Data in the form of pairwise comparisons arises in many domains, including preference elicitation, sporting competitions, and peer grading among others. We consider parametric ordinal models for such pairwise comparison data involving a latent vector $w^* \in \mathbb{R}^d$ that represents the “qualities” of the $d$ items being compared; this class of models includes the two most widely used parametric models---the Bradley-Terry-Luce (BTL) and the Thurstone models. Working within a standard minimax framework, we provide tight upper and lower bounds on the optimal error in estimating the quality score vector $w^*$ under this class of models. The bounds depend on the topology of the comparison graph induced by the subset of pairs being compared, via the spectrum of the Laplacian of the comparison graph. Thus, in settings where the subset of pairs may be chosen, our results provide principled guidelines for making this choice. Finally, we compare these error rates to those under cardinal measurement models and show that the error rates in the ordinal and cardinal settings have identical scalings apart from constant pre- factors.

Consider the problem of identifying the underlying qualities of a set of items based on measuring noisy comparisons between pairs of items. The Bradley-Terry-Luce (BTL) and Thurstone models are the most widely used parametric models for such pairwise comparison data. Working within a standard minimax framework, this paper provides sharp upper and lower bounds on the optimal error in estimating the underlying qualities under the BTL and the Thurstone models. These bounds are are topology-aware, meaning that they change qualitatively depending on the comparison graph induced by the subset of pairs being compared. Thus, in settings where the subset of pairs may be chosen, our results provide some principled guidelines for making this choice. Finally, we compare these error rates to those under cardinal measurement models and show that the error rates in the ordinal and cardinal settings have identical scalings apart from constant pre-factors. We use this result to investigate the relative merits of cardinal and ordinal measurement schemes.

In this paper, we propose a novel incentive mechanism for message delivery services based on delay tolerant networks (DTNs). In recent years, the amount of mobile data traffic has grown rapidly due to the increase in global smartphone shipments. Some research has focused on offloading the mobile data traffic through other types of networks, such as DTNs, Wi-Fi access points, and femtocells. From the viewpoint of mobile operators, smartphone users need to be involved in order to reduce the amount of mobile data traffic via 3G/LTE networks. As a way of supporting the delivery of messages by providing surplus resources from users' smartphones DTNs look like a promising solution. However, in a DTN-based message delivery service, smartphone users who do not want to obtain the messages may also join such a store-carry-forward scheme. Therefore, it follows that smartphone users need to be motivated to transmit the messages to other users. In this paper, we propose a reward-based incentive mechanism for such a service, where the mobile operator announces a total reward to be shared among smartphone users, and smartphone users provide surplus resources for disseminating messages to other users. Consequently, we show how a win-win relationship between mobile operator and smartphone users can be created with our incentive mechanism.

We propose, analyze and implement a general architecture for massively parallel VoD content distribution. We allow for devices that have a wide range of reliability, storage and bandwidth constraints. Each device can act as a cache for other devices and can also communicate with a central server. Some devices may be dedicated caches with no co-located users. Our goal is to allow each user device to be able to stream any movie from a large catalog, while minimizing the load of the central server. First, we architect and formulate a static optimization problem that accounts for various network bandwidth and storage capacity constraints, as well as the maximum number of network connections for each device. Not surprisingly this formulation is NP-hard. We then use a Markov approximation technique in a primal-dual framework to devise a highly distributed algorithm which is provably close to the optimal. Next we test the practical effectiveness of the distributed algorithm in several ways. We demonstrate remarkable robustness to system scale and changes in demand, user churn, network failure and node failures via a packet level simulation of the system. Finally, we describe our results from numerous experiments on a full implementation of the system with 60 caches and 120 users on 20 Amazon EC2 instances. In addition to corroborating our analytical and simulation-based findings, the implementation allows us to examine various system-level tradeoffs. Examples of this include: (i) the split between server to cache and cache to device traffic, (ii) the tradeoff between cache update intervals and the time taken for the system to adjust to changes in demand, and (iii) the tradeoff between the rate of virtual topology updates and convergence. These insights give us the confidence to claim that a much larger system on the scale of hundreds of thousands of highly heterogeneous nodes would perform as well as our current implementation.

Networking researchers complain that the current Internet is ossified, i.e. that it can hardly be changed. We believe that one of the fundamental reasons for that is the lack of appropriate incentives for providers to invest in new technology, especially in the absence of a compelling new architecture and a killer application that would benefit from an alternative architecture. There is a chicken-and-egg problem: In order to come up with exciting new applications, there needs to be an infrastructure supporting them. Researchers have proposed to build network testbeds (e.g. GENI/FIRE) to test new network architectures and protocols at larger scale. However, these testbeds appear to have little attraction for users, in particular for commercially oriented application developers. OpenFlow is an alternative approach enabling experimental protocols in production networks. However, one of its limitations is lack of addressing provider incentives. In this position paper, we therefore sketch the characteristics that we think a new Internet platform should have in order to be compelling. We argue for a platform that offers rich programmability at low performance cost and that separates traffic to enhance security and limit interference among applications. Moreover, the platform should be open and accessible to a wide community of users and have a high usability in terms of being easily programmable by application developers. Finally, we believe the new platform should provide support for running sophisticated applications across multiple provider domains.

Recently, Etkin, Tse, and Wang found the capacity region of the two-user Gaussian interference channel to within 1 bit/s/Hz. A natural goal is to apply this approach to the Gaussian interference channel with an arbitrary number of users. We make progress towards this goal by finding the capacity region of the many-to-one and one-to-many Gaussian interference channels to within a constant number of bits. The result makes use of a deterministic model to provide insight into the Gaussian channel. The deterministic model makes explicit the dimension of signal level. A central theme emerges: the use of lattice codes for alignment of interfering signals on the signal level.

We study a spectrum sharing problem in an unlicensed band where multiple systems coexist and interfere with each other. Due to asymmetries and selfish system behavior, unfair and inefficient situations may arise. We investigate whether efficiency and fairness can be obtained with self-enforcing spectrum sharing rules. These rules have the advantage of not requiring a central authority that verifies compliance to the protocol. Any self-enforcing protocol must correspond to an equilibrium of a game. We first analyze the possible outcomes of a one shot game, and observe that in many cases an inefficient solution results. However, systems often coexist for long periods and a repeated game is more appropriate to model their interaction. In this repeated game the possibility of building reputations and applying punishments allows for a larger set of self-enforcing outcomes. When this set includes the optimal operating point, efficient, fair, and incentive compatible spectrum sharing becomes possible. We present examples that illustrate that in many cases the performance loss due to selfish behavior is small. We also prove that our results are tight and quantify the best achievable performance in a non-cooperative scenario

ATM networks are moving to a state where large production networks are deployed and require a universal, open and efficient ATM network control platform (NCP). The emerging PNNI (Private Network to Network Interface) standard introduces an internetworking architecture which can also be used as an intranetwork interface. However, PNNI fails in the latter due to performance limitations, limited functionality and the lack of open interfaces for functional extensions. OPENET is an open high-performance NCP based on performance and functional enhancements to PNNI. It addresses the issues of scalability, high performance and functionality. OPENET focuses on intranetworking and is fully compatible with PNNI in the internetwork environment. The major novelties of the OPENET architecture compared to PNNI is its focus on network control performance. A particular emphasis is given to the increase of the overall rate of connection handling, to the reduction of the call establishment latency and to the efficient utilization of the network resources. These performance enhancements are achieved by the use of a native ATM distribution tree for utilization updates, lightweight signalling and extensive use of caching and pre-calculation of routes. OPENET also extends PNNI functionality. It utilizes a new signalling paradigm that better supports fast reservation and multicast services, a control communication infrastructure which enables the development of augmented services such as directory, hand-off, billing, security etc. OPENET was implemented by the High-Speed Networking group at Sun Labs and is under operational tests.

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.

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.>

Worst-case bounds on delay and backlog are derived for leaky bucket constrained sessions in arbitrary topology networks of generalized processor sharing (GPS) servers. The inherent flexibility of the service discipline is exploited to analyze broad classes of networks. When only a subset of the sessions are leaky bucket constrained, we give succinct per-session bounds that are independent of the behavior of the other sessions and also of the network topology. However, these bounds are only shown to hold for each session that is guaranteed a backlog clearing rate that exceeds the token arrival rate of its leaky bucket. A much broader class of networks, called consistent relative session treatment (CRST) networks is analyzed for the case in which all of the sessions are leaky bucket constrained. First, an algorithm is presented that characterizes the internal traffic in terms of average rate and burstiness, and it is shown that all CRST networks are stable. Next, a method is presented that yields bounds on session delay and backlog given this internal traffic characterization. The links of a route are treated collectively, yielding tighter bounds than those that result from adding the worst-case delays (backlogs) at each of the links in the route. The bounds on delay and backlog for each session are efficiently computed from a universal service curve, and it is shown that these bounds are achieved by "staggered" greedy regimes when an independent sessions relaxation holds. Propagation delay is also incorporated into the model. Finally, the analysis of arbitrary topology GPS networks is related to Packet GPS networks (PGPS). The PGPS scheme was first proposed by Demers, Shenker and Keshav (1991) under the name of weighted fair queueing. For small packet sizes, the behavior of the two schemes is seen to be virtually identical, and the effectiveness of PGPS in guaranteeing worst-case session delay is demonstrated under certain assignments.>

Worst-case bounds on delay and backlog are derived for leaky bucket constrained sessions in arbitrary topology networks of generalized processor sharing servers. When only a subset of the sessions are leaky bucket constrained succinct per-session bounds that are independent of the behavior of the other sessions and also of the network topology are given. However, these bounds are only shown to hold for each session that is guaranteed a backlog clearing rate that exceeds the token arrival rate of its leaky bucket. When all of the sessions are leaky bucket constrained, a much larger class of networks called consistent relative session treatment networks is analyzed. The session i route is treated as a whole, yielding tighter bounds than those that result from adding the worst-case delays (backlogs) at each of the servers in the route. The bounds on delay and backlog for each session are computed and shown to be achieved by staggered regimes when an independent sessions relaxation holds. Propagation delay is also incorporated into the model.>

The problem of allocating network resources to the users of an integrated services network is investigated in the context of rate-based flow control. The network is assumed to be a virtual circuit, connection-based packet network. It is shown that the use of generalized processor sharing (GPS), when combined with leaky bucket admission control, allows the network to make a wide range of worst-case performance guarantees on throughput and delay. The scheme is flexible in that different users may be given widely different performance guarantees and is efficient in that each of the servers is work conserving. The authors present a practical packet-by-packet service discipline, PGPS that closely approximates GPS. This allows them to relate results for GPS to the packet-by-packet scheme in a precise manner. The performance of a single-server GPS system is analyzed exactly from the standpoint of worst-case packet delay and burstiness when the sources are constrained by leaky buckets. The worst-case session backlogs are also determined.>

The problem of allocating network resources to the users of an integrated services network is investigated in the context of rate based flow control. The authors propose the use of a packet service discipline at the nodes of the network that is based on a multiplex scheme called generalized processor sharing (GPS). This service discipline is combined with leaky bucket rate admission control to provide flexible, efficient and fair use of the links. A single server GPS system is analyzed exactly, and tight bounds on worst case packet delay, output burstiness and backlog are derived for each session, when the sources are constrained by leaky buckets. The analysis yields a simple resource assignment scheme that allows the server to make worst case delay and rate guarantees to every session in the system. Extensions of this work to arbitrary topology networks are also discussed.>