THE ROLE OF THE MAC PROTOCOL IN OFFERING QOS TO IP SERVICES OVER SHARED ACCESS SYSTEMS
J. D. Angelopoulos, N. Leligou, Th. Orphanoudakis, G. Pikrammenos
National Technical University of Athens
Telecom Lab, Polytechniopolis Zographou
GR15753, Athens, Greece, e-mail:
Abstract
Shared access networks such as hybrid fiber/coaxial and passive optical networks have emerged as promising ways to reduce the cost of the transition to a broadband access infrastructure and provide a graceful upgrade path towards the photonization of the local loop. The TDMA multiplexing of traffic entering such a system is governed by the MAC protocol, which arbitrates the allocation of bandwidth to the shared feeder. At the same time the need to integrate telecom services presenting demanding quality requirements with plain best-effort services over the same infrastructure, brings new issues to the design of such an access mechanism. The differentiated services architecture with its bundling of behaviour aggregates is particularly suited to the H/W based MAC mechanisms enabling the support of diverse service needs, while aligning the system to the emerging Differentiated Services Internet strategy. The implementation of this mechanism in the AROMA research system and its evaluation using computer simulation are presented in this paper.
Introduction
Tree-shaped topologies present attractive cost advantages for broadband access networks by allowing many customers to share the expensive head-end equipment and the feeder section. In addition they offer re-use of the copper last drops to the customer, at least during the crucial introductory phase and probably for many years to come. Re-use of the existing infrastructure greatly reduces the initial investment outlay and provides a graceful upgrade path in step with service demand. Typical examples of tree topology systems are Hybrid Fiber Coaxial (HFC) and Passive Optical Networks (PON). The first use legacy CATV systems and their coaxial medium beyond the fiber node, while the latter re-use the twisted pairs of the telephone network enhanced with xDSL beyond the Optical Network Unit (ONU). In both systems, sharing of the upstream channel (from customer to network) is usually effected through TDMA multiplexing. A Medium Access Control (MAC) protocol is employed to arbitrate the access to the slotted common medium by allocating slots to customer terminations according to demand [5], [6].
This results in a distributed queuing system characterised by the long time required to pass control information from the queuing points to the service controller residing at the head-end. The allocation of upstream slots in a tree-topology access system is based on a reservation method which allows to dynamically adapt the bandwidth distribution to traffic fluctuations. The MAC controller collects access requests and then allocates the upstream slots by sending access permits. It is important to note that the service policy of the MAC governs the distributed multiplexing from a central point situated miles away at the head-end. This has the important implication that, as regards delay control, the acquisition of arrival information and buffer fill levels includes a considerable delay element not found in the centralized multiplexing typically operating in a switch queuing point. In contrast, drop policies must be distributed over all network terminations (cable modems) where the flow identity and fill levels are known. Because of the much larger reservation delay and statistical behaviour of the aggregations from many customers, special care must be taken to safeguard QoS to sensitive traffic. This requires a prioritization scheme and the differentiated services strategy of IETF [1] is a very suitable approach to handle the problem.
Service differentiation in an access multiplexer
From a traffic engineering point of view, the tree-shaped access system exhibits a very different behaviour between the upstream (from customers to core) and the downstream (from core to the customers) direction. In the downstream, the broadcast nature of the medium creates replicas of the signal (in the passive splitters of PONs or the taps of HFC systems) giving rise to privacy and security issues. These are typically dealt with by means of encryption, which is out of the scope of this work. In the upstream on the other hand, the MAC control function effects multiplexing / concentration of the traffic from all active modems as typically occurs in the output ports of any switching node. However in this case it is characterized by the distributed nature of the queuing points and the additional difficulties in the exchange of control information. Failure to recognize the packet precedence and behave accordingly would result in loss of QoS for any sensitive flow under high load. To apply any scheduling or priority discipline requires the correlation of the traffic from all multiplexed sources going to the one common egress point of the system.
The differentiated services (DS) strategy recently adopted by IETF as a scalable and relatively simple methodology towards enriching with QoS the IP services, is applicable and quite appropriate in the case of such tree-shaped access systems where IP services are dominant. To align such an access system to the DS concept requires the incorporation of provisions in the MAC function for the appropriate handling of each flow aggregation respecting its requirements. Each QoS class must encounter the specified Per-Hop-Behaviour (PHB) and this can only be embedded into the MAC control function.
The expedience that led to the DS architecture was the emergence of requirements for ISPs to offer QoS to certain flows in the presence of disturbing plain best-effort traffic without complex maintenance of state information in core routers. Eventually quality can not be guaranteed without some maintenance of state variables relating to reservations and flow intensities, but the strength of DS lies in the slow and graceful introduction of such complexities in line with revenues from a previous stage of introduction of such mechanisms. Dealing with behaviour aggregates and starting with static management based Service Level Agreements (SLAs) executed at slow timescales while keeping traffic conditioning at the edges of the network, enables a low cost starting phase while smaller granularity levels can be sought out at later stages of deployment. Non-compliant intermediate nodes can be transparent but at risk of reduced overall performance should they become the bottleneck in the route of the flow. Slow distributed access multiplexer such as an HFC system residing at the network edge can not be relied upon to operate in a transparent fashion as regards the DS strategy since the MAC directly affects the temporal properties of the egress stream.
Figure 1. Typical HFC system
Even before the emergence of the DS architecture it had become apparent that differentiation in the handling of traffic classes was required in the MAC function of PONs and HFC systems [5]. This was reflected in the adopted MAC solutions in the EU research projects PLANET and AROMA. These demonstrator systems supported both delay-sensitive and best-effort services. It was obvious that it was not acceptable to mix these two classes without unacceptable loss of grade of service for the former class, which employed preventive flow control. In contrast TCP/IP traffic uses flow control on the transport layer based on the detection of losses. Thus a TCP flow increases gradually the rate submitted to the network up to the point that buffer overflow occurs and re-transmission is triggered. Then the rate is reduced and a new balance point is sought. This action will disturb the other traffic classes, which can not benefit from re-transmission.
Given the fluidity of the standardisation for high QoS traffic and the several options open for PHB variants and the fact that the MAC protocol requires a significant hardware part, a flexible strategy is needed that can accommodate the changing environment. In the sections below, a MAC design approach is presented which can execute the currently available PHBs and adapt to a broader framework of service differentiation in the MAC protocol.
The basis of the approach is the use of access priorities in the reservation system which can be programmed to fit with required PHBs by means of S/W programming of the mapping of flows to priorities.
The system implemented in the framework of the AROMA project employs a TDMA slotting designed for ATM cells since native ATM services in addition to IP best-effort and QoS enhanced services are also supported. This required many more functions to map the IP flows to ATM VC/VPs using the Classical IP (CLIP) framework. However we do not deal with these other system implications of the ATM based solution which may have a narrower scope and we will simply consider the ATM size slots as a quantum of the MAC assigned bandwidth allocation. To accommodate IP frames, several slots are successively assigned by the MAC. However, even if IP is not offered over ATM, the legacy of short slots can be used to advantage in the context of the DS architecture over a slow shared link such as HFC. Namely, it allows to suspend the transmission of a lower priority packet, on the boundary of a slot (cell), and transmit delay-sensitive packets before resuming the transmission of the low priority one. In addition, fixed slots are easier to handle in a H/W based MAC. Such an ATM slotting is anyway the approach followed by the main standards bodies, like DAVIC 1.3 and 1.4, DVB/ETSI ETS 300 800, (now incorporated in DAVIC), and IEEE 802.14 [4].
The MAC as an executor of Per Hop Behaviours
As in the centralized multiplexer case, flows with demanding QoS (e.g. Expedited Forwarding or Assured Forwarding) must be identified and receive properly differentiated treatment from the plain best effort traffic. This is accomplished in the termination (ONU or cable modem) during packet classification by placing the corresponding cells in the high priority queue which will subsequently place higher priority requests and activate the higher priority permit allocation MAC algorithms.
To reduce complexity in this cost-sensitive residential access system, services are grouped into behaviour aggregates (classes) with a similar set of requirements. This is in line with the DS philosophy of flow aggregation for better scalability and flexibility. Given the fluidity of service class definition and the fact that the MAC is implemented mainly in H/W, it was deemed adequate to implement just 4 priority classes in the MAC, leaving finer aggregations and more elaborate forwarding policies, to be implemented in the router at the egress of the HFC system. All that is required from the MAC is not to deny quality to any groups of flows.
The characteristics of the four aggregation levels/priorities are the following:
· The high priority is devoted to delay-sensitive periodic CBR traffic which is supported by pre-allocated, unsolicited permits issued at fixed intervals by the MAC controller. This class is suitable for services with very strict delay requirements, which undergo strict traffic profile control (traffic conditioning) such as the EF (Expedited Forwarding) service [2].
· The second priority level is devoted to real-time variable rate flows, such as video services or VoIP and it is provided with peak rate policing for guaranteed QoS. MAC exercises a policing function by rate checking before issuing the permits. This is based on credit allocation at the time of subscription or connection set-up. In the DS context it could be used for the top AF (Assured Forwarding) class.
· The third priority is devoted to data services with higher requirements than best-effort. The traffic profile control assumed for this class aims at minimizing the loss of packets and the disturbance to other traffic. The credit scheme is used to guarantee a minimum rate (while credits last) while traffic exceeding this limit is relegated to the 4th priority permit generation. The 3rd priority mechanism is suited to the support of all four or the lower three AF classes [3]. (Drop policies can be independently applied at the modem queuing points).
· The fourth priority is reserved for plain best-effort services which employ loss based flow control at the TCP level and can be very disruptive to the other classes when sharing the same queue.
The last three priorities employ reservation while the first unsolicited permits. The reservations operate by means of three request fields totaling 8 bits. Two bits are used for the 2nd class and three bits for the other two creating a byte which is piggy-backed in every upstream slot. In the AROMA system this byte uses the place of the HEC field of the ATM cell which is not needed for cell delineation since an additional synchronization preamble is employed because of the burst mode operation. The total length of the upstream slot has in the AROMA system a size of 64 bytes accommodating except of the cell payload a synchronization preamble. The speed of the upstream is 3Mbps and each modem has a slot-by-slot agility over a triple of FDM channels which allows further dynamic statistical load balancing over all the customers of three channels, i.e. the MAC controller assigns not just a slot but also one of three channels. The frame structure in the downstream direction is dictated by several mandates of the physical layer (synchronization, flexibility over many modulation constellations and rates adapting to plant conditions, interleaving, Reed-Solomon FEC etc). These issues are covered in the relevant standards, e.g. [4], and will not be discussed here since we focus on the traffic multiplexing issues. For the operation of the MAC all that is important is that a number of permits is provided in a periodic basis corresponding to the number of upstream slots over the same period.