文档库

最新最全的文档下载
当前位置:文档库 > A new media access control protocol guaranteeing fairness among users in Ethernet-based pas

A new media access control protocol guaranteeing fairness among users in Ethernet-based pas

A New Media Access Control Protocol Guaranteeing Fairness Among Users in Ethernet-based Passive Optical

Networks

Fu-Tai An, Hopil Bae, Yu-Li Hsueh, Kyeong Soo Kim, Matthew S. Rogge, and Leonid G. Kazovsky Optical Communication Research Laboratory, Stanford University (http://www.wendangku.net/doc/71791ea10029bd64783e2cab.html)

Phone: 1.650.725.1110; e-mail: ftan@http://www.wendangku.net/doc/71791ea10029bd64783e2cab.html

Abstract: We propose a new EPON MAC protocol guaranteeing fairness among users by

allocating excess bandwidth proportional to their subscription rates. The novelty of the protocol is

in the use of scalable per-subscription-rate-queuing with round-robin scheduling and packet

reclassification at ONU.

?2000 Optical Society of America

OCIS codes: (060.4250) Networks

1. Introduction

Low cost and minimal protocol overhead make Ethernet-based Passive Optical Networks (EPONs) a promising candidate for next-generation broadband access networks. In implementing EPON systems, one of key challenges is designing Media Access Control (MAC) protocol to achieve network efficiency and meet various Quality of Service (QoS) requirements.

For efficient use of upstream bandwidth, EPON MAC protocol resorts to Dynamic Bandwidth Allocation (DBA) scheme [1]. The current DBA specification, based on per-priority queuing with reporting of queue lengths at Optical Network Units (ONUs) via GATE and REPORT messages, can differentiate service priorities among different types of traffic, but does not guarantee fairness among multiple users served by the same ONU [2]. In this case the bandwidth allocation to lower-priority traffic, especially for Best Effort (BE) traffic without any connection parameters, simply depends on the queue length reported. Hence, malicious users are able to gain more bandwidth than others by over-flooding the priority queues, which cause severe unfairness among those users served by the same ONT.

In this paper we propose and evaluate the performance of a new MAC protocol that solves the fairness issue in EPON. The proposed MAC protocol divides bandwidth for lower-priority traffic among users proportional to their subscription rates, preventing over-flooding from malicious users from affecting the performance of other users. 2. Proposed MAC Protocol and ONU Architecture

The current EPON DBA scheme is based on per-priority-queueing at ONU (up to 8 priority queues), and each queue contains packets from all the users served by the ONU. When the ONU sends a REPORT message for grant requests to the OLT, queue length is the only information that is conveyed. Therefore, the OLT has no idea about statistics of each user’s traffic. It grants the ONU a slot size that is proportional to the ONU’s queue length only. Hence, users of an ONU could gain more bandwidth by over-populating the priority queues.

To provide better fairness among multiple users at an ONU, in the proposed MAC protocol we assume that the OLT knows each user’s subscription rate. The OLT grants the associated bandwidth to ONUs, and ONUs then distribute the bandwidth to each user according to the subscription rate as follows:

1. When all the users in the network send traffic at rates equal to or more than their subscription rates and there is

no surplus bandwidth available, the OLT distributes bandwidth according to users’ subscription rates.

2. When actual traffic rates from some users are less than their subscription rates, any surplus bandwidth is

allocated to all other users proportional to their subscription rates.

3. After steps 1 & 2, if the OLT still finds any surplus bandwidth available, not used by some users, it allocates the

unused bandwidth again to users (as in step 2) who have packets to send in the queues.

The OLT aggregates the bandwidth belonging to users of an ONU and sends a grant reply that indicates the slot size for the ONU. The ONU divides the slot according to subscription rates and queue length for each user. For users of

the same subscription rate at an ONU, round-robin scheduling is used to guarantee fairness among them. Fig. 1 shows the ONU architecture employing the proposed MAC protocol with the format of the control message, where for simplicity, we assume two traffic classes, one for guaranteed traffic and the other for BE traffic.

A new media access control protocol guaranteeing fairness among users in Ethernet-based pas

Fig. 1. ONU architecture with the proposed MAC protocol and the control message format.

As shown in Fig. 1, a classifier directs incoming packets from each user either into a FIFO directly or into lower priority queues through round-robin scheduler based on their service classes. The FIFO is for higher-priority packets. A user’s subscription rate determines the size of these FIFOs. When the FIFO grows over a predetermined threshold, however, the packets originally classified as higher-class will be re-tagged to lower-class, in stead of being dropped, and redirected into lower-priority queues (L i), which is called packet reclassification. By doing this, the protocol can guarantee the instantaneous bandwidth is always bounded by the limit set by subscription rate. When the network load is low, the higher-priority packets can still be transmitted without reclassification. Packets from these FIFOs are then sent to a higher-priority queue, while there are k lower-priority queues given there are k subscription rates. By employing per-subscription-rate queueing for lower traffic packets, the proposed MAC protocol provides better scalability in terms of the number of queues and the size of control packet for reporting queue lengths than those based on per-user-queueing, especially when the number of users at ONU, M, is very large. Note that k is independent of M and can be kept smaller than M, while still guaranteeing fairness among users.

3. Simulation Results and Discussions

We model traffic from each user as a Pareto process to capture the self-similar nature of Ethernet traffic [3]. Packets are generated with packet size distribution matching that of a measurement trace in [4]. The guard time between packets is set to 10 ns. We assume three ONUs in the network, and four users per ONU.

To demonstrate the performance of the protocol, we measure and compare throughputs of users at ONU-1 (U1) and ONU-2 (U2) as we change the traffic demand of U2. Total available bandwidth for lower-priority (BE) traffic is set to 700Mb/s, and demands from all other users are kept constant and set to 500Mb/s in total. The subscription rates of U1 and U2 are set to 4 and 1, respectively.

We show the simulation results in Fig. 2, where the results in (a) and (c) are for the generic DBA MAC and those in (b) and (d) are for the proposed MAC. In Fig. 2 (a), when U2’s demand is less than 100 Mb/s, the throughput of U1 remains the same. The throughput of U2 grows linearly with its demand, since the network is not fully loaded. The throughput of U1 slightly decreases as the demand of U2 exceeds 100 Mb/s, since the total offered load of the network is over 1 in this case. Notice that the sum of throughputs grows beyond 200 Mb/s when U2’s demand is more than 100 Mb/s, which means a large amount of traffic from U2 affects the performance (i.e., throughput) of others. On the other hand, Fig. 2 (b) shows that U1’s throughput is constant (100 Mb/s) regardless of U2’s demand. After U2’s demand reaches 100 Mb/s, the throughput saturates at about 60Mb/s. The reason is that given the available 200Mb/s bandwidth for both U1 and U2, 160 Mb/s belongs to U1 and 40 Mb/s belongs to U2 (corresponding to 4:1 subscription rate ratio). However, since U1 cannot consume all the bandwidth, the protocol re-allocates this bandwidth to U2. The reason U2 does not fully achieve 100 Mb/s is due to overhead from the packet length variation and guard time.

(a) (c)

A new media access control protocol guaranteeing fairness among users in Ethernet-based pas

A new media access control protocol guaranteeing fairness among users in Ethernet-based pas

(b) (d)

Fig. 2. Throughputs of User 1 and User 2 for the case of (a) generic MAC with demand of U1 = 100 Mb/s, (b) proposed MAC with demand of U1 = 100 Mb/s, (c) generic MAC with demand of U1 = 200 Mb/s, and (d) proposed MAC with demand of U1 = 200 Mb/s.

In Fig. 2 (c), U1’s throughput drops with the increase in demand of U2, as expected, while in Fig. 2 (d), U1’s throughput remains at 150Mb/s when U2’s demand is over 40Mb/s, which shows a good division of available bandwidth between U1 and U2 by the proposed MAC protocol based on their subscription rates. Note that throughputs of other users are not affected with the proposed MAC protocol.

Another simulation with the identical settings but U1 and U2 now served by the same ONU also shows almost identical results to those shown in Fig. 2, which indicates that the proposed MAC protocol guarantees fairness of throughputs among users irrespective of their locations.

4. Conclusion

The proposed MAC protocol based on per-subscription-rate queuing with round-robin scheduling and packet reclassification, guarantees fairness among users served by the same ONU by allocating excess bandwidth according to their subscription rates. This protocol provides high network efficiency and supports QoS, while providing good scalability. The control message format for reporting queue length is also proposed to minimize the overhead. In this paper we assume two traffic classes for simplicity, but it can be easily extended to more traffic classes. References

[1] IEEE 802.3ah EFM Task Force P2MP (EPON) Baseline Proposals, http://www.wendangku.net/doc/71791ea10029bd64783e2cab.html/3/efm/baseline/p2mpbaseline.html.

[2] G. Kramer et al., “EPON scheduling protocol requirements,” IEEE 802.3ah EFM Task Force Meeting, Jun. 2002.

[3] W. Willinger et al., “Self-Similarity through High-Variability: Statistical Analysis of Ethernet LAN Traffic at the Source Level,” Proc.

ACM SIGCOMM 1995, pp. 100-113.

[4] WAN packet size distribution, http://www.wendangku.net/doc/71791ea10029bd64783e2cab.html/NA/Learn/packetsizes.htm.