Robert W. Walthall and Dr. Mark Clement, Electrical Engineering
This research was conducted to determine the effectiveness of a method of controlling the flow of traffic on the information highway. A new network protocol, Asynchronous Transfer Mode, is being developed to support applications that will run on the infohighway. These applications will include current data transfers such as surfing the net and carrying local area network traffic across the country, but it will also support things like video on demand, distance learning, and video conferencing.
Sending video across a data network presents many unique challenges. One of the toughest challenges is timing. It is currently possible to get on the Internet and watch a video from MIT, but the viewer gets a few seconds of video, 1 and then waits. It may be several minutes before the next few seconds of video will arrive. While the Internet is running at full capacity, the problem will not be completely solved by getting bigger pipes to carry the traffic. Asynchronous Transfer Mode (ATM) networks solve this problem by allowing the user to specify the type of service that he/she is requesting.
The ATM protocol is very flexible and will allow users to specify constraints for any type of traffic. Essentially the user can specify that the application being used is delay sensitive and can only tolerate so much delay. The user will request that level of service from the network. And if the network is able to support that type of a connection at that time, the network will accept the request and the connection will be made.
The ATM protocol has been designed to guarantee that the probability for losing certain types of the data in transmission (i.e. full motion video) will be less that 10-8. This type of traffic is referred to as guaranteed traffic. The result of this guarantee was that the full capacity of the pipes could not be used. In fact some studies showed that utilization would be limited to 30% of the capacity of the pipes.2
Another type of service was developed to fill the remaining capacity. It is called best effort traffic. If there is no capacity for the traffic, then it must wait. The wait would probably be less than a tenth of a second, and would be unnoticeable to many users (such as those surfing the net). A control mechanism had to be developed to tell the best effort traffic when to increase or decrease the amount of traffic that was being injected into the network. This turned out to be one of the most challenging aspects of the development of the ATM protocol.
Two types of control mechanisms were developed 1) Explicit Forward Congestion Indicator (EFCI) 2) Explicit Proportional Rate Control Algorithm (EPRCA)
The EPRCA was developed because it was determined that the first control mechanism (EFCI) did not react quickly enough to changes in the network load. But by the time the EFCI mechanism was found to be inadequate, it had already been installed on many ATM switches. Therefore, the EPRCA mechanism was developed to be backwards compatible. In other words, if a user had a switch that used the EFCI mechanism he/she would be able to use that switch even if the network was using the EPRCA mechanism.
This research evaluates the performance of EPRCA in a network consisting of edge switches using the EFCI mechanism and interior switches using EPRCA. Performance is evaluated in terms of throughput, buffer occupancy, fairness, and fast convergence. A network simulator developed at MIT and then modified by the National Institute of Standards and Technology was used. The simulator was modified so as to be able to accept actual video streams as input for traffic. The basic idea is to use the video traffic to generate noise, and then see if the control mechanisms can utilize the remaining capacity without overflowing the buffers in the switches.
We expected to find that users of an EFCI edge switch would experience a loss of performance in terms of, increased time to converge on the fair share, decreased throughput, increased delay due to buffer occupancy, and unfair allocation of rate to greedy best effort traffic sources. Our results indicate that this is the case. We also expected to find that accepting a switch using the EFCI mechanism would have a negative impact on the rest of the users in terms of increased buffer occupancy and unfairness of rate allocation. Our results indicate that this is not the case. In all simulations run so far the buffer occupancy of the interior switches remained low. And while the rate allocated to users of the EPRCA switches was unfair, it was beneficial to the users of those switches in that they received a higher rate than what they would have normally been given. The full results of this research were published and presented at the Fourth International Conference on Telecommunications Systems Modeling and Analysis,3 and the paper is available over the Internet.4 Overall the mechanisms work surprisingly well together, and, depending on the types of applications being used, it is very possible that the degraded service would be unnoticeable to the end user.
References
- L. Roberts. (1995). Can ABR Service Replace VBR Service in ATM Networks, IEEE Computer Society International Conference.
- Telemedia, Networks and Systems Group, http://www.tns.lcs.mit.edu/vs/rover/rover.html
- Proceedings of the 4th International Conference on Telecommunication Systems Modeling and Analysis, March 21-24, 1996
http://www.vanderbilt.edu/Owen/gavish/tsm96proceedings.html - ABR Bibliography, http://nebo.cs.byu.edu/˜walthalr/congestion.html