For well over a quarter century, detection systems have been driven by models learned from input features collected from real or simulated environments. An artifact (e.g., network event, potential malware sample, suspicious email) is deemed malicious or non-malicious based on its similarity to the learned model at runtime. However, the training of the models has been historically limited to only those features available at runtime. In this paper, we consider an alternate learning approach that trains models using privileged information--features available at training time but not at runtime--to improve the accuracy and resilience of detection systems. In particular, we adapt and extend recent advances in knowledge transfer, model influence, and distillation to enable the use of forensic or other data unavailable at runtime in a range of security domains. An empirical evaluation shows that privileged information increases precision and recall over a system with no privileged information: we observe up to 7.7% relative decrease in detection error for fast-flux bot detection, 8.6% for malware traffic detection, 7.3% for malware classification, and 16.9% for face recognition. We explore the limitations and applications of different privileged information techniques in detection systems. Such techniques provide a new means for detection systems to learn from data that would otherwise not be available at runtime.

Traffic anomaly detection is primarily concerned with identifying malicious traffic patterns in a much larger stream of benign traffic. Traditionally, this is achieved by selecting a very specialized set of traffic-based features that are used for both training a model, as well as for detection at runtime. This paper introduces a novel method of anomaly detection that breaks the assumption that the same set of features should be used for both training and runtime detection. Building upon the concept of Learning Using Privileged Information (LUPI), this paper shows how to build a detection model that uses an extended set of features, some that are available only during training, and some also available at runtime, with results significantly superior to the traditional learning methods. We demonstrate the usage of the privileged features in detecting malicious botnet activities in a network, activities shown to be challenging to detect in past works.

This article describes a method for constructing a special rule (we call it synergy rule) that uses as its input information the outputs (scores) of several monotonic rules which solve the same pattern recognition problem. As an example of scores of such monotonic rules we consider here scores of SVM classifiers. In order to construct the optimal synergy rule, we estimate the conditional probability function based on the direct problem setting, which requires solving a Fredholm integral equation. Generally, solving a Fredholm equation is an ill-posed problem. However, in our model, we look for the solution of the equation in the set of monotonic and bounded functions, which makes the problem well-posed. This allows us to solve the equation accurately even with training data sets of limited size. In order to construct a monotonic solution, we use the set of functions that belong to Reproducing Kernel Hilbert Space (RKHS) associated with the INK-spline kernel (splines with Infinite Numbers of Knots) of degree zero. The paper provides details of the methods for finding multidimensional conditional probability in a set of monotonic functions to obtain the corresponding synergy rules. We demonstrate effectiveness of such rules for 1) solving standard pattern recognition problems, 2) constructing multi-class classification rules, 3) constructing a method for knowledge transfer from multiple intelligent teachers in the LUPI paradigm.

We introduce a method called multi-scale local shape analysis for extracting features that describe the local structure of points within a dataset. The method uses both geometric and topological features at multiple levels of granularity to capture diverse types of local information for subsequent machine learning algorithms operating on the dataset. Using synthetic and real dataset examples, we demonstrate significant performance improvement of classification algorithms constructed for these datasets with correspondingly augmented features.

We introduce a constructive setting for the problem of density ratio estimation through the solution of a multidimensional integral equation. In this equation, not only its right hand side is approximately known, but also the integral operator is approximately defined. We show that this ill-posed problem has a rigorous solution and obtain the solution in a closed form. The key element of this solution is the novel V-matrix, which captures the geometry of the observed samples. We compare our method with previously proposed ones, using both synthetic and real data. Our experimental results demonstrate the good potential of the new approach.

Radial basis function (RBF) kernels for SVM have been routinely used in a wide range of classification problems, delivering consistently good performance for those problems where the kernel computations are numerically feasible (high-dimensional problems typically use linear kernels). One of the drawbacks of RBF kernels is the necessity of selecting the proper value of the hyperparameter γ in addition to the standard SVM penalty parameter C — this process can lead to overfitting. Another (more obscure) drawback of RBF is its inherent non-optimality as an approximation function. In order to address these issues, we propose to extend the concept of polynomial splines (designed explicitly for approximation purposes) to multidimensional normalized splines with infinite number of knots and use the resulting hyperparameter-free kernel SVMs instead of RBF kernel SVMs. We tested our approach for a number of standard classification datasets used in the literature. The results suggest that new kernels deliver mostly better classification performance than RBF kernel (for problems of moderately large dimensions), but allow faster computation (if measured on large cross-validation grids), with less chance of overfitting.

Due to its obvious importance, fault detection and localization is a well-studied problem in communication networks, as attested by the many techniques designed to address this problem. The inherent variability, limited component reliability, and constrained resources of MANETs (Mobile Ad hoc Networks) make the problem not just more important, but also critical. Practical development and deployment considerations imply that fault detection and localization methods must i) avoid relying on overly detailed models of network protocols and traffic assumptions and instead rely on actual cross-layer measurements/observations, and ii) be applicable across different network scales and topologies with minimum adjustments. This paper demonstrates the feasibility of such goals, and proposes an important and as yet unexplored approach to fault management in MANETs: network-invariant fault detection, localization and diagnosis with limited knowledge of the underlying network and traffic models. We show how fault management methods can be derived by observing statistical network/traffic measurements in one network, and subsequently applied to other networks with satisfactory performance. We demonstrate that a carefully designed but widely applicable set of local and weak global indicators of faults can be efficiently aggregated to produce highly sensitive and specific methods that perform well when applied to MANETs with varying sizes, topologies, and traffic matrices.

Load balancing in publish/subscribe (pub/sub) broker networks is challenging as the workload is multi-dimensional and content-dependent. In this paper we present the framework design of a middleware, called Shuffle, to achieve optimal load balancing in a pub/sub broker network. Shuffle features a suite of active workload management schemes within a single overlay topology on message parsing, matching, delivery, and forwarding, the four types of workload in a publish/subscribe service affected by two inputs-streaming events and stored subscriptions. Shuffle leverages its traffic randomization scheme and Chord, a DHT substrate, to build overlay trees for active workload aggregation and distribution, and we show the optimality property of the load balancing scheme upon any input traffic distribution on individual Shuffle aggregation trees. We also show the NP-hardness of the workload management problem when it has to be done among multiple correlated aggregation trees, and present a heuristic accordingly. Through extensive simulations we validated the design of Shuffle upon dynamic and heavy workload.

Service hosting platforms typically run a large number of third-party applications that are composed of multiple communicating components distributed on a dynamic set of hosting servers. Understanding the real-time behaviors of these applications and the intricate interactions/dependency relationships among these application components is very important to service management tasks such as load balancing, capacity planning, performance debugging and fault diagnosis. In this paper, we present the scalable real-time application monitoring and diagnosis (SRAMD) tool, for applications consisting of "black box" components: software without source code available, and usually without desired logging instrumentation. SRAMD runs at application layer and requires no modification to existing applications, middleware, or messages. For each application component collocated at its hosting server, a SRAMD monitor traces the component's packet-level traffic unobtrusively, summarizes its local resource utilization and performance (e.g. response time) online, performs interactive queries (e.g. per- request resource utilization) to locate possible bottlenecks on- demand, and discovers inter-component dependence relationships statistically. The SRAMD controller simply aggregates reports from distributed monitors to construct real-time application topologies with rich runtime information. We have developed mechanisms to decentralize the computation overhead and minimize the communication cost in the monitoring and diagnosis process, and two schemes to discover application component dependency relationships in different scenarios. The SRAMD tool offers an alternative to server logs and message-level traces for service monitoring and performance diagnosis.

In this paper, we exploit the flexible features in the medium access control (MAC) layer of WiMax for construction and transmission of MAC protocol data units (MPDU) for supporting multiple VoIP streams over a WiMax link. Quality of VoIP calls, usually given by R-score, is studied with respect to delay and loss of packets. We observe that loss is more sensitive than delay, hence we trade delay for loss. We propose a combination of techniques that exploit the flexibility of the WiMax MAC layer to strike a balance between loss and delay. These techniques are forward error correction, automatic repeat request, MPDU aggregation, and minislot allocation. Simulation experiments are conducted to test the performance of the proposed mechanisms. We assume a three-state Markovian channel model and study the performance with and without retransmissions. We show that the feedback-based technique coupled with retransmissions, aggregation, and variable length MPDUs are effective and increases the R-score by about 40%.

Abstract — We present Sieve – an infrastructure for fast content-based data distribution to interested users. The ability of Sieve to filter and forward high-bandwidth data streams stems from its distributed pipelined architecture. The complex message filtering task is broken-up into a sequence of light-weight filtering components resulting in high end-to-end throughput. Furthermore, since each component is assigned to a node based on its resource constraints, the queue buildup inside the nodes is minimal resulting in low end-to-end latency. Our experimental results based on real system implementation show that Sieve can sustain a throughput of more than 5000 messages per second for 100000 subscriptions with predicates of 10 attributes. Index Terms — Content-based Information Dissemination, Publish-Subscribe System, Event Stream Filtering

Abstract — Performance in multihop wireless networks is known to degrade with the number of hops for both TCP and UDP traffic. For VoIP, the wireless network presents additional challenges as the perceived quality is dependent on both loss and delay. We investigate several methods to improve voice quality and present experimental results from an 802.11b testbed optimized for voice delivery. Use of multiple interfaces, path diversity and aggregation are shown to provide a combined improvement of 13 times in number of calls supported in our 15 node 802.11 mesh system. VoIP, multihop, mesh I.

In this paper we design and evaluate Ganges, a wireless mesh network architecture that can efficiently transport real-time video streams from multiple sources to a central monitoring station. Video quality suffers from deterioration in the presence of bursty network losses and due to packets missing their playback deadline. Ganges spatially separates the paths to reduce inter-flow contention. It finds out a fair rate allocation for the different video sources. The wireless routers in the mesh network implement several optimizations in order to reduce the end-to-end delay variation. Ganges improves the network capacity by over 10-15% over a shortest path tree, and improves the video picture quality by 7-10 dB

Performance in multihop wireless networks is known to degrade with the number of hops for both TCP and UDP traffic. For VoIP, the wireless network presents additional challenges as the perceived quality is dependent on loss, delay and jitter. Furthermore, small size of VoIP packets aggravates network utilization of the 802.11 based wireless mesh network. In this work, we evaluate the packet aggregation algorithms to reduce VoIP protocol overhead, compare the aggregation algorithms and propose a distributed packet aggregation mechanism. The experimental results demonstrates that the proposed distributed packet aggregation algorithm improves the VoIP quality and results in significant increase in the number high quality VoIP calls supported over multihop mesh.

Current web search engines are retrospective in that they limit users to searches against already existing pages. Prospective search engines, on the other hand, allow users to upload queries that will be applied to newly discovered pages in the future. We study and compare algorithms for efficiently matching large numbers of simple keyword queries against a stream of newly discovered pages.

In the recent past, there has been a tremendous increase in the popularity of VoIP services as a result of huge growth in broadband access. The same voice-over-Internet protocol (VoIP) service poses new challenges when deployed over a wireless mesh network, while enabling users to make voice calls using WiFi phones. Packet losses and delay due to interference in a multiple-hop mesh network with limited capacity can significantly degrade the end-to-end VoIP call quality. In this work, we discuss the basic requirements for efficient deployment of VoIP services over a mesh network. We present and evaluate practical optimizing techniques that can enhance the network capacity, maintain the VoIP quality and handle user mobility efficiently. Extensive experiments conducted on a real testbed and ns-2 provide insights into the performance issues and demonstrate the level of improvement that can be obtained by the proposed techniques. Specifically, we find that packet aggregation along with header compression can increase the number of supported VoIP calls in a multihop network by 2-3 times. The proposed fast path switching is highly effective in maintaining the VoIP quality. Our fast handoff scheme achieves almost negligible disruption during calls to roaming clients

A substantial portion of the emerging wireless data service consists of non-real-time applications such as content download. The existing mechanisms based on per-packet performance guarantees used mainly for voice and streaming media do not suffice for the elastic nature of non-real-time traffic. For a non-real-time user data services, the key performance measure of interest is the total download time. In this paper, we propose a novel scheduling framework for wireless content service. Specifically, we present a two-layer scheduling architecture that combines content-aware scheduling with opportunistic scheduling. In terms of content-awareness, the proposed scheduling policy provides guarantees on the download time of content. In the second stage, the instantaneous channel conditions of different users are exploited in an opportunistic manner so as to maximize the throughput of the system. We define service differentiation in two modes - differential and guaranteed - and provide polynomial time algorithms for both that manipulate the stretch ratio but within allowable limits. Extensive simulations are conducted that verify the efficiency of the proposed schemes and provide insights into the behavior of the scheduling algorithms for non-real-time data.

We present SPIDER-a system for fast replication or distribution of large content from a single source to multiple sites interconnected over Internet or via a private network. In order to exploit spatial diversity of the underlying network, SPIDER uses an overlay structure composed of dedicated transit nodes (TNs). The data transport mechanism in SPIDER leverages this overlay structure to provide a coordinated approach that minimizes the maximum time to replicate to all destination sites (the make span of content replication). In order to achieve this objective, SPIDER employs two orthogonal components: a) creation of multiple dynamic distribution trees using the transit nodes b) end-to-end reliable data transport with flow control on these trees by chaining point-to-point TCPs. We further present simulations based results to quantify benefits of tree construction algorithms in random topologies. We evaluate the real implementation of the SPIDER in Planet Lab and observe a 2-6 times speed up compared to different existing schemes.

Non-uniform coverage and low system throughput due to location dependent fading in cellular networks is a major bottleneck in providing quality of service. Moreover, the growing number of stand-alone Wi-Fi hotspots is creating competition for 3G cellular networks. We propose an architecture which augments the existing 3G downlink technology with Wi-Fi like relay nodes leveraging benefits of both forms to improve the 3G system performance and provide uniform coverage throughout the cell. The relay nodes in the proposed WRN (Wi-Fi like relay network) architecture are equipped with a dual radio interface to communicate with the base station and other relay nodes. We compare the performance of the existing CDMA/HDR-based 3G cellular network with the proposed WRN architecture in terms of throughput and blocking probability. We demonstrate how pre-engineered deployment of relay nodes can yield better performance than random deployment of relay nodes. Simulation experiments are conducted to corroborate our analytical findings. The system throughput is not only enhanced, but the fairness among users is also maintained.

The IEEE 802.16 standard (commonly known as WiMAX), which has emerged as a broadband wireless access technology, is capable of delivering very high data rates. However, providing performance guarantees to delay sensitive applications like streaming media is still a challenge. We study the media access control (MAC) layer of WiMAX and exploit its flexible features to construct MAC packet data units (MPDU) dynamically. The sizes of the MPDUs are constantly modified based on channel state information. The desired payload is obtained either by aggregation or fragmentation of the upper layer data units. The robustness of MPDUs is also made tunable by means of cyclic redundancy code bits. We consider both the scenarios - with and without feedback. We adhere to the 802.16 specifications and propose adapting the MPDU length for streaming media for better performance. Three metrics are defined: restore probability; goodput; dropping probability. Simulation experiments are conducted which show the performance enhancements of the proposed ARQ-enabled adaptive algorithm in terms of these three metrics.

Hybrid hierarchical optical cross-connects leverages the benefits of both all-optical (OOO) and optical-electrical (OEO) switching for a cost-effective solution in future optical networks. The design goal using hybrid cross-connects is to minimize the sum of two cost metrics: OEO switching cost and OOO switching cost. This paper proposes two different waveband merging architectures and minimum cost routing and merging algorithms for offline traffic scenario. The simulation based results show that the proposed algorithms provide higher cost benefit as compared to existing algorithms that do not rely on waveband merging architecture.

Non-real time content delivery constitutes a major portion wireless data services in 3G networks. The per-packet performance guarantees required for real-time applications, such as voice and streaming media, is achieved through conventional flow-based scheduling mechanisms. However, for data services, such scheduling mechanisms cannot be used since the key performance measure for non-real time applications is the total download time. In this paper, we present a two-layer scheduling architecture that combines content-aware scheduling with opportunistic scheduling. In terms of content-awareness, the proposed scheduling policy provides guarantees to the download time of a content. Furthermore, through parallelizing the schedule and using opportunistic channel usage, the scheme achieves higher throughput. Extensive simulations verify the efficiency of our proposed schemes and provide insights into the behavior of the scheduling algorithms.

Non-uniform coverage is a major concern in cellular data networks based on HSDPA/HDR access technologies. Poor coverage lowers the overall utilization of the cell and results in location-dependent downlink throughput for mobile users. We focus on the planning of ad hoc relay network (ARN) in providing an improved cellular coverage. Specifically, we present and discuss issues and approaches for relay node placement in cellular space. Through extensive simulation modeling, we provide the evaluation of the improvement in the location dependent cellular data rate by employing the ARN.

The huge demand and growth of high-speed mobile data applications drive the development of the next generation wireless systems. The emerging 3G cellular networks provide new wireless access technologies for high-speed downlink packet access. The variable nature of wireless channel quality brings up challenges and opportunities for wireless system design. By utilizing channel state information, the next generation wireless packet systems provide high data rate communications with adaptive modulation and coding. To provide quality of service (QoS) in such systems is an important engineering issue. In this paper, we propose a novel scheduling algorithm that supports assured QoS. Due to the unpredictability and volatility of wireless medium, wireless QoS design tends to focus on soft QoS guarantee instead of hard one. The proposed scheme can effectively guarantee soft rate reservation and provide high-speed connections with channel-aware radio resource allocation without compromising fairness.

We present a QoS-enhanced routing framework designed for an administrative system that runs a QoS-aware link-state routing protocol supporting MPLS. This routing framework is based on using only edge nodes and does not require any modifications at the core. The framework also does not employ any bandwidth reservation; instead, congestion is relieved by adaptive rerouting of aggregated traffic flows. Route oscillation is prevented by using randomization-based back-off techniques for triggering route changes. The proposed mechanism is distributed and non-cooperative, which makes it suitable for practical implementation. The results of simulations demonstrate that a significant reduction of packet drop ratio can be achieved with a moderate number of path shifts.

Aggregation of individual wavelengths into wavebands for their subsequent switching and routing as a single group is an attractive way for scalable and cost-efficient optical networks. We analyze the implications of this waveband hierarchy for a single optical node by analyzing two issues: the proper selection of waveband sizes and the assignment of wavebands for a limited set of input-output patterns of traffic. We formulate a general model and propose optimal algorithmic solutions for both problems. The performance of resulting sets of nonuniform wavebands is studied for several representative cases (a single node, an optical ring network, an optical mesh network). The results demonstrate improved optical throughput and reduced cost of switching and routing when using nonuniform waveband hierarchy.

Dynamic replication schemes for cost-effective, scalable content delivery networks (CDNs) are discussed. Based on content replication granularity, current dynamic replication approaches can be classified into entire replication (entire content is replicated) and object replication (individual objects are replicated) schemes. The drawbacks of these schemes are low replication performance and low scalability, respectively. We propose two schemes delivering replication performance close to that of object replication scheme with replication granularity close to that of entire replication scheme. The schemes aggregates objects in the origin server into multiple content groups and then replicate them on per-group granularity. The content groups are either statically created in advance or dynamically created upon replication. Simulation of the proposed scheme showed that the scheme of dynamic group creation can achieve almost the same performance as the optimal scheme of object replication. Even the scheme of static group creation provides much better performance than entire replication under skewed request distribution. The effect of the number of content groups on performance is also investigated.

We study the path provisioning as a mechanism to deliver service level agreements in IP Differentiated Services networks. There is no known polynomial time solution for this problem, so we propose and analyze (by simulations) several heuristic algorithms for solving the problem. As our simulations demonstrate, a centralized server consistently delivers a better performance than a distributed solution. We also show that the performance of one of the proposed algorithms, the greedy algorithm with backtracking, can be very close to the optimal one, while being computationally feasible.

We consider hierarchical optical networks with different granularities of paths (wavelengths and wavebands). We introduce classification layers for routing models in these networks (separated and integrated routing, homogeneous and heterogeneous networks, online and offline routing). We also propose and analyze two heuristic routing and aggregation algorithms (online and offline) to be used for homogeneous networks in the separate routing framework. The analysis is based on ring topology; its results are supported by simulations of large scale networks (Europe and USA). These simulations demonstrate a significant cost reduction achieved by employing hierarchical routing (from 33% in the online algorithm to almost 60% in the offline one).

Faster processors are quickly enabling a new class of computationally intensive applications that actively transform information flows. Performing such computation at the terminal end is contrary to current trends toward low-power terminal devices. Moreover, scalability and efficiency concerns are also encouraging service providers to outsource computation when operating under loaded conditions. To address the problem of deploying such applications, we introduce the JOURNEY network model, which provides computation as an integrated network service. Contrary to other distributed computing models, JOURNEY does not attempt to guarantee that a given computational job will indeed be completed. Instead, the JOURNEY model trades off such hard guarantees in favor of architectural simplicity, and hence scalability. In order for the JOURNEY model to be applicable to real-time multimedia flows, we introduce the notion of soft quality-of-service (QoS) that provides a probabilistic bound on the unprocessed packet rate (UPR). Based on this notion, we describe a packet-processing admission control algorithm that additionally takes into consideration a flow's real-time constraints. We also propose a computing router architecture that is based on cluster technology. This architecture can track technology advances in both routing and computing independently. We further present a motivating multimedia application that employs a semantically driven video transcoding service within the JOURNEY framework we implemented, and describe our experience along with performance measurements.

We propose a PIM-SM based IP-multicast routing framework for delivering heterogeneous quality of service to multicast receivers. We propose two tree construction algorithms: TIQM (relying on the full availability of tree-specific information about a multicast group) and NUQM (not requiring any tree specific information). While TIQM is able to compute pseudo-optimal QoS-constrained trees, it has a control-scalability problem. This problem is overcome in NUQM by restricting the amount of information used during tree computation. For realistic networks, under low to moderate network load and receivers' bandwidth requirements, performance of NUQM is close to that of TIQM. A QoS-extended intra-domain PIM-SM framework is also presented.

Since real-time variable bit rate (VBR) traffic is inherently bursty, dynamic bandwidth allocation is necessary for ATM streams that carry VBR traffic. In order to provide quality-of-services (QoS) guarantees and to reduce the computational complexity, an hybrid of guaranteed and dynamic adaptive allocation scheme requires to be implemented. Typical dynamic allocations to competing streams are done in the form of linear proportions to the bandwidth requirements. We show that during temporary link congestion such proportional arrangements can give rise to unequal queue growth and, subsequently, degraded QoS. This is found to be true even for streams that belong to the same VBR class and share identical long term traffic characteristics and QoS requirements. In this paper, four allocation algorithms are presented and analyzed in terms of their fairness and QoS potential for real-time VBR traffic. We propose and show that a novel allocation strategy, termed Minmax, solves the mentioned problem of unfairness within a class. By maintaining a fair distribution of buffer length across the streams of a class, the proposed policy can achieve better and fairer QoS performance compared to the traditional methods. We present analytical results, proofs and a simulation study of the described algorithms. Four allocation policies for handling MPEG VBR video streams are simulated in the context of a wireless ATM (WATM) medium access control. The results show that in certain scenarios, the Minmax strategy can reduce losses by an order of magnitude, while decreasing delays substantially.

In this paper, we propose a technique for reducing asynchronous transfer mode (ATM) call blocking which is achieved by splitting wide-band connections into multiple low-bandwidth subconnections and routing them independently through the network. The essence of the mechanism is to use fragmented network bandwidth for supporting calls which are otherwise blocked by conventional routing. ATM bandwidth fragmentation may take place in a situation when a connection occupies a part of a link bandwidth and the rest of it is not sufficient for another connection. The unused bandwidth becomes fragmented. We provide a detailed cell-level design for the split-scheduling algorithms, which use a special type of ATM resource management cell for maintaining cell ordering. The analysis and simulation of the scheduling algorithms show that connection splitting is capable of delivering acceptable cell-level quality of service to multiple traffic classes. We also deliver a solution for implementing splitting without requiring any protocol changes within the network. We show that it is sufficient to modify the data and control plane protocols only within the end stations. Finally, a set of routing-level simulations for splitting demonstrates that splitting can reduce blocking by up to 45% for high-bandwidth calls in a moderately loaded network. Considering that it does not incur any network expenses or protocol changes, we propose splitting as an efficient means for reducing connection blocking.

We describe a mobile call admission control algorithm (M-CAC) for mobile multimedia terminals roaming within coverage areas consisting of multiple microcells. The method targets consistent soft-QoS for admitted connections with high utilization of radio resources, exploiting both temporal and spatial statistical multiplexing. We compare the performance of M-CAC to the performance of the default algorithm (all calls are accepted, provided there is enough bandwidth in the originating microcell). Results of our simulations demonstrate that a reasonably good connection-level soft-QoS can be achieved with moderate increase of cost of connections (determined by efficiency of bandwidth utilization).

We propose a new private network-to-network interface (PNNI) aggregation algorithms for hierarchical ATM networks. Standard PNNI routing provides a scalable hierarchical routing capability to support a large ATM network, and its hierarchy mechanism can decrease the network complexity significantly. However, in order to support effective QoS-based routing across foreign subnetworks, the topology aggregation mechanism, representing the lower level network topology to higher level, becomes a key issue. PNNI specification defines general notions of this aggregation mechanism, leaving enough opportunities for subnetwork administrators to choose the amount of aggregated information that they advertise. So far, no feasible PNNI compliant algorithms have been proposed. We propose a new three step aggregation scheme to solve the problem. We evaluated the proposed scheme, based on large-scale simulations with different network topologies, traffic scenarios, and performance metrics. The proposed scheme delivers good network performance while significantly reducing the network complexity.

The performance of two previously proposed asymptotically stable unicast flow control algorithms for high-speed networks is compared. We extend one of them to the multicast case. The multicast algorithm is modeled by a set of delay-differential equations. Despite the difficulties in analyzing the second order system with multiple different delays, the algorithm is analytically proven to be stable, and this is supported by extensive simulations. This work should prove useful in the design of multicast flow control algorithms.

Current designs of high-speed networks assume that all customers are tolerant of some amount of losses. However, it is possible that some applications may require very high reliability, and would be willing to pay more for it, if such a service were available. Motivated by this, we propose a design of a hybrid network which can guarantee zero cell loss probability for type 1 traffic while allowing some losses for type 2 traffic. This paper has three contributions. Our first contribution in this paper is to propose a service discipline (which can be implemented easily on a specific switch architecture) which guarantees zero losses for type 1 traffic. Our second contribution is to propose an algorithm for a scheduling strategy which reduces the number of buffers required at the output pods of the switches to zero for type 1 traffic. Our last contribution is to solve a difficult queueing problem involving service interruptions, which characterizes the performance of type 2 traffic.

The authors analyze a threshold priority access statistical multiplexing technique with a Markov modulated fluid source model for variable-bit-rate (VBR) video traffic, and an exact constant-bit-rate (CBR) model. Exact expressions for loss probabilities and delay are obtained. Based on these quality-of-service parameters, buffer size requirements of the statistical multiplexer are obtained, and the performance of a fixed threshold algorithm is compared with that of a rate-based threshold algorithm. The results provide the tradeoffs between buffer size and processing complexity for control. The exact analytical results can be used to calculate buffer size requirements.>