Recent advances in machine learning and hardware have produced embedded devices capable of performing real-time object detection with commendable accuracy. We consider a scenario in which embedded devices rely on an onboard object detector, but have the option to offload detection to a more powerful edge server when local accuracy is deemed too low. Resource constraints, however, limit the number of images that can be offloaded to the edge. Our goal is to identify which images to offload to maximize overall detection accuracy under those constraints. To that end, the paper introduces a reward metric designed to quantify potential accuracy improvements from offloading individual images, and proposes an efficient approach to make offloading decisions by estimating this reward based only on local detection results. The approach is computationally frugal enough to run on embedded devices, and empirical findings indicate that it outperforms existing alternatives in improving detection accuracy even when the fraction of offloaded images is small. Code for the paper's solution is available at https://github.com/qiujiaming315/edgeml-object-detection.

This paper considers networks where user traffic is regulated through deterministic traffic profiles, e.g., token buckets, and requires cleanrequires guaranteed clean hard delay bounds. The network’s goal is to minimize the resources it needs to meet those requirements clean bounds cleanbounds. The paper explores how reprofiling, i.e., proactively modifying how user traffic enters the network, can be of benefit. Reprofiling produces “smoother” flows but introduces an up-front access delay that forces tighter network delays. The paper explores this trade-off and demonstrates that, unlike what holds in the single-hop case, reprofiling can be of benefit cleanthat, unlike what holds in the single-hop case, reprofiling can be of benefit even when “optimal” clean“optimal” sophisticated clean schedulers are available at each hop cleanat each hop.

The need to guarantee hard delay bounds to traffic flows with deterministic traffic profiles, e.g., token buckets, arises in several network settings. It is of interest to offer such guarantees while minimizing network bandwidth. The paper explores a basic building block, namely, a single hop configuration, towards realizing such a goal. The main results are in the form of optimal solutions for meeting local deadlines under schedulers of varying complexity and therefore cost. The results demonstrate how judiciously modifying flows’ traffic profiles, i.e., reprofiling them, can help simple schedulers reduce the bandwidth they require, often performing nearly as well as more complex ones.

IoT devices are increasingly the source of data for machine learning (ML) applications running on edge servers. Data transmissions from devices to servers are often over local wireless networks whose bandwidth is not just limited but, more importantly, variable. Furthermore, in cyber-physical systems interacting with the physical environment, image offloading is also commonly subject to timing constraints. It is, therefore, important to develop an adaptive approach that maximizes the inference performance of ML applications under timing constraints and the resource constraints of IoT devices. In this paper, we use image classification as our target application and propose progressive neural compression (PNC) as an efficient solution to this problem. Although neural compression has been used to compress images for different ML applications, existing solutions often produce fixed-size outputs that are unsuitable for timing-constrained offloading over variable bandwidth. To address this limitation, we train a multi-objective rateless autoencoder that optimizes for multiple compression rates via stochastic taildrop to create a compression solution that produces features ordered according to their importance to inference performance. Features are then transmitted in that order based on available bandwidth, with classification ultimately performed using the (sub)set of features received by the deadline. We demonstrate the benefits of PNC over state-of-the-art neural compression approaches and traditional compression methods on a testbed comprising an IoT device and an edge server connected over a wireless network with varying bandwidth.

This article considers a setting where embedded devices are used to acquire and classify images. Because of limited computing capacity, embedded devices rely on a parsimonious classification model with uneven accuracy. When local classification is deemed inaccurate, devices can decide to offload the image to an edge server with a more accurate but resource-intensive model. Resource constraints, e.g., network bandwidth, however, require regulating such transmissions to avoid congestion and high latency. This article investigates this offloading problem when transmissions regulation is through a token bucket, a mechanism commonly used for such purposes. The goal is to devise a lightweight, online offloading policy that optimizes an application-specific metric (e.g., classification accuracy) under the constraints of the token bucket. This article develops a policy based on a deep$Q$-network (DQN), and demonstrates both its efficacy and the feasibility of its deployment on embedded devices. Of note is the fact that the policy can handle complex input patterns, including correlation in image arrivals and classification accuracy. The evaluation is carried out by performing image classification over a local testbed using synthetic traces generated from the ImageNet image classification benchmark. Implementation of this work is available athttps://github.com/qiujiaming315/edgeml-dqn.

As the role of virtualization technology becomes more prevalent, the range of applications deployed in virtualized systems is steadily growing. This increasingly includes applications with soft real-time requirements that benefit from low and predictable latency, even when co-located with other virtualized hosts with arbitrary traffic patterns. In this paper, we examine the policies and mechanisms affecting communication latency between virtual machines based on the Xen platform, and identify limitations that can result in long or unpredictable network stack latency for virtual machines deployed on this platform. To address these limitations, we propose and implementVATC, aVirtualization-Aware Traffic Controlframework that supports differentiation (via rate-limited prioritization) of outbound and inbound network traffic from co-located virtualized hosts. Results of our experiments show how and why VATC can offer predictable (soft) latency guarantees to applications running on virtualized hosts with minimum overhead.

The cloud's flexibility and promise of seamless auto-scaling notwithstanding, its ability to meet service level objectives (SLOs) typically calls for some form of control in resource usage. This seemingly traditional problem gives rise to new challenges in a cloud setting, and in particular a subtle yet significant trade-off involving load-distribution decisions (the distribution of workload across available cloud resources to optimize performance), and rate limiting (the capping of individual workloads to prevent global over-commitment). This paper investigates that trade-off through the design and implementation of a real-time messaging system motivated by Internet-of- Things (IoT) applications, and demonstrates a solution capable of realizing an effective compromise. The paper's contributions are in both explicating the source of this trade-off, and in demonstrating a possible solution.

The development and adoption of new protocols (or of extensions to existing protocols) is arguably central to the Internet's evolution. However, and in spite of over 40 years of experience with this process, we have limited understanding of what factors may contribute to a protocol's success. A sound technical design and a well-grounded purpose are obviously important, but we have many examples of failures that met those two criteria. What other factors affect a protocol's likelihood of success, and under what circumstances? We investigate this question through a statistical approach, based on analyzing a set of about 250 Internet standard documents, Internet engineering task force request for comments (RFCs). We characterize these RFCs using a number of key features, which we then seek to associate with positive or negative odds when it comes to success. Our high-level results are intuitive, e.g., protocols that call for Internet-wide adoption face greater challenges. Focusing on more targeted subsets of protocols reveals more subtle and possibly more interesting differences between areas of the Internet landscape. We also apply our prediction framework to IPv6, and use different “what-if” scenarios to explore what might have affected its deployment.

Multipath solutions have been shown to help improve throughput, reliability and/or load balancing. This paper seeks to understand if and when they benefit rate stability. Rate stability is important to many real-time, interactive applications, e.g., streaming video, but whether multipath solutions can help is unclear. Of relevance is the time-scale at which bandwidth changes are detected and acted upon to rebalance transmissions across paths. Consider two boundary cases: instantaneous detection and rate re-allocation, and a static rate assignment based on long-term path statistics. When transmissions can be instantaneously rebalanced across paths based on real-time link rate information, a multipath solution trivially improves rate stability (it all but eliminates rate variations). In contrast, when rate allocations are static, we find that multipath cannot improve upon the best single-path solution when buffers are large. When buffers are small (and coding is used to overcome losses), a multipath solution can, however, be beneficial even under a static rate allocation. The paper provides insight into when and how multipath solutions can help improve rate stability.

Network services often exhibit positive and negative externalities that affect users' adoption decisions. One such service is “user-provided connectivity” or UPC. The service offers an alternative to traditional infrastructure-based communication services by allowing users to share their “home base” connectivity with other users, thereby increasing their access to connectivity. More users means more connectivity alternatives, i.e., a positive externality, but also greater odds of having to share one's own connectivity, i.e., a negative externality. The tug of war between positive and negative externalities together with the fact that they often depend not just on how many but also which users adopt make it difficult to predict the service's eventual success. Exploring this issue is the focus of this paper, which investigates not only when and why such services may be viable, but also explores how pricing can be used to effectively and practically realize them.

This paper documents and to some extent elucidates the progress of IPv6 across major Internet stakeholders since its introduction in the mid 1990s. IPv6 offered an early solution to a well-understood and well-documented problem IPv4 was expected to encounter. In spite of early standardization and awareness of the issue, the Internet's march to IPv6 has been anything but smooth, even if recent data point to an improvement. This paper documents this progression for several key Internet stakeholders using available measurement data, and identifies changes in the IPv6 ecosystem that may be in part responsible for how it has unfolded. The paper also develops a stylized model of IPv6 adoption across those stakeholders, and validates its qualitative predictive ability by comparing it to measurement data.

What determines the eventual success of a protocol? Are certain features or properties more important? Do those vary according to a protocol's type? We explore these questions by applying data mining techniques to a rich repository of protocol specifications; IETF RFCs. While the investigation is still preliminary, some interesting findings have emerged. It confirms a number of intuitive results such as backward compatibility being key for protocol extensions and new versions, but not for new protocols. Similarly, the ability to improve performance is the single most important factor in the success of data plane protocols. Less intuitive findings, however, also emerge. Adding value to other protocols was the most significant factor in the success of new protocols, while extensions targeting security were the most likely to fail among new application and transport layer protocols. The paper offers a brief overview of our methodology and of the initial results it has afforded.

As virtualization technology becomes ever more capable, large-scale distributed applications are increasingly deployed in virtualized environments such as data centers and computational clouds. Many large-scale applications have soft real-time requirements and benefit from low and predictable latency, even in the presence of diverse traffic patterns between virtualized hosts. In this paper, we examine the policies and mechanisms affecting communication latency between virtual machines based on the Xen platform, and identify limitations that could result in long or unpredictable network traffic latencies. To address these limitations, we propose VATC, aVirtualization-Aware Traffic Controlframework for prioritizing network traffic in virtualized hosts. Results of our experiments show how and why VATC can improve predictability and reduce delay for latency sensitive applications, while introducing limited overhead.

The paper seeks to broaden our understanding of MPTCP and focuses on the impact that initial sub-path selection can have on performance. Using empirical data, it demonstrates that which sub-path is chosen to start an MPTCP connection can have unintuitive consequences. Using numerical analysis and a model-driven investigation, the paper elucidates and validates the empirical results, and highlights MPTCP's non-linear coupling between paths as a primary cause for this behavior. The findings are both of operational interest and may help design better MPTCP schedulers, as they are also exposed to complex interactions with MPTCP's congestion control.

The need for a larger Internet address space was acknowledged early on, and a solution (IPv6) standardized years ago. Its adoption has, however, been anything but easy and still faces significant challenges. The situation begs the questions of “why has it been so difficult?” and “what could have been (or still be) done to facilitate this migration?” There has been significant recent interest in those questions, and the paper builds on a line of work based on technology adoption models to explore them. The results confirm the impact of several known factors, but also provide new insight. In particular, they highlight the destabilizing effect of Internet Service Providers (ISPs) offering competing alternatives (to IPv6), and demonstrate the benefits of even minimum coordination among them in offering IPv6 as an option. The findings afford additional visibility into what affects technology transition in large systems with complex dependencies such as the Internet.

Today’s communication networks and networked systems are highly complex and heterogeneous \nand are often owned by profit-making entities. For new technologies or \ninfrastructure designs to be adopted, they must not only be based on sound engineering \nperformance considerations but also present the right economic incentives. Recent \nchanges in regulations of the telecommunication industry make such economic considerations \neven more urgent. For instance, new concerns such as network neutrality \nhave a significant impact on the evolution of communication networks. \nAt the same time, communication networks and networked systems support increasing \neconomic activity based on applications and services such as cloud computing, \nsocial networks, and peer-to-peer networks. These applications pose new challenges \nincluding the development of good pricing and incentive mechanisms to promote effective \nsystem-wide behavior. Similarly, the security and privacy of these applications are \nthemselves heavily dependent on economic considerations, which therefore need to be \nfully understood. \nTo address these questions, this special issue brings together a relevant set of state-of-the-art research contributions on complementary topics including communication \nnetworks, wireless networks, web content and security, and the use of multidisciplinary \napproaches ranging from game theory and economic modeling to algorithms \nand mechanism design, and including empirical studies.

The goal of this article is to develop a principled understanding of when it is beneficial to bundle technologies or services whose value is heavily dependent on the size of their user base, that is, exhibits positive exernalities. Of interest is how the joint distribution, and in particular the correlation, of the values users assign to components of a bundle affect its odds of success. The results offer insight and guidelines for deciding when bundling new Internet technologies or services can help improve their overall adoption. In particular, successful outcomes appear to require a minimum level of value correlation. Categories and Subject Descriptors (2012): Networks -- Network Algorithms -- Network economics; Networks -- Network properties -- Network dynamics; Information systems -- Information systems applications -- Collaborative and social computing systems and tools; Security and privacy -- Human and societal aspects of security and privacy; Human-centered computing -- Collaborative and social computing theory, concepts and paradigms

Wireless mesh networks hold the promise of rapid and flexible deployments of communication facilities. This potential notwithstanding, the often erratic behavior of multihop wireless transmissions is limiting the range of applications that such networks can target. In this paper we investigate the feasibility and benefits of a routing protocol explicitly aimed at making wireless mesh networks more predictable while preserving their efficiency and flexibility. The protocol's basic premise is the classical idea that a multipath solution can offer resiliency to unexpected link variations. The paper's contributions are in demonstrating how this can be effectively realized in a wireless context, and in offering initial evidences of its efficacy. In particular, the paper illustrates how routing decisions that account for link variability can be computed in a distributed fashion, and the benefits they afford in improving the stability of end-to-end transmission rates even in the presence of random network fluctuations.

User-provided connectivity (UPC) services offer a possible alternative, or complement, to existing infrastructure-based connectivity. A user allows other users to occasionally connect through its “home base” in exchange for reciprocation, or possibly compensation. This service model exhibits strong positive and negative externalities. A large user base makes the service more attractive, as it offers more connectivity options to roaming users, but it also implies a greater volume of (roaming) traffic passing through a user's home base, which can increase congestion. These interactions make it difficult to predict the eventual success of such a service offering, and in particular how to effectively price it. This paper investigates a two-price policy where the first price is an introductory price that expires once service adoption reaches a certain level. The paper uses a simplified analytical model to investigate pricing strategies under this policy, and their sensitivity to changes in system parameters. The insight and practical guidelines this yields are validated numerically under more realistic conditions.

Transitioning an infrastructure the size of the Internet is no small feat. We are in the midst of such a transition, i.e., from IPv4 to IPv6. IPv6 was standardized 15 years ago, but until recently there were few incentives to adopt it. The allocation of the last large block of IPv4 addresses changed that, and migrating to an IPv6 Internet has become more urgent. This migration is, however, still rife with uncertainties and challenges. The goal of this paper is to provide insight into this transition, and possibly make it smoother. The focus is on the "network," and the paper reports on extensive measurements that compare and contrast IPv6 and IPv4. Two important hypotheses, denoted as H1 and H2, were identified and validated. H1 argues that the IPv6 and IPv4 data planes now perform by and large comparably. In contrast, H2 points to routing differences as the primary culprit behind occurrences of poorer IPv6 performance. In other words, promoting IPv6 and IPv4 peering parity is probably the single most effective step towards equal IPv6 and IPv4 performance

With network components increasingly reliable, routing is playing an ever greater role in determining network reliability. This has spurred much activity in improving routing stability and reaction to failures and rekindled interest in centralized routing solutions, at least within a single routing domain. Centralizing decisions eliminates uncertainty and many inconsistencies and offers added flexibility in computing routes that meet different criteria. However, it also introduces new challenges, especially in reacting to failures where centralization can increase latency. This paper leverages the flexibility afforded by centralized routing to address these challenges. Specifically, we explore when and how standby backup forwarding options can be activated while waiting for an update from the centralized server after the failure of an individual component (link or node). We provide analytical insight into the feasibility of such backups as a function of network structure and quantify their computational complexity. We also develop an efficient heuristic reconciling protectability and performance, and demonstrate its effectiveness in a broad range of scenarios. The results should facilitate deployments of centralized routing solutions.

With network components increasingly reliable, routing is playing an ever greater role in determining network reliability. This has spurred much activity in improving routing stability and reaction to failures, and rekindled interest in centralized routing solutions, at least within a single routing domain. Centralizing decisions eliminates uncertainty and many inconsistencies, and offers added flexibility in computing routes that meet different criteria. However, it also introduces new challenges; especially in reacting to failures where centralization can increase latency. This paper leverages the flexibility afforded by centralized routing to address these challenges. Specifically, we explore when and how standby backup forwarding options can be activated, while waiting for an update from the centralized server after the failure of an individual component (link or node). We provide analytical insight into the feasibility of such backups as a function of network structure, and quantify their computational complexity. We also develop an efficient heuristic reconciling protectability and performance, and demonstrate its effectiveness in a broad range of scenarios. The results should facilitate deployments of centralized routing solutions.

This paper presents a new economic approach for studying competition and innovation in a complex and highly interactive system of network providers, users, and suppliers of digital goods and services (i.e., service providers). It employs Cournot and Bertrand games to model competition among service providers and network providers, respectively, and develops a novel unified model to capture the interaction and competition among these players in a "service-oriented" Internet. Incentives for service and network innovation are studied in this model.

Modern networks face the challenging task of handling increasingly diverse traffic that is displaying a growing intolerance to disruptions. This has given rise to many initiatives, and in this paper we focus on multiple topology routing as the primary vehicle for meeting those demands. Specifically, we seek routing solutions capable of not just accommodating different performance goals, but also preserving them in the presence of disruptions. The main challenge is computational, i.e., to identify among the enormous number of possible routing solutions the one that yields the best compromise between performance and robustness. This is where our principal contribution lies, as we expand the definition of critical links - a key concept in improving the efficiency of routing computation - and develop a precise methodology to efficiently converge on those solutions. Using this new methodology, we demonstrate that one can compute routing solutions that are both flexible in accommodating different performance requirements and robust in maintaining them in the presence of failures and traffic fluctuations.

Commercial agreements drive the routing policies used in today's Internet. The two most extensively studied commercial agreements are transit and peering; however, they are only two of many diverse and continuously evolving commercial agreements that ISPs enter into. So far, the only known practical safe and robust routing policy is Gao and Rexford's policy guideline, which is applicable to transit and peering agreements only. It is, therefore, of importance to identify routing policies that are safe and robust and, at the same time, capable of accommodating the diverse commercial agreements existing in the Internet. In particular, this paper investigates the extent to which routing policies can be devised to accommodate complex mutual transit agreements. We propose a series of policy guidelines that allow mutual transit agreements with progressively broader semantics to be established. Those policy guidelines guarantee routing safety and robustness as long as the autonomous system (AS) graph satisfies a corresponding set of precise topological constraints. An experimental evaluation of the proposed policy guidelines demonstrates the benefits they would likely afford in terms of routing reliability if adopted in the current Internet.

New network technologies constantly seek to displace incumbents. Their success depends on technological superiority, the size of the incumbent's installed base, users' adoption behaviors, and various other factors. The goal of this paper is to develop an understanding of competition between network technologies and identify the extent to which different factors, in particular converters (a.k.a. gateways), affect the outcome. Converters can help entrants overcome the influence of the incumbent's installed base by enabling cross-technology interoperability. However, they have development, deployment, and operations costs and can introduce performance degradations and functionality limitations, so that if, when, why, and how they help is often unclear. To this end, the paper proposes and solves a model for adoption of competing network technologies by individual users. The model incorporates a simple utility function that captures key aspects of users' adoption decisions. Its solution reveals a number of interesting and at times unexpected behaviors, including the possibility for converters to reduce overall market penetration of the technologies and to prevent convergence to a stable state, something that never arises in their absence. The findings were tested for robustness, e.g., different utility functions and adoption models, and found to remain valid across a broad range of scenarios.

802.11 wireless mesh networks are ubiquitous, but suffer from severe performance degradations due to poor synergy between the 802.11 CSMA MAC protocol and higher layers. Several solutions have been proposed that either involve significant modifications to the 802.11 MAC or legacy higher layer protocols, or rely on 802. MAC models seeded with off-line measurements performed during network downtime.

Modern networks face the daunting task of handling increasingly diverse traffic that is displaying a growing intolerance to disruptions. This has given rise to many initiatives, and in this paper we focus on multiple topology routing as the primary vehicle for meeting those demands. Specifically, we seek routing solutions capable of not just accommodating different performance goals, but also preserving them in the presence of disruptions. The main challenge is computational, i.e., to identify among the enormous number of possible routing solutions the one that yields the best compromise between performance and robustness. This is where our principal contribution lies, as we expand the definition of critical links -- a key concept in improving the efficiency of routing computation -- and develop a precise methodology to efficiently converge on those solutions. Using this new methodology, we demonstrate that one can compute routing solutions that are both flexible in accommodating different performance requirements and robust in maintaining them in the presence of failures and traffic fluctuations.

The Internet inter-domain routing protocol, BGP, experiences frequent routing disruptions such as transient routing loops or loss of connectivity. The goal of this paper is to address this issue while preserving BGP's benefits in terms of operational maturity and flexibility in accommodating diverse policies. In realizing this goal, we apply to inter-domain routing a common concept in the design of highly reliable systems, namely, the use of redundancy, which we introduce in a manner that maximizes compatibility with the existing BGP protocol. The basic idea is to run several, mostly unchanged BGP processes that compute complementary routes, so that in the presence of network instabilities a working path remains available to any destination. The paper outlines the design of this approach and compares it to previously proposed alternatives. The benefits of the scheme are demonstrated using actual BGP data and realistic simulations.

The convergence on IP of a wide variety of traffic types has strengthened the need for service differentiation. Service differentiation relies on two equally important components: (i) resource allocation, i.e., what resources does a given service class have access to; and (ii) contention resolution, i.e., how is access to shared resources arbitrated between services classes. The latter has been well studied with numerous mechanisms, e.g., scheduling and buffer management, supporting it in modern routers. In contrast, relatively few studies exist on the former, and in particular on the impact of routing that determines the resources a given service class is assigned to. This is the focus of the paper, which seeks to investigate how routing influences a network's ability to efficiently support different service classes. Of particular interest is the extent to which the ability to route service classes separately is beneficial. This question is explored for a base configuration involving two classes with either similar or entirely different service objectives (cost functions). The paper's contributions are in demonstrating and quantifying the benefits that the added flexibility of different (dual) routing affords, and in developing an efficient heuristic for computing jointly optimal routing solutions. The former can motivate the deployment of newly standardized multi-topology routing (MTR) functionality. The latter is a key enabler for the effective use of such capability.

Ethernet's plug-&-play feature is built on its use of flat (location independent) addresses and use of broadcasts to resolve unknown MAC addresses. While plug-&-play is one of Ethernet's most attractive features, it also affects its scalability. As the number of active MAC addresses in the network grows beyond the capacity of forwarding caches in bridges, the odds of "cache-misses," each triggering a broadcast, grow as well. The resulting increase in broadcast bandwidth consumption affects scalability. To address this problem, we propose a simple address resolution scheme based on an adaptation of distributed hash tables where a single query suffices in the steady state. The new scheme is implemented on advanced bridges maintaining backward compatibility with legacy bridges and eliminating reliance on broadcasts for address discovery. Comparisons with a legacy, broadcast-based scheme are carried out along several metrics that demonstrate the new scheme's robustness and ability to improve scalability.

1106-1118

Large-scale peer-to-peer systems span a wide range of Internet locations. Such diversity can be leveraged to build overlay "detours" to circumvent periods of poor performance on the default path. However, identifying which peers are "good" relay choices in support of such detours is challenging, if one is to avoid incurring an overhead that grows with the size of the peer- to-peer system. This paper proposes and investigates the Earliest Branching Rule (EBR) to perform such a selection. EBR builds on the Earliest Diverging Rule (EDR) that selects relay nodes whose AS path diverges from the default path at the earliest possible point, but calls for monitoring a much smaller number of paths. As a result, it has a much lower overhead. The paper explores the performance and overhead of EBR, and compares them to that of EDR. The results demonstrate that EBR succeeds in selecting good relay nodes with minimum control overhead. Hence, providing a practical solution for dynamically building good overlays in large peer-to-peer systems.

Packet-level diversity, or distributing packet transmissions over multiple, diverse channels, offers benefits in improving communication performance and robustness to channel variations. Previous works have analyzed and quantified those benefits, and developed transmission policies to realize them. However, translating those benefits into practice still faces numerous challenges from uncertainty in the adequacy of the channel models used to develop policies, to implementation dificulties in realizing the precise transmission schedules they mandate. This work is a first step in assessing what remains of those benefits once confronted with such practical challenges. Our investigation is carried out over an 802.11 testbed, where diversity is realized through the different frequency bands available for transmissions between hosts and access points. We use the testbed to evaluate the impact of transmission policies, channel characteristics, channel correlation, and various end-system constraints that affect our ability to precisely control transmissions timing. Our investigation reveals that in spite of the many gaps that exist between theory and practice, packet-level diversity still provides a simple solution to improve transmission performance and robustness across a broad range of configurations.

IP networks have seen tremendous growth in not only their size and speed, but also in the volume of traffic they carry. Over-provisioning is commonly used to protect network performance against traffic variations, be they caused by failures or transient surges. This paper investigates the influence that increasing network size has on the efficacy of over-provisioning in absorbing a certain range of traffic variations and preserving performance guarantees. For that purpose, we develop a general model that accounts for network topology, base offered traffic, and traffic variations, and allows us to explore how their combination behaves as the network and the traffic it carries grow. The model's generality enables us to investigate several representative scenarios and to identify critical thresholds in the relation between network and traffic growth, which delineate regions where a given amount of over-provisioning provides increasingly better protection against traffic variations. The results offer insight into how to grow IP networks in order to enhance their robustness

The current best-effort Internet cannot readily provide the service guarantees that VoIP applications often require. Path switching can potentially address this problem without requiring new network mechanisms, simply by leveraging the robustness to performance variations available from connectivity options such as multi-homing and overlays. In this paper, we evaluate the effectiveness and benefits of path switching in improving the quality of VoIP applications, and demonstrate its feasibility through the design and implementation of a prototype gateway. We argue for an application-driven path switching system that accounts for both network path characteristics and application-specific factors (e.g., codec algorithms, playout buffering schemes). We also develop an application path quality estimator based on the ITU-T E-model for voice quality assessment, and an application-driven path switching algorithm that dynamically adapts the time scales over which path switching decisions are made to maximize voice quality. Through network emulation and experiments over a wide-area multi-homed test bed, we show that, with sufficient path diversity, path switching can yield meaningful improvements in voice quality. Hence by exploiting the inherent path diversity of the Internet, application-driven path switching is a viable option in providing quality-of-service to applications.

In order to achieve better scalability, inter-domain signaling protocols rely on aggregation to reduce the amount of state information that routers are required to maintain. Nonetheless, they do not address another scalability key factor, the signaling load associated with establishing and maintaining reservations. Such load can be reduced if bandwidth is over-reserved. Over-reservation allows accommodating reservations without exchanging signaling messages, but may result in additional blocking. In this paper, we carry out a systematic investigation of the impact of over-reservation in different aggregation approaches, evaluating such impact in terms of the achieved signaling reduction, and blocking.

This paper investigates the problem of assessing the quality of video transmitted over IP networks. Our goal is to develop a methodology that is both reasonably accurate and simple enough to support the large-scale deployments that the increasing use of video over IP are likely to demand. For that purpose, we focus on developing an approach that is capable of mapping network statistics, e.g., packet losses, available from simple measurements, to the quality of video sequences reconstructed by receivers. A first step in that direction is a loss-distortion model that accounts for the impact of network losses on video quality, as a function of application-specific parameters such as the video codec and loss recovery technique, coded bit rate, packetization, video characteristics, etc. The model, although accurate, is poorly suited to large-scale, on-line monitoring, because of its dependency on many parameters that are difficult to estimate in real-time. As a result, we introduce a "relative quality" metric that bypasses this problem by measuring video quality against a quality benchmark that the network is expected to provide. The approach offers a lightweight video quality monitoring solution that is suitable for large-scale deployments. We assess its feasibility and accuracy through extensive simulations and experiments.

We investigate issues that Bluetooth may face in evolving from a simple wire replacement to a large-scale ad hoc networking technology. We do so by examining the efficacy of Bluetooth in establishing a connected topology, which is a basic requirement of any networking technology. We demonstrate that Bluetooth experiences some fundamental algorithmic challenges in accomplishing this seemingly simple task. Specifically, deciding whether there exists at least one connected topology that satisfies the Bluetooth constraints is NP-hard. Several implementation problems also arise due to the internal structure of the Bluetooth protocol stack. All these together degrade the performance of the network, or increase the complexity of operation. Given the availability of efficient substitute technologies, Bluetooth's use may end up being limited to small ad hoc networks.

Traffic engineering aims to distribute traffic so as to "optimize" some performance criterion. This optimal distribution of traffic depends on both the routing protocol and the forwarding mechanisms in use in the network. In IP networks running the OSPF or IS-IS protocols, routing is over shortest paths, and forwarding mechanisms distribute traffic "uniformly" over equal cost shortest paths. These constraints often make achieving an optimal distribution of traffic impossible. In this paper, we propose and evaluate an approach that can realize near optimal traffic distribution without changes to routing protocols and forwarding mechanisms. In addition, we explore the tradeoff that exists between performance and the configuration overhead that our solution requires. The paper's contributions are in formulating and evaluating an approach to traffic engineering in IP networks that achieves near-optimal performance while preserving the existing infrastructure.

This paper explores the differences that can exist between individual and aggregate loss guarantees in an environment where guarantees are only provided at the aggregate level. The focus is on understanding which traffic parameters are responsible for inducing possible deviations and to what extent. In addition, we seek to evaluate the level of additional resources, e.g., bandwidth or buffer, required to ensure that all individual loss measures remain below their desired target. This paper's contributions are in developing analytical models that enable the evaluation of individual loss probabilities in settings where only aggregate losses are controlled, and in identifying traffic parameters that have a major influence on the differences between individual and aggregate losses. The latter allows us to further construct practical tools and guidelines for rapidly assessing if specific traffic sources can be safely multiplexed into a common service class.

This work explores the feasibility of improving the performance of end-to-end data transfers between different sites through path switching. Our study is focused on both the logic that controls path switching decisions and the configurations required to achieve sufficient path diversity. Specifically, we investigate two common approaches offering path diversity multi-homing and overlay networks - and investigate their characteristics in the context of a representative wide-area testbed. We explore the end-to-end delay and loss characteristics of different paths and find that substantial improvements can potentially be achieved by path switching, especially in lowering end-to-end losses. Based on this assessment, we develop a simple path-switching mechanism capable of realizing those performance improvements. Our experimental study demonstrates that substantial performance improvements are indeed achievable using this approach.

Estimating end-to-end packet loss on Internet paths is important not only to monitor network performance, but also to assist adaptive applications make the best possible use of available network resources. There has been significant prior work on measuring and modeling packet loss in the Internet, but most of those techniques do not focus on providing real-time information and on assessing path performance from an application standpoint. In this paper, we present an online probing-based approach to estimate the loss performance of a network path, and extend this estimate to infer the performance that an application using the path would see. The approach relies on a hidden Markov model constructed from performance estimates generated from probes, which is then used to predict path performance as an application would experience. The accuracy of the model is evaluated using a number of different metrics, including loss rate and loss burstiness. The sensitivity of the results to measurement and computational overhead is also investigated, and an extension of the base approach using a layered model is explored as a possible solution to capturing time-varying channel behavior while keeping computational complexity reasonably low. The results we present show that the approach is capable of generating accurate, real-time estimates of path performance, and of predicting the performance that applications would experience if routed on the path

Today's Internet carries an ever broadening range of application traffic with different requirements. This has stressed its original, one-class, best-effort model, and has been one of the main drivers behind the many efforts aimed at introducing QoS. Those efforts have, however, experienced only limited success because their added complexity often conflict with the scalability requirements of the Internet. This has motivated many proposals that try to offer service differentiation while keeping complexity low. This paper shares similar goals and proposes a simple scheme, BoundedRandomDrop (BRD), that supports multiple service classes. BRD focuses on loss differentiation, as although both losses and delay are important performance parameters, the steadily rising speed of Internet links is progressively limiting the impact of delay differentiation. BRD offers strong loss differentiation capabilities with minimal added cost. BRD does not require traffic profiles or admission controls. It guarantees each class losses that, when feasible, are no worse than a specified bound, and enforces differentiation only when required to meet those bounds. In addition, BRD is implemented using a single FIFO queue and a simple random dropping mechanism. The performance of BRD is investigated for a broad range of traffic mixes and shown to consistently achieve its design goals.

The focus of this paper is on improving the quality of streaming video transmitted over the Internet. The approach we investigate assumes the availability of multiple paths between the source and the destination, and dynamically selects the best one. Although this is not a new concept, our contribution is in estimating the "goodness" of a path from the perspective of the video stream, instead of relying only on raw network performance measures. The paper starts by showing that the use of raw network performance data to control path switching decisions can often result in poor choices from an application perspective, and then proceeds to develop a practical approach for evaluating, in real-time, the performance of different paths in terms of video quality. Those estimates are used to continuously select the path that yields the best possible transmission conditions for video streaming applications. We demonstrate the feasibility and performance of the scheme through experiments involving different types of videos.

No abstract available.

Traffic engineering is aimed at distributing traffic so as to "optimize" a given performance criterion. The ability to carry out such an optimal distribution depends on both the routing protocol and the forwarding mechanisms in use in the network. In IP networks running the OSPF or IS-IS protocols, routing is over shortest paths, and forwarding mechanisms are constrained to distributing traffic uniformly over equal cost shortest paths. These constraints often make achieving an optimal distribution of traffic impossible. In this paper, we propose and evaluate an approach, based on manipulating the set of next hops for routing prefixes, that is capable of realizing near optimal traffic distribution without any change to existing routing protocols and forwarding mechanisms. In addition, we explore the tradeoff that exists between performance and the overhead associated with the additional configuration steps that our solution requires. The paper's contributions are in formulating and evaluating an approach to traffic engineering for existing IP networks that achieves performance levels comparable to that offered when deploying other forwarding technologies such as MPLS.

Current quality of service models such as those embodied in the differentiated services proposal, rely on data path aggregation to achieve scalability. Data path aggregation bundles into a single aggregate multiple flows with the same quality requirements, hence decreasing the amount of state to be kept. A similar scalability concern exists on the control path, where the state required to account for individual reservations needs to be minimized. There have been several proposals aimed at control path aggregation, and the goal of the paper is to expand on these works in an attempt to gain a better understanding of the various parameters that influence the efficiency of different approaches. In particular, we focus on inter-domain control aggregation, and compare an autonomous system (AS) sink-tree based approach with two examples of a shared AS segment based approach, in terms of the amount of state kept, both per AS and per edge router Our main contributions are in providing a greater understanding into the design of efficient control path aggregation methods.

In this paper, we wish to derive analytic models that predict the performance of TCP flows between specified endpoints using routinely observed network characteristics such as loss and delay. The ultimate goal of our approach is to convert network observables into representative user and application relevant performance metrics. The main contributions of this paper are in studying which network performance data sources are most reflective of session characteristics, and then in thoroughly investigating a new TCP model based on Padhye et al. (2000) that uses non-invasive network samples to predict the throughput of representative TCP flows between given end-points.

This paper explores, primarily by means of analysis, the differences that can exist between individual and aggregate loss guarantees in an environment where guarantees are only provided at an aggregate level. The focus is on understanding which traffic parameters are responsible for inducing possible deviations and to what extent. In addition, we seek to evaluate the level of additional resources, e.g., bandwidth or buffer, required to ensure that all individual loss measures remain below their desired target. The paper's contributions are in developing analytical models that enable the evaluation of individual loss probabilities in settings where only aggregate losses are controlled, and in identifying traffic parameters that play a dominant role in causing differences between individual and aggregate losses. The latter allows the construction of guidelines identifying what kind of traffic can be safely multiplexed into a common service class.

In this paper, we introduce and investigate a "new" path optimization problem that we denote the all hops optimal path (AHOP) problem. The problem involves identifying, for all hop counts, the optimal, i.e., minimum weight, path(s) between a given source and destination(s). The AHOP problem arises naturally in the context of quality-of-service (QoS) routing in networks, where routes (paths) need to be computed that provide services guarantees, e.g., delay or bandwidth, at the minimum possible "cost" (amount of resources required) to the network. Because service guarantees are typically provided through some form of resource allocation on the path (links) computed for a new request, the hop count, which captures the number of links over which resources are allocated, is a commonly used cost measure. As a result, a standard approach for determining the cheapest path available that meets a desired level of service guarantees is to compute a minimum hop shortest (optimal) path. Furthermore, for efficiency purposes, it is desirable to precompute such optimal minimum hop paths for all possible service requests. Providing this information gives rise to solving the AHOP problem. The paper's contributions are to investigate the computational complexity of solving the AHOP problem for two of the most prevalent cost functions (path weights) in networks, namely, additive and bottleneck weights. In particular, we establish that a solution based on the Bellman-Ford algorithm is optimal for additive weights, but show that this does not hold for bottleneck weights for which a lower complexity solution exists.

Over the past few years, there have been a number of proposals aimed at introducing different levels of service in the Internet. One of the more recent proposals is the Differentiated Services (Diff-Serv) architecture, and in this paper we explore how the policing actions and associated rate guarantees provided by the Expedited Forwarding (EF) translate into perceived benefits for applications that are the presumed users of such enhancements. Specifically, we focus on video streaming applications that arguably have relatively strong service quality requirements, and which should, therefore, stand to benefit from the availability of some form of enhanced service. Our goal is to gain a better understanding of the relation that exists between application level quality measures and the selection of the network level parameters that govern the delivery of the guarantees that an EF based service would provide. Our investigation, which is experimental in nature, relies on a number of standard streaming video servers and clients that have been modified and instrumented to allow quantification of the perceived quality of the received video stream. Quality assessments are performed using a Video Quality Measurement tool based on the ANSI objective quality standard. Measurements were made over both a local Diff-Serv testbed and across the QBone, a QoS enabled segment of the Internet2 infrastructure. The paper reports and analyzes the results of those measurements.

This paper provides an initial look at how support for advance reservations affects the complexity of the path selection process in networks. Advance reservations are likely to become increasingly important as networks and distributed applications become functionally richer and there have been a number of previous works and investigations that explored various related aspects. However, the impact or advance reservations on path selection is a topic that has been left largely untouched. This paper investigates several service models for advance reservations, which range from the traditional basic model of reserving a given amount of bandwidth for some time in the future, to more sophisticated models aimed at increasing the flexibility of services available through advance reservations. The focus is primarily on the issue of computational complexity when supporting advance reservations, and in that context, we derive a number of algorithms and/or intractability results for the various models we consider.

We discuss an implementation of QoS routing extensions to the open shortest path first (OSPF) routing protocol and evaluate its performance over a wide range of operating conditions. Our evaluations are aimed at assessing the cost and feasibility of QoS routing in IP networks. The results provide insight into the respective weights of the two major components of QoS routing costs, processing cost and protocol overhead and establish strong empirical evidence that the cost of QoS routing is well within the limits of modern technology and can be justified by the performance improvements.

This paper investigates the problem of making QoS guarantees available in access devices such as edge routers, that are commonly deployed in today's IP networks. We propose a specific design which we evaluate by carrying out a complete implementation, whose performance we then measure in the context of an experimental testbed. The results indicate that a reasonable level of service differentiation, i.e., rate and delay guarantees, can be provided with a minimal impact on the raw packet forwarding performance of edge devices.

In this paper, we review the basic mechanisms used in packet networks to support Quality-of-Service (QoS) guarantees. We outline the various approaches that have been proposed, and discuss some of the trade-offs they involve. Specifically, the paper starts by introducing the different scheduling and buffer management mechanisms that can be used to provide service differentiation in packet networks. The aim is not to provide an exhaustive review of existing mechanisms, but instead to give the reader a perspective on the range of options available and the associated trade-off between performance, functionality, and complexity. This is then followed by a discussion on the use of such mechanisms to provide specific end-to-end performance guarantees. The emphasis of this second part is on the need for adapting mechanisms to the different environments where they are to be deployed. In particular, fine grain buffer management and scheduling mechanisms may be neither necessary nor cost effective in high speed backbones, where “aggregate” solutions are more appropriate. The paper discusses issues and possible approaches to allow coexistence of different mechanisms in delivering end-to-end guarantees.

This paper investigates the problem of routing flows with quality-of-service (QoS) requirements through one or more networks, when the information available for making such routing decisions is inaccurate. Inaccuracy in the information used in computing QoS routes, e.g., network state such as link and node metrics, arises naturally in a number of different environments that are reviewed in the paper. The goal is to determine the impact of such inaccuracy on the ability of the path-selection process to successfully identify paths with adequate available resources. In particular, we focus on devising algorithms capable of selecting path(s) that are most likely to successfully accommodate the desired QoS, in the presence of uncertain network state information for the purpose of the analysis, we assume that this uncertainty is expressed through probabilistic models, and we briefly discuss sample cases that can give rise to such models. We establish that the impact of uncertainty is minimal for flows with only bandwidth requirements, but that it makes path selection intractable when end-to-end delay requirements are considered. For this latter case, we provide efficient solutions for special cases of interest and develop useful heuristics.

Recent studies provide evidence that Quality of Service (QoS) routing can provide increased network utilization compared to routing that is not sensitive to QoS requirements of traffic. However, there are still strong concerns about the increased cost of QoS routing, both in terms of more complex and frequent computations and increased routing protocol overhead. The main goals of this paper are to study these two cost components, and propose solutions that achieve good routing performance with reduced processing cost. First, we identify the parameters that determine the protocol traffic overhead, namely (a) policy for triggering updates, (b) sensitivity of this policy, and (c) clamp down timers that limit the rate of updates. Using simulation, we study the relative significance of these factors and investigate the relationship between routing performance and the amount of update traffic. In addition, we explore a range of design options to reduce the processing cost of QoS routing algorithms, and study their effect on routing performance. Based on the conclusions of these studies, we develop extensions to the basic QoS routing, that can achieve good routing performance with limited update generation rates. The paper also addresses the impact on the results of a number of secondary factors such as topology, high level admission control, and characteristics of network traffic.

In recent years, a number of link scheduling algorithms have been proposed that greatly improve upon traditional FIFO scheduling in being able to assure rate and delay bounds for individual sessions. However, they cannot be easily deployed in a backbone environment with thousands of sessions, as their complexity increases with the number of sessions. In this paper, we propose and analyze an approach that uses a simple buffer management scheme to provide rate guarantees to individual flows (or to a set of flows) multiplexed into a common FIFO queue. We establish the buffer allocation requirements to achieve these rate guarantees and study the trade-off between the achievable link utilization and the buffer size required with the proposed scheme. The aspect of fair access to excess bandwidth is also addressed, and its mapping onto a buffer allocation rule is investigated. Numerical examples are provided that illustrate the performance of the proposed schemes. Finally, a scalable architecture for QoS provisioning is presented that integrates the proposed buffer management scheme with WFQ scheduling that uses a small number of queues.

An important challenge for the future growth of the Internet is to design routers that can forward the exponentially increasing volume of traffic, and at the same time provide the service differentiation needed by new applications. In this paper, we describe the architecture, implementation, and initial experiences with a system designed to meet this challenge. This system, which we call a QoS capable switch-router (QSR), combines the salient features of switching and routing technologies to provide high throughput and support the different classes of service being defined by the IETF. It consists of a core (ATM) switch fabric connecting intelligent adapters, each capable of both routing and switching pockets. A control engine is responsible for routing, RSVP signalling, and resource management. We have built a prototype network of 3 systems connected to several UNIX hosts, and have conducted preliminary performance measurements on this network.

We investigate the problem of routing connections with QoS requirements across one or more networks, when the information available for making routing decisions is inaccurate and expressed in some probabilistic manner. This uncertainty about the actual state of a node or network arises naturally in a number of different environments, that are reviewed in the paper. The main focus is to determine the impact of such inaccuracies on the path selection process, whose goal is then to identify the path that is most likely to satisfy the QoS requirements.

This paper is motivated by the need to provide per-session quality of service guarantees in fast packet-switched networks. We address the problem of characterizing and designing scheduling policies that are optimal in the sense of minimizing buffer and/or delay requirements under the assumption of commonly accepted traffic constraints. We investigate buffer requirements under three typical memory allocation mechanisms which represent tradeoffs between efficiency and complexity. For traffic with delay constraints we provide policies that are optimal in the sense of satisfying the constraints if they are satisfiable by any policy. We also investigate the tradeoff between delay and buffer optimality, and design policies that are "good" (optimal or close to) for both. Finally, we extend our results to the case of "soft" delay constraints and address the issue of designing policies that satisfy such constraints in a fair manner. Given our focus on packet switching, we mainly concern ourselves with nonpreemptive policies, but one class of nonpreemptive policies which we consider is based on tracking preemptive policies. This class is introduced and may be of interest in other applications as well.

Ring network architectures that employ spatial reuse permit concurrent transmissions of messages over different links. While spatial reuse increases network throughput, it may also cause starvation of nodes. To alleviate this problem, various policies have been suggested in the literature. In this paper, we concentrate on a class of such policies that achieves fairness by allocating transmission quotas to nodes. For such policies, we provide mechanisms for improving delays and increasing overall throughput without compromising fairness.

This paper addresses the problem of providing per-connection end-to-end delay guarantees in a high-speed network. We assume that the network is connection oriented and enforces some admission control which ensures that the source traffic conforms to specified traffic characteristics. We concentrate on the class of rate-controlled service (RCS) disciplines, in which traffic from each connection is reshaped at every hop, and develop end-to-end delay bounds for the general case where different reshapers are used at each hop. In addition, we establish that these bounds can also be achieved when the shapers at each hop have the same "minimal" envelope. The main disadvantage of this class of service disciplines is that the end-to-end delay guarantees are obtained as the sum of the worst case delays at each node, but we show that this problem can be alleviated through "proper" reshaping of the traffic. We illustrate the impact of this reshaping by demonstrating its use in designing RCS disciplines that outperform generalized processor sharing-based service disciplines.

In this paper, we investigate some issues related to the efficient provision of end-to-end delay guarantees in the context of the Guaranteed (G) Services framework [16]. First, we consider the impact of reshaping traffic within the network on the end-to-end delay, the end-to-end jitter, as well as per-hop buffer requirements. This leads us to examine a class of traffic disciplines that use reshaping at each hop, namely rate-controlled disciplines. In this case, it is known that it is advantageous to use the Earliest Deadline First (EDF) scheduling policy at the link scheduler [8]. For this service discipline, we determine the appropriate values of the parameters that have to be exported, as specified in [16]. Subsequently, with the help of an example, we illustrate how the G service traffic will typically underutilize the network, regardless of the scheduling policy used. We then define a Guaranteed Rate (GR) service, that is synergetic with the G service framework and makes use of this unutilized bandwidth to provide rate guarantees to flows. We outline some of the details of the GR service and explain how it can be supported in conjunction with the G service in an efficient manner.

This paper addresses the problem of providing per-connection end-to-end delay guarantees in a high-speed network. We consider a network comprised of store-and-forward packet switches, in which a packet scheduler is available at each output link. We assume that the network is connection oriented and enforces some admission control which ensures that the source traffic conforms to specified traffic characteristics. We concentrate on the class of rate-controlled service (RCS) disciplines, in which traffic from each connection is reshaped at every hop, and develop end-to-end delay bounds for the general case where different reshapers are used at each hop. In addition, we establish that these bounds can also be achieved when the shapers at each hop have the same "minimal" envelope. The main disadvantage of this class of service discipline is that the end-to-end delay guarantees are obtained as the sum of the worst-case delays at each node, but we show that this problem can be alleviated through "proper" reshaping of the traffic. We illustrate the impact of this reshaping by demonstrating its use in designing RCS disciplines that outperform service disciplines that are based on generalized processor sharing (GPS). Furthermore, we show that we can restrict the space of "good" shapers to a family which is characterized by only one parameter. We also describe extensions to the service discipline that make it work conserving and as a result reduce the average end-to-end delays.

Addresses the problem of designing optimal buffer management policies in shared memory switches when packets already accepted in the switch can be dropped (pushed-out). The goal is to maximize the overall throughput, or equivalently to minimize the overall loss probability in the system. For a system with two output ports, the authors prove that the optimal policy is of pushout with threshold type (POT). The same result holds if the optimality criterion is the weighted sum of the port loss probabilities. For this system, the authors also give an approximate method for the calculation of the optimal threshold, which they conjecture to be asymptotically correct. For the N-ported system, the optimal policy is not known in general, but it is shown that for a symmetric system (equal traffic on all ports) it consists of always accepting arrivals when the buffer is not full, and dropping one from the longest queue to accommodate the new arrival when the buffer is full. Numerical investigations show that under the optimal POT policy the loss probability of a port is insensitive to traffic fluctuations in the other port.

We address the problem of designing optimal buffer management policies in shared memory switches when packets already accepted in the switch can be dropped (pushed-out). Our goal is to maximize the overall throughput, or equivalently to minimize the overall loss probability in the system. For a system with two output ports, we prove that the optimal policy is of push-out with threshold type (POT). The same result holds if the optimality criterion is the weighted sum of the port loss probabilities. For this system, we also give an approximate method for the calculation of the optimal threshold, which we conjecture to be asymptotically correct. For the N-ported system, the optimal policy is not known in general, but we show that for a symmetric system (equal traffic on all ports) it consists of always accepting arrivals when the buffer is not full, and dropping one from the longest queue to accommodate the new arrival when the buffer is full. Numerical results are provided which reveal an interesting and somewhat unexpected phenomenon. While the overall improvement in loss probability of the optimal POT policy over the optimal coordinate-convex policy is not very significant, the loss probability of an individual output port remains approximately constant as the load on the other port varies and the optimal POT policy is applied, a property not shared by the optimal coordinate-convex policy.>

Ring network architectures that employ spatial reuse permit concurrent transmissions of messages over different links. While spatial reuse increases network throughput, it may also cause starvation of nodes. To alleviate this problem, various policies have been suggested in the literature. In the paper the authors concentrate on a class of such policies that achieve fairness by allocating transmission quotas to nodes. For such policies, they provide mechanisms for improving delays and increasing overall throughput without compromising fairness.>

This paper is motivated by the need to support multiple service classes in fast packet-switched networks. The authors address the problem of characterizing and designing scheduling policies that are optimal in the sense of minimizing buffer and/or delay requirements under the assumption of commonly accepted traffic constraints. They investigate the buffer requirements under three typical memory allocation mechanisms, that represent trade-offs between efficiency and complexity. For classes with delay constraints they provide policies that are optimal in the sense of satisfying the constraints if they are satisfiable by any policy, and they also have low buffer requirements. They also address the issue of designing policies that satisfy delay constraints in a fair manner. They mainly concern ourselves with non-preemptive policies. One of the proposed policies is based on a class of non-preemptive policies that tracks preemptive policies. This class is introduced in this paper and may be of interest in other applications as well.>

Studies buffering policies which provide different loss priorities to packets/cells, while preserving packet ordering (space priority disciplines). These policies are motivated by the possible presence, within the same connection, of packets with different loss probability requirements or guarantees, e.g., voice and video coders or rate control mechanisms. The main contribution of the paper is the identification and evaluation of buffering policies which preserve packet ordering and guarantee high priority packets performance (loss probability), irrespective of the traffic intensity and arrival patterns of low priority packets. Such policies are termed protective policies. The need for such policies arises from the difficulty to accurately characterize and size low priority traffic, which can generate large and unpredictable traffic variations over short periods of time. The authors review previously proposed buffer admission policies and determine if they satisfy such "protection" requirements. Furthermore, they also identify and design new policies, which for a given level of protection maximize low priority throughput.>

Fast packet-switching has been chosen as the basis for future high speed, "universal" networks. The successful deployment of such networks will clearly depend on a wide range of factors such as cost and technology, but the authors believe that foremost among all is how well they will support existing and future applications. Emphasizing an application oriented perspective is one of the main motivation of the paper. The authors denote by "transparency" the ability of a network to transport application information while altering or manipulating it as little as possible, and believe it will be key to the acceptance of high-speed networks. While the concept is clearly not new, they articulate the need for it and illustrate its feasibility and the advantages it affords through the example of the plaNET network. In particular, they argue that a "transparent" data transfer mechanism can be provided that is both compatible with current standard proposals such as ATM and frame relay, and offers applications the choice of the data transfer mode that best meet their needs. A number of examples are used to illustrate these claims.>

Buffering policies that provide different loss priorities to packets/cells with no change in packet ordering (space priority disciplines) are studied. These policies are motivated by the possible presence, within the same connection, of packets with different loss probability requirements or guarantees. Examples of such applications are voice and video coders that generate information of unequal importance, and rate control mechanisms that mark excess traffic with a low priority rate violation tag. The focus is on the identification and evaluation of buffering policies that can guarantee performance, i.e. loss probability, to high priority packets irrespective of the traffic intensity and arrival patterns of low priority packets, while preserving the original ordering among packets. Such policies are termed protective policies.>

A family of queues for which the service time B/sub n/ of customer n depends on the interarrival time I/sub n/ between customers n-1 and n and the random variables I/sub n/ and B/sub n/ exhibit a proportionality relation is studied. In particular, the focus is on dependencies that arise naturally in communication systems, where the finite speed of the communication links constrains the amount of data that can be received in a given time interval. The simple case of a deterministic proportionality relation between the service time of a customer and its preceding interarrival time is considered and extended to allow the addition of an independent, generally distributed overhead to the service time. Several models that capture the on-off behavior of communication links in packet networks are then addressed. In all cases, expressions for the delay experienced by a packet in the system and illustrative numerical examples are provided.>

A family of queuing systems in which the interarrival time I/sub n+1/ between customers n and n+1 depends on the service time B/sub n/ of customer n is considered. Specifically, cases where the dependency between I/sub n+1/ and B/sub n/ is a proportionally relation and B/sub n/ is an exponentially distributed random variable is considered. Such dependencies arise in the context of packet-switched networks from employing rate policing functions which regulate the amount of data that can arrive at a link within any given time interval. The models developed and the associated solutions are, however, of independent interest and potentially applicable to other environments. Several scenarios that consist of adding an independent random variable to the interarrival time, allowing the proportionality to be random, and the combination of the two are considered. Numerical results are compared to those for an equivalent system without dependencies.>

A hierarchical decomposition of network nodes which permits the networkwide topology database and algorithms to be independent of node internals, yet allows (implicit) access to special intranodal features when required is described. A complementary procedure for the establishment of bandwidth-reserved connections is outlined that makes efficient use of this node structure to support network-level optimization through use of node-level features. Based on a significantly increased concurrency of path computation and bandwidth reservation, the execution time of the setup procedure is independent of the network's complexity (such as number of nodes and links) and is essentially bounded from above by the round-trip delay only.>

The authors investigate various performance measures of interest, when fast packet-switched networks that operate with either fixed or variable packet sizes are compared. These performance measures include queue length distribution, packet loss probability, and user frame loss probability. The focus is on identifying key parameters that influence the outcome of this comparison, and on quantifying the potential benefits of each approach.>

The authors present an analysis of a rate-based access control mechanism for high-speed networks that is based on the buffered leaky bucket scheme. The analysis assumes a discrete time environment representative of asynchronous transfer mode (ATM) networks and a batch arrival process that captures cell arrivals generated by segmentation of large user packets or superposition of a number of arrival streams. The solution method is based on matrix analytic techniques, but the particular structure of the system allowed for a number of important improvements. It is shown that the problem can be partitioned and that the matrix G, central to the matrix analytic technique, can be computed using exact recursive procedures instead of the traditional iterative approach. These improvements not only extend the range of systems that can be handled, but also eliminate computational issues such as convergence rate and stopping criterion.>

A family of queues is studied for which the service time B/sub n/ of customer n and the interarrival time I/sub n/ between customers n-1 and n exhibit some sort of proportionality. The focus is on dependencies that arise naturally in the context of communication systems, where the finite speed of the communication links constrains the amount of data that can be received in a given time interval. The simple case of a deterministic proportionality relation between the service time of a customer and its preceding interarrival time is considered. This is extended to allow the addition of an independent, generally distributed overhead to the service time of each customer. Several models that capture the ON-OFF behavior of communication links in packet networks are considered. In all cases, expressions for the delay experienced by a packet in the system are provided. Numerical examples illustrate the impact of dependencies through comparison with less accurate models. The results should be of relevance to environments other than communication as well.>

Considers a slotted ring in which simultaneous transmission of messages by different stations is allowed, a property referred to as spatial reuse. Ring networks with spatial reuse can achieve significantly higher throughput than standard token rings but they also introduce the possibility of starvation for some nodes on the ring. To alleviate this problem, various policies have been suggested in the literature. The present objective is to characterize the node throughputs achievable by general transmission policies in ring networks with spatial reuse and then to evaluate the throughput trade-off for a class of policies that has been proposed in the literature in order to avoid starvation. Specifically, the authors study a policy that is based on the idea of allocating transmission quotas to the nodes. Each node is guaranteed transmission of his quota within a specified interval. The authors show that by appropriately allocating the quotas, policies that satisfy general optimality criteria-in particular criteria related to fairness-can be designed. They also study the asymptotic behavior of the quota policy when either the quotas or the number of nodes increase.>

Asynchronous transfer mode (ATM) is being suggested as the basis for future high speed, universal networks. A key requirement for future ATM networks will be transparency, i.e. for the network to alter or manipulate the user information as little as possible. A transparent transport mechanism, plaNET, based on extensions of the current ATM standard and IBM's earlier PARIS technology, is proposed. It is shown how plaNET can satisfy the universal transport requirement of ATM, while avoiding some demonstrated deficiencies. In fact, plaNET could be viewed as an enhanced version of ATM that may be more suitable for the late 1990s than the current standard.>

The authors present an approach to computing access control parameters as a function of both source characteristics and bandwidth allocation in the network. The technique relies on a fluid-flow model for the source, the leaky bucket rate control system, and the bandwidth allocation process. The methodology is suitable for any high-speed packet switching architecture. The approach allows for the real-time setting of the access control at a connection setup. It ensures that the access control mechanism is near transparent as long as connections behave as expected, while protecting the network from most misbehaving connections. Although some extreme cases have the potential to affect the network, they are essentially due to limitations inherent in the leaky bucket itself.>

The output queues of an M*N packet switch are studied using a Markov-modulated flow model. The switching element is a central server which sequentially routes packets from the inputs to the outputs. The focus is on systems in which the server speed is such that the bulk of the queuing takes place in the output queues. The conventional point process approach neglects the impact of switching and transmission time. An attempt is made to account for these finite system speeds by using a Markov-modulated continuous flow to approximate the arrival process to an output queue. This model captures the dependency between arrivals at different outputs and reflects the fact that packet arrivals and departures are not instantaneous. The output queue content distribution is obtained, for both infinite and finite buffer systems, from the spectral expansion of the solution of a system of differential equations. Numerical examples and comparisons with the results of an M/M/1 approximation are presented.>

The authors propose a computationally simple approximate expression for the equivalent capacity or bandwidth requirement of both individual and multiplexed connections, based on their statistical characteristics and the desired grade-of-service (GOS). The purpose of such an expression is to provide a unified metric to represent the effective bandwidth used by connections and the corresponding effective load of network links. These link metrics can then be used for efficient bandwidth management, routing, and call control procedures aimed at optimizing network usage. While the methodology proposed can provide an exact approach to the computation of the equivalent capacity, the associated complexity makes it infeasible for real-time network traffic control applications. Hence, an approximation is required. The validity of the approximation developed is verified by comparison to both exact computations and simulation results.>

An N*N nonblocking packet switch with input queues and two priority classes that can be used to support traffic with different requirements is described. The switch operation is slotted and, at each time slot, fixed-size packets arrive at the inputs with distinct Bernoulli distributions for both the high- and low-priority classes. Two policies are discussed. In the first policy, packets of both priority classes are queued when waiting for service. In the second policy, only low-priority packets are queued, and high-priority packets not delivered at the first attempt are dropped from the system. Under both policies, high-priority packets prevail over low-priority packets at the inputs as well as the outputs. An approximate analysis that is based on independence assumptions and uses an equivalent queueing system to estimate the service capability seen be each input is presented. Using this approach, an expression for the input queue length distribution is obtained. The maximum system throughput is derived and shown to exceed that of a single priority switch. Numerical results are compared to simulations and are found to agree.>

A study is made of the output queues of an M*N packet switch using a Markov-modulated flow model. The switching element is a central server which sequentially routes packets from the inputs to the outputs. The authors focus on systems where the server speed is such that the bulk of the queueing takes place in the output queues. For such systems, accurate sizing of the output buffers is an important design issue and requires a correct characterization of the arrival processes to the output buffers. The conventional point process approach neglects the impact of switching and transmission time. An attempt is made to account for these finite system speeds by using a Markov-modulated continuous flow to approximate the arrival process to an output queue. This model captures the dependency between arrivals at different outputs and reflects the fact that packet arrivals and departures are not instantaneous. The output queue content distribution is obtained, for both infinite and finite buffer systems, from the spectral expansion of the solution of a system of differential equations. Numerical examples and comparisons with the results of an M/M/1 approximation are presented.>

A system of multiple primary queues with finite waiting rooms, where blocked jobs are allowed to overflow onto a common secondary queue, also with finite waiting room, is considered. Poisson arrivals and exponentially distributed service times, possibly different ones, are assumed at each primary service facility. Service facilities consist of a single server or multiple servers. Based on a simple iterative expression for the Laplace-Stieltjes transform of the interoverflow time distribution, the peakedness of the overflow process from primary queues is studied as a function of the number of primary servers and buffers as well as the original primary load. Behaviors and relations are identified for extreme cases (low offered load or large waiting rooms) and illustrated with examples. An approximation consisting of a natural extension of Hayward's formula to finite waiting room systems is proposed to estimate the blocking seen by the overflow traffic offered to the secondary service facility. Numerical examples covering a wide range of systems show that the approximation is good over a wide range of loads for systems with a single primary queue, as well as at high load for systems with multiple primary queues.>

An N*N nonblocking packet switch with two priority classes which can be used to support traffics with different requirements is analyzed. Real-time traffic such as voice is assigned high priority to satisfy its strict delay requirement, while data traffic uses a lower priority and utilizes the remaining system capacity. The switch operation is slotted, and at each time slot packets arrivals at the inputs have Bernoulli distributions with different probabilities for high-priority and low-priority classes. Packets of both priority classes can be queued when waiting for service; the service discipline is FCFS (first-come first-served) within each priority class. The authors provide explicit expressions for blocking probabilities and average delays. They also study system stability and show that the maximum throughput can in fact be higher than with no priority.>