Research challenges in QoS routing

Abstract

Quality of Service Routing is at present an active and remarkable research area, since most emerging network services require specialized Quality of Service (QoS) functionalities that cannot be provided by the current QoS-unaware routing protocols. The provisioning of QoS based network services is in general terms an extremely complex problem, and a significant part of this complexity lies in the routing layer. Indeed, the problem of QoS Routing with multiple additive constraints is known to be NP-hard. Thus, a successful and wide deployment of the most novel network services demands that we thoroughly understand the essence of QoS Routing dynamics, and also that the proposed solutions to this complex problem should be indeed feasible and affordable. This article surveys the most important open issues in terms of QoS Routing, and also briefly presents some of the most compelling proposals and ongoing research efforts done both inside and outside the E-Next Community to address some of those issues.

Introduction

The concept of Quality of Service (QoS) in communication systems is closely related to the network performance of the underlying routing system. To establish a common understanding for network QoS and particularly QoS Routing we depart from the ITU's definition of Quality of Service [1].

Definition: ‘Quality of Service—the collective effect of service performance which determines the degree of satisfaction of a user of the service.’

Fig. 1 shows the four major building blocks introduced in [1]: quality of service, serveability, trafficability performance, and dependability. To allow for implementation, the high-level concept of QoS can be mapped to service related primitives as described with the concept of serveability. The service performance is directly affected by the network performance. This ability of the network to meet the traffic demands is described by the concept of trafficability performance. Finally, dependability is a critical point impacting on the whole QoS network performance.

Routing can decisively contribute to the provision of QoS, and to the improvement of traffic performance and dependability in the ITU model. Although the merit of QoS Routing has long been recognized [2], a full-scale deployment is still missing. In this article we present an accurate description of the current state-of-the-art and enumerate the main open QoS Routing issues where significant effort and research is needed.

Next, we briefly introduce the reader to the main open QoS issues focussing on QoS Routing, writing down the main algorithmic, dynamic, architectural and dependability aspects. Subsequently these open issues are discussed in detail.

It is fair to state that the concept of Quality of Service (QoS) with its multidimensional service requirements was born in the late 1980 with the advent of ATM. Some years ago, QoS has been introduced in the Internet by a series of IETF contributions like Intserv, Diffserv, RSVP and MPLS. Currently, the IETF working group on traffic engineering is continuing to shape QoS induced features from the network provider's perspective. The interactivity of multimedia communication in the Internet is still increasing: real-time communication and QoS-awareness are regarded as valuable. Today, it is unclear what the role of QoS will be in newer types of networking such as mobile ad-hoc networks, sensor networks, WIFI and UMTS, grid computing, and overlay networking. In wired networks and especially in traditional telephony, network operators are facing the problem of replacing their relatively old classical telephony equipment, since the end of lifetime of switching fabrics is looming at the near horizon (2010). Their concern is the question whether it is possible or not to offer large-scale telephony (VoIP) over the current Internet with the conservation of the accustomed toll quality. In spite of the apparent importance of QoS, there does not seem to exist yet a business model for a QoS aware Internet. Perhaps the main importance of QoS lies in its lever function between economy (pricing) and technology (QoS Routing, QoS control, and QoS network management). But, undoubtedly the main disadvantage of QoS is the notorious complexity, which causes that QoS will only be implemented abundantly if we fully understand the QoS dynamics and can demonstrate its feasibility (in practice) and the associated economic gain.

The IETF QoS Routing working group was established as a continuation of the Birds of a Feather (BOF) session held at the IETF in June 1996 to discuss issues in Quality of Service Routing. The IETF QoS Routing working group has been stopped in the late 1990 s, mainly because the thorough understanding of the problem was still lacking. The moral seems to be that a theory and conceptual understanding of the problem is needed before the standards and not vice versa. Nevertheless, QoS Routing is a logically required architectural functionality, because all current IETF standards rely on traditional QoS-unaware routing. From this perspective, QoS Routing is the missing piece in a full-fledged QoS architecture for the Internet.

A conceptual difficulty with QoS in general starts already with the definition, and the same holds for a subpart of QoS, QoS Routing, to which this article is devoted. If we take the viewpoint that routing consists of a routing algorithm (static) and routing protocol (dynamics), then a QoS Routing algorithm solves the Multi-Constrained (Optimal) Path (MC(O)P) routing problem. In the MCP problem, each link u_v in a given graph is characterized by a link weight vector $w(u\to v)=\left[w_1,w_2,\dots ,w_m\right]$ with m positive real numbers w_i (u_v)≥0 as components. The MCP problem asks for a path P from a source node to a destination node that satisfies Eq. (1) for all 1≤i≤m QoS metrics, where L_i are the QoS constraints on the path.

$$w_i(P)\stackrel{def}{=}\sum _{(u\to v)\in P}{w}_i(u\to v)\le L_i$$

A path that satisfies all m constraints is often referred to as a feasible path. There may be many different paths in the graph that satisfy the constraints and, therefore, it might be desirable to retrieve the path with smallest length l(P) from the set of feasible paths. The problem that additionally optimizes some length function l(P) is called the Multi-Constrained Optimal Path (MCOP) problem. In addition to satisfying Eq. (1), the MCOP problem minimizes some length criterion such that l(P)≤l(P'), all paths P' between source and destination. A flexibility in the MC(O)P problem is the length criterion l(P) - the cost optimization function-which only needs to obey the properties of a norm of a vector. A difficulty of the MC(O)P problem is that it is NP-complete [3]. This classification essentially means that the time required to solve the MC(O)P problem exactly cannot, in the worst case, be upper-bounded by a polynomial function. Therefore the MC(O)P problem has been interpreted as intractable, which, in turn, has spurred the proposals of many heuristics. Only a few exact QoS Routing algorithms such as SAMCRA (Self-Adaptive Multiple Constraints Routing Algorithm [5]) exist. Although QoS Routing algorithmic issues still require attention, the larger part seems to be reasonably well understood.

The second component in QoS Routing, the QoS Routing protocol responsible for information exchange and for routing dynamics, is believed to be a far more difficult problem as outlined below. In short, the QoS Routing protocol consists of all the actions that inform individual nodes with a consistent and updated view on the network and the link weight structure.

Being the missing piece in the IETF QoS architectures and needing solutions for the MC(O)P problem and for the routing information dissemination protocol QoS Routing is definitely an excellent research subject in the area of computer networks. In order to substantiate this statement, the following subsections present some topics that deserve further study.

As stated in the previous Section, the algorithmic problem in QoS Routing, called the MC(O)P problem is NP-complete. Some of the proposed heuristics only target special cases of the MC(O)P problem. For instance, when bandwidth is one of the constraints that must be satisfied by the path computation algorithm, the MCP problem is defined as a Bandwidth Restricted Path (BRP) problem [6], [7], [8], [9], [10]. Another popular subproblem is called Restricted Shortest Path (RSP) problem [11], [12], [13]. In this case, all the paths that satisfy the constraint associated with one of the two metrics are computed and then the shortest path according to the second metric is selected. A straightforward method for heuristically solving the general MCP problem is via Metrics Combination (MC) [4], [14], [15], [16]. By combining a set of QoS metrics in a single metric, it is possible to use existing polynomial-time path computation algorithms, such as Bellman–Ford or Dijkstra.

Of course, when using exact QoS algorithms, QoS guarantees can be made, which is not possible (or can only be approximated) with heuristics. It is therefore desirable to be exact, but this may come at a high price in terms of execution time. Fortunately, the theory of NP-completeness is based on a worst-case analysis, and knowing what kind of network scenarios constitute a worst case is valuable (both in theory and in practice). Kuipers and Van Mieghem have distinguished in [17] several conditions that must hold simultaneously in order for worst cases to emerge: (1) the underlying topology must have a large expected hop-count, (2) the link weights can grow arbitrarily large or have an infinite granularity, which is not the case in practice, (3) there is a very negative correlation among the link weights, and (4) the constraint values are not too large nor too strict. These conditions are highly unlikely to reflect typical (practical) cases, suggesting that exact QoS Routing is feasible in practice.

In [18] and [19] the most relevant of QoS algorithms are described and evaluated via simulations: SAMCRA performed best. However, SAMCRA may possibly be improved, which requires a good understanding of the complexity of QoS Routing itself. If it can be demonstrated (rigorously) that QoS Routing possesses an acceptable complexity (hence, feasible in practice), then it may be regarded as a fundamental cornerstone and the consequences may be far-reaching.

To conclude, concerning the algorithmic aspects of QoS Routing several questions are still open. Some of them are the following:

• Can the computational efficiency of exact QoS Routing algorithms such as SAMCRA still be improved? If so, how can it be done?

• Can topologies be pruned a priori in order to reduce the computational effort?

• Can new computationally more efficient data structures be used instead of the Fibonacci-heaps used in SAMCRA [20]?

• Can ‘NP-complete’ topologies be detected a priori [17] and [21]? If so, by assigning proper link weights, network management may avoid these hard cases.

• Extensions to multicast QoS Routing such as MAMCRA [22] need to be explored further.

• Extensions to link-disjoint QoS Routing such as DIMCRA [18] need to be explored further.

The answer to the above questions is the subject of ongoing research work.

The current toughest problem that hampers the implementation of QoS in the Internet concerns the QoS Routing protocol. To enable QoS Routing, it is necessary to implement state-dependent, QoS-aware networking protocols. An example of such a protocol is PNNI, which uses link-state routing, in which every node tries to acquire a ‘map’ of the underlying network topology and the available resources via flooding. The available resources on a link are expressed by values, called link weights. Although simple and reliable, flooding involves unnecessary communications and causes inefficient use of resources, particularly in the context of QoS Routing that requires frequent distribution of multiple, dynamic parameters. Monitoring any change along the Internet is simply not possible and even not desirable, because not all changes are important. Two possible changes are considered:

• Infrequent changes due to joining/leaving of nodes. In the current Internet, only this kind of topology changes is considered. Its dynamics are relatively well understood.

• Frequent changes, which are typically related to the consumption of resources or to the traffic flowing through the network.

The link weight coupling to state information seriously complicates the dynamics of flooding because, contrary to infrequent changes, the flooding convergence time can be longer than the change rate of some metric (such as available bandwidth). The identification of the QoS characteristics and their characterization is determinant to the conception of QoS-aware routing protocols. QoS characteristics used to support the routing decision usually include bandwidth, loss rate, delay and jitter. Choosing the metrics upon which to base the routing decision is one of the main issues that must be addressed in a routing strategy because it determines simultaneously the characteristics that are offered to traffic and the complexity of the path computation algorithm. The selection of metrics must be done in order to increase the network self-awareness and service awareness.

The definition of issues related to metrics should contribute to increase the self-awareness and service awareness through the definition of the decisions concerning metrics selection and the mechanisms for metrics manipulation. The computation of QoS-aware paths requires that the routers obtain information about the state of the network in terms of the chosen metrics. The state of the network is composed of the local state of each node and of the global state that pertains to existing paths. The global state maintained by each node is obtained by the distribution of local states of the nodes that constitute the network.

An optimal update strategy for the infrequent changes is highly desirable in future multimedia networks that are characterized by the broad variability in traffic profiles and QoS requirements. No detailed update strategy for the infrequent changes has been published yet, although some descriptive papers have already appeared. Therefore the following points still deserve attention:

• What are the link weights w₁, w₂,…, w_m?. Type of metrics, number of metrics or relative significance of metrics.

• What is the influence of variations or inaccuracies (instabilities) on the link weights on the properties of the shortest (QoS) path? How can we handle it?

• Precision of metrics on the routing decision place

• What is the impact of aggregating routing information on the processing overhead? Would it be possible to reduce this processing overhead by means of path pre-computation?

• How do we determine, update and flood the link weight vectors? Is prediction possible?

• Proof of the QoS Routing conjecture ‘QoS Routing is near to optimal load balancing’. More precisely, consider a network that is loaded by reserving resources per source-destination pair using an exact QoS Routing algorithm on an instantaneously updated topology. If a steady state is reached, we conjecture that the consumption of the network resources will be close to an optimally loaded network. If true, dynamic QoS Routing would imply load balancing and load balancing need not be treated as a separate optimization step.

Further there is a topology range of interest: not all details of the entire global Internet are needed to determine a path from A to B. A sub-network encompassing A and B seems sufficient. In this respect, the properties of a network topology are very important. The Internet is shown to possess a power-law like degree distribution, while Ad-Hoc networks may vary from lattice structures to random graphs. Since paths strongly depend on both link weight structure and graph properties, the network dynamics will depend on these factors, even to the extent that some control strategies successful in a certain class of graphs may not work properly in other graphs.

The combination of QoS Routing algorithm and QoS Routing protocol forms the basis for a QoS architecture for the Internet. However, several issues are still open:

• Hierarchical QoS Routing: intra- and inter-domain QoS Routing

• How do we manage the QoS Routing fairness (co-existence of QoS flows/classes and best-effort)?

• How do we design a future save QoS Routing architecture (=both algorithm and protocol)?

• What is the level of detail required of the packet-level? What of the flow level? (=Architectural Issues & current RFCs)

• We need test bed verifications of proposed QoS Routing protocols and the influence of other control mechanisms as e.g. TCP.

• QoS Routing in wireless and Peer-to-peer networks.

• QoS Routing protocols are mostly evaluated by simulation, but how far can simulation go? Is a prototype implementation necessary?

The origins of dependability can be traced back to the early days of computing and communication as described in [23]. In the context of the early and pioneering work of Babbages, Larnder in 1834 proposed to eliminate errors in computation by using separate and independent computers and even more decisive by using different computation methods. Later, the first electronic computers and communication systems used highly unreliable components. As a result the research focussed on enhancing the reliability and dependability of operation—a first step towards QoS. Basic theories of redundancy to enhance the reliability of logical structures and to enhance the quality of communication have been developed from von Neumann, Moore, Shannon and their successors and are still the basis for our work. Today, the fundamental concept of dependability in computer/communication systems is discussed from a technical perspective in various research groups and committees including the joint initiative of the International Federation for Information Processing (IFIP WG 10.4 on dependable computing and fault tolerance) and the IEEE computer society (IEEE TC-FTC—technical committee on fault-tolerant computing) IEEE [24].

We now focus on the aspects of dependability that are closely related to communication networks and especially the Internet. Dependable operation of the routing system is part of the QoS Routing agenda since the early days of the Internet. For example, the predecessor of the Internet, the ARPANET, suffered from catastrophic failures because of its routing protocol, which could only be repaired with manual intervention (see, for example [25] for details of this malfunction). Based on this experience, the Internet community decided to require routing protocols to fulfil some basic dependability criteria such as, for example, the ability of the protocol to stabilize after the failure condition is removed (self-stabilization). Influenced by the failure of the ARPANET, routing protocols for the Internet have been kept very simple, though. Even today and despite the fact of high application QoS demands, the Internet lives without QoS and without QoS-capable routing mechanisms. We conclude that the dependability and survivability of the core transport functionality even under extreme conditions makes up one important point in the QoS Routing open agenda. We define routing dependability to be:

‘Routing dependability is the trustworthiness of a routing system such that reliance can justifiably be placed on the consistency of behaviour and performance of the routing service it delivers.’ [26]

To be able to design dependable QoS Routing systems, it is necessary to better understand the dimensions of routing dependability. These dimensions are not fixed, however, but are influenced by the characteristics of the investigated network. For the example of mobile and wireless communications (see also Section 5 of this article) we find some important characteristics to influence routing dependability to be [26]:

• User and end system mobility.

• The wireless nature of the communication channel.

• The routing strategies/algorithms and routing protocols, i.e., the adaptation to changing network conditions on various time-scales as well as the overhead induced.

• The infrastructure-based, infrastructure-less, or hybrid nature of the routing systems.

• The limitations in energy-resources.

• Asymmetric capabilities of nodes in heterogeneous networks.

• Cooperation vs. non-cooperation of network nodes in ad hoc networks.

• External forces, like environmental conditions.

We have described the main open QoS issues and have clearly justified the need for QoS Routing, given that the main goals of this article are in fact both to state those open issues as well as to present the most recent and significant contributions (some of them from E-Next partners) addressing such issues. This article is split in different sections covering a significant spectrum of the recent and future work to be done in QoS Routing. Section 2 focuses on intra-domain routing, describing recent work and new proposals in such a routing scenario. Section 3 extends the QoS Routing problem to inter-domain routing, also describing the most recent activities carried out on this topic. Optimization issues are analyzed in Section 4. Afterwards, in Section 5 we extend QoS Routing to wireless networks. Being aware of the main target of this article, we introduce in Section 6, as a brief summary, the main points of interests of the partners of E-Next involved in the writing of this article. Finally, Section 7 concludes the article.