followed by an intensity modulator. Data recovering is achieved by a set of electrical bandpass filters. Each channel is received in a different frequency. Each frequency corresponds to the beat between each possible wavelength signal and the local oscillator signal. The data of each channel is recovered by a microwave envelop detector followed by a buffer. It is obvious that the economical practicability is presently a weakness of our proposal but the foreseen concepts of multichannel protocols and their efficacy might stimulate coherent technology in the future.
3. PROTOCOL DESCRIPTION The protocol is designed to support LAN traffic by means of adequate average delay and throughput performance. The protocol presented is based in the Ethernet1 protocol, which is a Carrier Sense Multiple Access (CSMA) with Collision Detection (CD) Medium Access Control (MAC) protocol. MAC protocols have to solve channel access conflicts. In the case of a CSMA/CD protocol channel access is performed as fairly as possible for all the stations. To achieve that fairness there is a collision detection at the transmitter which interrupts the transmission, sends a jam signal and waits a random uniformly distributed time. WDM is used to carry multiple data channels on a single fiber and it provides a multichannel quality to develop next generation system technologies for high-speed optical networks. Taking Ethernet as a starting point is not only a purpose of compatibility with existing standards, but also the intention to provide as much fairness as possible to the users. In the multichannel case, protocol acts like Ethernet for each channel. However, when a certain number of collisions has been detected while attempting to transmit a packet, the station can change to another wavelength and thus to another transmission channel. All parameters are the same as those of Ethernet (frame length, headers, slot time, etc.), except those related to the multichannel facilities8: the number of consecutive collisions (C) that one host has to wait before changing to another wavelength, and the laser tuning time (T), which is the time it takes the laser to jump from one wavelength to another.
4. PROTOCOL PERFORMANCE EVALUATION Ethernet traffic is statistically long-range dependent, also called self-similar2 (it exhibits similar-looking traffic bursts), and thus shows different statistical properties than those predicted by the stochastic models currently considered. It is crucial to analyze the performance of this new protocol using source models which exhibit self-similarity. To evaluate the impact of self-similar traffic on the performance, a tracing data file with real Ethernet packet arrivals has been used to generate fictitious traces by multiplexing this real traffic. Finally, these traces are the traffic sources for the simulations. A protocol simulator8 has been developed to analyze and evaluate the performance of our new MAC protocol under congestion conditions. Other protocols, particularly Fast Ethernet at 100 Mbit/s, are compared to our protocol with the same total capacity (10 wavelengths of 10 Mbit/s each). Compared to Fast Ethernet, the maximum distance is 10 times longer and some performance measures (average delay and packets loss rate) are remarkably better. Simulations are done with a large number of users connected and with a high traffic demand per user. We define the average delay as the time between the first attempt to transmit the packet and the time the transmission actually starts. This average delay does not include the waiting time of the packet in the buffer before being sent. Throughput is the number of packets transmitted successfully per unit of time. Finally, a packet is considered to be lost when 16 consecutive collisions have been encountered while attempting to transmit it. It is then returned to the above Control Level (to be resent to the MAC Level while giving an error at the Application Level). The delay of ‘lost packets’, as defined above, is taken into account for the average delay computation. We have simulated the protocol behavior for an offered incoming traffic of 2 Mbit/s per station and we have increased the number of station until the total capacity (100 Mbit/s) has been reached. The following figures compare the performance measures for the considered protocols. When throughput is plotted versus offered total traffic (fig. 4-3), a maximum throughput of 67 Mbit/s is found for Fast-Ethernet at an offered load of 70 Mbit/s. The multichannel protocol reaches a maximum throughput of 84 Mbit/s at an offered load
of 95 Mbit/s. While the throughput of Fast Ethernet decreases under congestion conditions, it remains constant in the multichannel protocol. These results show that our protocol reaches to the congestion conditions for a higher traffic and it shows a better throughput (the difference of the throughput of the two protocols at an offered traffic of 100 Mbit/s is 25 Mbit/s, which is the offered traffic of about 12 stations).
Throughput [Mbit/s]
100 80 60 40 Fast Ethernet Multichannel
20 0
0
20
40
60
80
100
Traffic offered by all stations [Mbit/s] Fig. 4-3: Throughput comparison between Fast Ethernet and multichannel Ethernet versus the total offered traffic.
Not only the throughput has better results but also the average delay is better than in the case of Fast Ethernet: the multichannel protocol delay is always below the Fast Ethernet delay (fig. 4-5).
Packet loss [packets/s]
30 25 Fast Ethernet Multichannel
20 15 10 5 0
0
20
40
60
80
100
Traffic offered by all stations [Mbit/s] Fig. 4-4: Packet loss comparison between Fast Ethernet and multichannel Ethernet versus the total offered traffic.
When a packet is sent to the channel by a source it can be successfully transmitted or it can suffer a collision with other packets of other data sources. The Ethernet protocol counts the number of consecutive collisions. The packets waits a number of slots (512 bits) which is related to the number of consecutive collisions: (2#collisions-1) slots; after 16 consecutive collisions the packet is ‘lost’: the MAC Layer sends a warning to the upper Layer (Logical Link Control) to say that there is congestion on the network. This ‘lost’ packet will be resent again provided that the upper layers do not stop the transmission. Because these packets are resent we compute their delay before been lost for the average delay results. As it is shown in fig. 4-4, the number of resent packets is ten times higher for Fast Ethernet compared to the multichannel protocol. Consequently, Fast Ethernet
delay must be higher than the delay in the multichannel Ethernet. There is also the fact that when collisions occur in a single channel protocol, the channel is kept occupied, and actually the collision reduce the accessible capacity. This can be seen as if the virtual capacity of the channel was reduced. In the case of a multichannel protocol collisions are divided among all channels so that the number of collisions per channel is reduced. The virtual capacity of each channel is slightly reduced and also the offered traffic per channel is N times lower in average. These considerations justify the good performance of the multichannel protocol compared to a single channel one. For an offered traffic of 100 Mbit/s the average delay of the multichannel protocol is 0.4 ms below the Fast Ethernet average delay. This leaves room to include the jumping time T, which is the time it takes the laser to jump from one wavelength to another when there is a collision. The difference between delays with the null jumping time and those with a jumping time of 50 ns are less than an Ethernet bit time. As the channels spacing are less than 2 GHz, the time to jump from one wavelength to an other is kept low. Thus all the multichannel results presented are simulated with T= 0 s. In the case of a slow tuning laser, the number of consecutive collisions before jumping (C) can be adjusted to achieve optimal performance. Packets will stay in the same wavelength before jumping to reduce delay. −4
x 10 Average delay [s]
12
Fast Ethernet Multichannel
10 8 6 4 2 0 0
20
40
60
80
100
Traffic offered by all stations [Mbit/s] Fig. 4-5: Average delay comparison between Fast Ethernet and multichannel Ethernet versus the total offered traffic.
An evaluation of different channel-jumping strategies is made. Stations have to select a new wavelength among the possible ones after a given number of packet collisions. A set of different policies is considered: 1.- all wavelengths have the same probability (random choice), 2.- choosing the wavelength with the longest idle time, 3.- a round-robin policy, 4.- no backoff waiting time after jumping. The average delay and throughput of the first three jumping policies are plotted on fig. 4-6. The fourth jumping policy shows worse performance than the others, due to the fact that if there is no backoff waiting time and all packets will collide again after jumping to another channel. The performance of these protocols are fairly the same, but these results foresee the possibility of other options that have a lower average delay and thus could be used for real time services. Other possibilities are being considered, like choosing a random choice over idle wavelengths, or reducing the backoff waiting time after a channel change.
−4
100 longest idle Throughput [Mbit/s]
Average delay [s]
8
x 10
6 round-robin
4 longest idle
2
0
20
40
60
random choice
80 round-robin
70 60
random choice
0
90
80
100
50 50
Traffic offered by all stations [Mbit/s]
60
70
80
90
100
Traffic offered by all stations [Mbit/s]
Fig. 4-6: Average delay and throughput comparison between different jumping policies versus the total offered traffic.
5. CONCLUSIONS A multichannel protocol that seek to exploit the high capacity of the fiber by creating multiple channels on the same fiber is presented. The advantages of our protocol are the compatibility with Ethernet, the scalability (simply by adding more wavelengths to obtain higher bit rates) and the possibility of increasing the total capacity while keeping the same physical topology by increasing the bit rate per channel (thus replacing Ethernet by Fast Ethernet). CSMA/CD networks lose efficiency when the total traffic offered on the medium approaches the medium’s capacity. Nevertheless, in the multichannel case the efficiency does not deteriorate so much as in the case of a single channel. Even, the average delay is much lower and the number of lost packets is a smaller by one order of magnitude when offered traffic is near the limit of the channel capacity. We also consider some suggestions for congestion control in high-capacity networks and therefore we show that a multichannel protocol shows better performance than a single channel one. New optical LAN protocol exploiting WDM and coherent transmission techniques is analyzed. Conceived as Ethernet evolution our multichannel protocol shows better performance than single channel one with the same total capacity.
ACKNOWLEDGMENTS This work is supported by the Swiss National Fund under contract 21-42085.94. The authors are grateful to Prof. P.-G. Fontolliet for his helpful comments and to Jürgen Ehrensberger for his generously given constructive comments.
REFERENCES 1. ISO/IEC 8802.3 ANSI/IEEE Std 802.3, ‘Carrier Sense Multiple Access with Collision Detection (CSMA/CD) access method and physical layer specifications’, IEEE publications, New York, 1992. 2. Will E. Leland, Murad S. Taqqu, Walter Willinger, Daniel V. Wilson, ‘On the self-similar nature of Ethernet traffic’, ACM SIGComm’93, San Francisco, CA, USA, September 1993.
3. Thomas K. Fong, Delfin Jay M. Sabido IX, Robert F. Kalman, Masafumi Tabara, and Leonid G. Kazovsky, ‘Linewidth-Insensitive Coherent AM Optical Links: Design, Performance, and Potential Applications’, Journal of Lightwave Technology, vol. 12, no. 3, pp. 526-534, March 1994. 4. Sadakuni Shimada, ‘Coherent lightwave communications Technology’, Capman & Hall, 1995. 5. Milorad Cvijetic, ‘ Coherent and nonlinear lightwave communications’, Artech House, 1996. 6. Leonid Kazovsky, Sergio Benedetto, Alan Willner, ‘Optical fiber communication systems’, Artech House, 1996. 7. G. P. Agrawal, ‘Fiber-Optic Communication Systems’, Wiley, 1992. 8. D. Rodellar, C. Bungarzeanu, H. Garcia, C. Brisson, P.-A. Nicati, ‘Performance Analysis of a Multiwavelength Ethernet Optical Local Area Network’, Proceedings of the European Conference on Networks and Optical Communications 1997 (NOC’97), pp. 63-70, Antwerpen, June 1997.
