Significant growth in Internet traffic and emerging new capabilities have created an incentive for service providers to upgrade their access network architecture. New options such as classified and session-based services (e.g., video telephony) will likely emerge from the development of a next-generation technology1 that will carry all information and services in a single network. Most of these functionalities depend on flow-based approaches (i.e., facilitating sequences of related data packets moving in a single direction between two endpoints) to guarantee quality of service (QoS). A core network can support flow-based services by employing the multiprotocol label-switching (MPLS) technique, but access networks, which connect subscribers to their providers, need additional schemes to accommodate them.
Dynamic flow allocation (DFA) is the mechanism by which an optical line terminal (OLT) allocates bandwidth to an optical network unit (ONU) and also assigns flow tags to individual data flows originating from a customer's equipment. After core network resources are reserved, the flow tag given by an OLT is determined by the customer's policy and service-level agreement. This tag is requested and assigned dynamically through the option field of the multipoint control protocol.
Figure 1 illustrates the DFA process. First, by choosing a flow type, a customer tries to send data through an access network (assuming the data type is not designated as short messages). Then, the customer premises equipment (CPE) transmits the data packets, along with the flow-type information, to an ONU. On receipt, the ONU sends a request for flow allocation and bandwidth assignment to an OLT in the form of a REPORT message. The OLT transfers the request to an edge router to reserve bandwidth and get a corresponding flow tag. Because the router belongs to a core network, which runs on the flow-based protocol, it must itself have the transfer resource control function to support bandwidth reservation and other resource controls. If it does not, the OLT should transfer the request to other control equipment that does.2
Figure 1. In the flow-based access mechanism, data sent through a customer's equipment is tagged and bandwidth is reserved as the data is passed to an edge router. CPE: Customer premises equipment. ONU: Optical network unit. OLT: Optical line terminal. ER: Edge router. BW: Bandwidth.
In addition, the OLT can guarantee bandwidth through both end-to-end and access networks because it gets the flow tag information from the edge router, which has already reserved core network resources. Thus, to avoid conflicts, connection admission control (CAC) should be required to block incoming traffic when needed, i.e., in the worst cases of network congestion. CAC will alert customers immediately when the CPE chooses an unavailable traffic type or tries to transmit without permission at the same time as another customer.
General weighted fair queuing (WFQ) is not appropriate for use with the flow-based access mechanism, since high-priority packets must yield precedence to low-priority packets when the latter belong to a flow being transmitted through an Ethernet passive optical network. In such cases, the high-priority data must wait to guarantee minimum bandwidth for the low-priority data. To meet these requirements, the WFQ should be modified to a constraint-based WFQ (C-WFQ). Figure 2 illustrates the C-WFQ algorithm. When the classified packets come into an ONU from the CPE, the ONU sorts them. If they belong to the flow being transmitted, the ONU will attach the same label and allocate the same bandwidth. Otherwise, the ONU allocates bandwidth by the WFQ algorithm.
The ONU also needs to compare the allocated bandwidth and the corresponding service-level agreement. If the terms of the agreement are satisfied, the packets are sent to an OLT. If, however, the ONU doesn't have sufficient resources to assign, it should reduce the bandwidth being used for other flows. For the most part, the ONU would retrench bandwidth belonging to lower-priority classes. If all the reducible resources belong to higher-priority classes, the ONU will cut those resources off and reallocate the bandwidth. In the worst case, if there is no marginal resource, the ONU should raise an alarm and drop incoming packets.
Figure 2. The scheduling algorithm for a C-WFQ identifies and routes packets. SLA: Service level agreement. WFQ: Weighted fair queuing.
DFA is the next-generation network's version of dynamic bandwidth allocation. Its main advantage is that it can allocate bandwidth to a specific customer's flows and also guarantee minimum bandwidth in the access area in accordance with the service-level agreement. DFA allows the service provider to offer guaranteed QoS before network congestion occurs. My colleagues and I have observed that the advantage afforded by the C-WFQ involves a tradeoff in terms of overall system performance, owing to the complexity and time required for resource reservations before calculating the algorithm. In the future, we plan to use flow tags to exploit the advantages of MPLS such as explicit routing.
Kyung-Hoon Bae received his BS degree from the department of electronic engineering at Kwangwoon University, Seoul, Korea, in 2001, and his MS and PhD degrees in electronic engineering there in 2003 and 2006, respectively. He also earned an MBA degree from Columbia Southern University, AL, in 2004. He currently works as a senior researcher. His research interests include communication, 3D vision, and robotics