Multi-agent LLM systems routinely generate multiple candidate responses that are aggregated by an LLM judge.To reduce the dominant prefill cost in such pipelines, recent work advocates KV cache reuse across partially shared contexts and reports substantial speedups for generation agents.In this work, we show that these efficiency gains do not transfer uniformly to judge-centric inference.Across GSM8K, MMLU, and HumanEval, we find that reuse strategies that are effective for execution agents can severely perturb judge behavior: end-task accuracy may appear stable, yet the judge's selection becomes highly inconsistent with dense prefill.We quantify this risk using Judge Consistency Rate (JCR) and provide diagnostics showing that reuse systematically weakens cross-candidate attention, especially for later candidate blocks.Our ablation further demonstrates that explicit crosscandidate interaction is crucial for preserving dense-prefill decisions.Overall, our results identify a previously overlooked failure mode of KV cache reuse and highlight judge-centric inference as a distinct regime that demands dedicated, risk-aware system design.1

An essential goal of the static random access memory (SRAM) array termination design is to both terminate as well as maintain a homogeneous environment for the active edge cells in the array. Local layout effects (LLEs) in the array termination design can exert influence on the active array SRAM devices in close proximity to the termination region, which can lead to undesirable inhomogenuities in the array. The impact of LLEs, originating from the array termination design, on SRAM read performance and Vminfail count, are examined using a 14nm FinFET technology. Large-scale SRAM read performance statistics are analyzed to identify elevated read currents and low-voltage fail counts associated with the array termination. The root cause and modulating factors are explored, and potential solution paths are discussed.

Call detail records (CDRs) collected at telecommunication networks have been well studied to reveal human behaviors such as voice service usage, contact regularities and mobility patterns. With the advances in big data technology, carriers can now collect, store and analyze more data, such as records of user mobile web activities, at scales much larger than CDRs. In this paper, we study 14 features extracted from both call and web records through statistical modeling. We also analyze the impact of locations on voice usage, and examine the correlations between the usage of voice and web services. The results can be used in user profiling, service usage prediction, and network dimensioning.

Determining the best fit statistical distribution with parameters for a dataset is a common task in many datadriven applications. Existing approaches to the problem usually combine parameter estimation and goodness-of-fit (GOF) tests or statistics, fitting the dataset to each of the candidate distributions and choosing the best one based on the GOF. With big data technologies collecting and storing more and more datasets, the efficiency of the existing solutions becomes a problem. In this paper, we decompose the model fitting task into two independent parts: selecting the distribution and estimating the parameters. We then treat the distribution selection as a classification problem. A classifying engine is proposed to efficiently determine the distribution for a dataset among a set of candidate distributions, after which parameter estimation can be performed. Our approach eliminates the need of estimating parameters for each candidate distribution and thus reduces complexity significantly.

Next generation analytics will be all about graphs, though performance has been a fundamental challenge for large scale graph processing. In this paper, we present an industrial graph processing engine for exploring various large scale linked data, which exhibits superior performance due to the several innovations. This engine organizes a graph as a set of edge-sets, compatible with the traditional edge-centric sharding for graphs, but becomes more amenable for large scale processing. Each time only a portion of the sets are needed for computation and the data access patterns can be highly predictable for prefetch for many graph computing algorithms. Due to the sparsity of large scale graph structure, this engine differentiates logical edge-sets from the edge-sets physically stored on the disk, where multiple logical edge-sets can be organized into a same physical edge-set to increase the data locality. Besides, in contrast to existing solution, the data structures utilized for the physical edge-sets can vary from one to another. Such heterogeneous edge-set representation explores the best graph processing performance according to local data access patterns. We conduct experiments on a representative set of property graphs on multiple platforms, where the proposed system outperform the baseline systems consistently.

The concepts of Big Data have became intertwined with those of the Internet of Things, creating mental pictures of a fully connected, all-encompassing, cyber-physical world, where each and every object will contribute with information to a "fully aware" society. Academic works are presenting this as the natural evolution for our current technologies. The panel looks at these promises from the hard perspective of reality: what is being done, how much it cost, what needs to be developed, and what can be expected in the near and mid-term.

As the smartphone/cellphone market has exploded, the war on which smartphone platform will dominate has become fiercer than ever. In that vain, the goal of this paper is to answer two fundamental questions: What are the adoption trends for smartphones? And how can we estimate the demand for new smartphones? We answer these two questions by collecting a dataset of 3 million subscribers from a nationwide telecom operator. A key aspect of our work is that we have demographic information per user, such as income level, and age, which we correlate with phone usage patterns. Interestingly, we find that in all demographic groups, Android is leading platform top in all age groups and income levels. A key question is whether the “social influence” affects the choice of phone, which we find more pronounced in business plans. Finally, we develop a predictor to infer the phone a user will switch to considering: (a) the type of previous phone, (b) the social influence, and (c) the demographics of the user. Compared with the reference method, our predictor is effective in: (a) reducing the prediction error in number of phones by 1/3, and (b) in the case of minimizing phone costs, the monetary cost by half. Apart from its interest in observations, our work could help telecom operator forecast their inventory more accurately by pointing to the right properties to consider.

Wireless carriers have various churn models that are mainly based on profiling the customers and assigning churn probabilities to them. Profiling is usually limited to their individual data, such as their subscription history, demographics, usage, etc. However, our analysis of a major wireless carrier data shows that such churn prediction methods do not fully model wireless subscriber churn, and that the subscribers can be influenced by other subscribers' churn in their social network. We propose a novel method to identify `churn influencers', whose influence makes their social contacts churn subsequently. To build our model, we scored the subscribers' influence level in a way that can take current churn models into account. We further used large scale call records to identify social network and communication features that abstract the strong influencers. Using real world churn data, we trained classification tools to classify high influencers with up to ninety nine percent precision.

As the usage of wireless devices rapidly increases, mobile phones have become an important aid for people to maintain social relationships. Analysis on cellular networks through mobile communication records, especially when such records contain temporal and spatial information, can potentially unveil fundamental laws that govern the dynamics of social networks. In this paper, we use call detail records (CDRs) collected from a nation-wide cellular network in North America for more than one month. We analyze contact patterns over one million pairs of people using real call data as looking into pairwise call records for each pair. First, we investigate the impacts of social relationships on contacts and discover that family members show different characteristics from non-family pairs in terms of contact regularity or duration. Next, we characterize inter-contact time and contact duration using finite mixture models of Gaussian, Lognormal, and Gamma distributions for each pair. Our mixture models for pairwise communication patterns of cellular users capture and demonstrate burstiness, periodicity, and inhomogeneity of human communication.

How do people use phone calls and text messages for their communication needs? Most studies so far study each mode of communication in isolation. Here, we study the interplay of multi-modal communications. We analyze more than a billion call and text records from a Chinese city and San Francisco Area between 2007 and 2011. First, we provide some definitions towards a framework for analyzing multi-modal communications. Then,we study the relationship of the two communication modes and quantify several aspects of correlation and inference. For a communicating pair, we find that the existence of texting during the weekend is the strongest indicator that the pair will communicate at other times with texts or calls. We compare the behavior between China and the U.S. and we find several similarities and differences. For example, we find evidence of an after-lunch siesta among Chinese users. Finally, we study the evolution of the two modes over time. We find that texting has taken over in sheer number of ”events” by flipping the number of calls over that of texts from 2: 1 in 2007 to 1:2 in 2011.

Homophily refers to the phenomenon where people who are socially-connected share many characteristics including demographic and behavioral properties. The goal of this paper is to see whether homophily exists in call networks and if so, to what degree we can infer a cellphone user's demographic properties by knowing the demographic information of the people that s/he talks to. We focus on three types of demographic information: a) home location, b) age group, and c) income level. The novelty is two-folds. First, we use both communication metrics and structural properties of call graphs to identify those “important” friends for each user with whom (s)he is most likely to be in homophily. Second, we assess the importance of different time slices such as weekdays, or nights and weekends for capturing different user relationships. We conduct our study on a real data trace with 20M subscribers during one month from a nationwide cellular carrier. Our first contribution is that we quantify the extent of homophily on the call graph and identify the correlations between homophily and communication and structural features. As a second contribution, we develop effective methods to infer demographic information for a cellular user using linear regression to select the most homophily-like friend of her/him. We find that we can predict home location within 20km radius with 80% accuracy, and age group and income level with 78% and 72% accuracy, respectively.

Malicious activities involving Android applications are rising rapidly. As prior work on cyber-crimes suggests, we need to understand the economic incentives of the criminals to design the most effective defenses. In this paper, we investigate application plagiarism on Android markets at a large scale. We take the first step to characterize plagiarized applications and estimate their impact on the original application developers. We first crawled 265,359 free applications from 17 Android markets around the world and ran a tool to identify similar applications ("clones"). Based on the data, we examined properties of the cloned applications, including their distribution across different markets, application categories, and ad libraries. Next, we examined how cloned applications affect the original developers. We captured HTTP advertising traffic generated by mobile applications at a tier-1 US cellular carrier for 12 days. To associate each Android application with its advertising traffic, we extracted a unique advertising identifier (called the client ID) from both the applications and the network traces. We estimate a lower bound on the advertising revenue that cloned applications siphon from the original developers, and the user base that cloned applications divert from the original applications. To the best of our knowledge, this is the first large scale study on the characteristics of cloned mobile applications and their impact on the original developers.

Malicious activities involving Android applications are rising rapidly. As prior work on cyber-crimes suggests, we need to understand the economic incentives of the criminals to design the most effective defenses. In this paper, we investigate application plagiarism on Android markets at a large scale. We take the first step to characterize plagiarized applications and estimate their impact on the original application developers. We first crawled 265,359 free applications from 17 Android markets around the world and ran a tool to identify similar applications ("clones"). Based on the data, we examined properties of the cloned applications, including their distribution across different markets, application categories, and ad libraries. Next, we examined how cloned applications affect the original developers. We captured HTTP advertising traffic generated by mobile applications at a tier-1 US cellular carrier for 12 days. To associate each Android application with its advertising traffic, we extracted a unique advertising identifier (called the client ID) from both the applications and the network traces. We estimate a lower bound on the advertising revenue that cloned applications siphon from the original developers, and the user base that cloned applications divert from the original applications. To the best of our knowledge, this is the first large scale study on the characteristics of cloned mobile applications and their impact on the original developers.

Many of today's applications such as cellular network management, prediction and control of the spread of biological and mobile viruses, etc., depend on the modeling and prediction of human locations. However, having widespread wireless localization technology, such as pervasive cell-tower/GPS location estimation available for only the last few years, many factors that impact human mobility patterns remain underresearched. Furthermore, many industries including telecom providers are still in need of low-cost and simple location/place prediction methods that can be implemented on a large scale. In this paper, we focus on “temporal factors” and demonstrate that they significantly impact randomness, size, and probability distribution of people's movements. We also use this information to make simple and inexpensive prediction models for subscribers' visited places. We monitored individuals for a month and divided days and hours into segments for each user to obtain probability distribution of their places for each segment of time intervals and observed major improvement in future “time-based” predictions of their location compared to when temporal factors were not considered. In addition to quantifying the improvement in place prediction, we show that significant improvements can actually be achieved through an intuitive division of time intervals with no added computational complexity.

Fine-grained dynamic population estimation is in an increasingly high demand as it has numerous applications in wireless network engineering, urban planning, location-based services and mobile applications, and advertisement. In this paper, we introduce a framework that dynamically estimates the wireless subscriber population of an arbitrary fine-grained area based on the current cellular phone usage. This framework takes advantage of strong regularities, low variance, and low information entropy in human mobility and phone usage patterns; thus simplifying the estimation for wireless carriers and other big entities while maintaining a high accuracy. We implemented our `regularity-based' framework using empirical data. Comparison with experimentally collected data shows a significant improvement in the accuracy of population estimation compared to population count based on cellular phone usage.

Location-based services enabled by broadband wireless access play an increasingly important role in people's daily navigation and coordination. Location-based applications frequently report user location to Internet servers, and location accuracy is essential to the utility of these services. However, regular and accurate location updates impact efficient usage of network resources and also users' privacy, which is not directly observed by the users. In this paper, we conducted a large scale measurement study to understand location accuracy and communication frequency of such applications. We found that while most location reports are accurate enough, some applications run in the background, reporting user locations with a higher accuracy and frequency than needed for the user's purpose. For example, while from the user perspective, hourly zip code updates are enough location information to obtain local weather conditions, some weather forecast applications report user location every five minutes or less and at GPS-level accuracy. We found that location reports from many phone applications are accurate enough and frequent enough to enable the inference of users home and work addresses, and potentially their identity, exacerbating user privacy concerns.

Recent technological advances have resulted in a dramatic change in the market shares of cellular mobile devices. However, little is known about the impact of these changes on the landscape of cellular network traffic. Using anonymized traces from one million cellular subscribers, we conduct a comparative study of the usage characteristics of three different types of mobile devices: feature phones, air cards, and smart phones. Our study covers three aspects: traffic volume in terms of data, voice, and short messages and corresponding temporal fluctuations, applications breakdown in data access, and the presence of malicious traffic. Our study reveals some similarities as well as distinct differences among the three device types. These insights into the modern cellular network traffic could influence how cellular carriers manage and provision their networks.

We examine a very large-scale data set of more than 30 billion call records made by 25 million cell phone users across all 50 states of the US and attempt to determine to what extent anonymized location data can reveal private user information. Our approach is to infer, from the call records, the "top N" locations for each user and correlate this information with publicly-available side information such as census data. For example, the measured "top 2" locations likely correspond to home and work locations, the "top 3" to home, work, and shopping/school/commute path locations. We consider the cases where those "top N" locations are measured with different levels of granularity, ranging from a cell sector to whole cell, zip code, city, county and state. We then compute the anonymity set, namely the number of users uniquely identified by a given set of "top N" locations at different granularity levels. We find that the "top 1" location does not typically yield small anonymity sets. However, the top 2 and top 3 locations do, certainly at the sector or cell-level granularity. We consider a variety of different factors that might impact the size of the anonymity set, for example the distance between the "top N" locations or the geographic environment (rural vs urban). We also examine to what extent specific side information, in particular the size of the user's social network, decrease the anonymity set and therefore increase risks to privacy. Our study shows that sharing anonymized location data will likely lead to privacy risks and that, at a minimum, the data needs to be coarse in either the time domain (meaning the data is collected over short periods of time, in which case inferring the top N locations reliably is difficult) or the space domain (meaning the data granularity is strictly higher than the cell level). In both cases, the utility of the anonymized location data will be decreased, potentially by a significant amount.

In this paper, we present a general technique based on Bayesian inference to locate mobiles in cellular networks. We study the problem of localizing users in a cellular network for calls with information regarding only one base station and hence triangulation or trilateration cannot be performed. In our call data records, this happens more than 50% of time. We show how to localize mobiles based on our knowledge of the network layout and how to incorporate additional information such as round-trip-time and signal to noise and interference ratio (SINR) measurements. We study important parameters used in this Bayesian method through mining call data records and matching GPS records and obtain their distribution or typical values. We validate our localization technique in a commercial network with a few thousand emergency calls. The results show that the Bayesian method can reduce the localization error by 20% compared to a blind approach and the accuracy of localization can be further improved by refining the a priori user distribution in the Bayesian technique.

Modern cellular networks commonly deploy rapid channel rate adaptation to vary the wireless capacity in response to channel conditions while maintaining a fixed target error rate (typically 1%). Although desirable in terms of throughput for inelastic applications that do not adapt to network conditions, a low fixed target error rate incurs the expense of significant power consumption, especially at high transmission rates. In this work, we show that elastic traffic, in particular TCP, benefits greatly from the perspective of power efficiency when we also incorporate target error rate adaptation. More specifically, TCP behavior, although sensitive to packet errors, is not uniformly so. When TCP has a small window, it requires extremely low packet error rates. However, for large windows, especially with a buffer, TCP can tolerate larger loss rates. The contribution of this work is in conducting a detailed and realistic investigation into how beneficial target error rate adaptation is for TCP in terms of reducing power and impact on throughput. Our work differs from past contributions in that we explicitly take into account the impact of the buffer and a variable channel. We devise simple local power-adaptation policies based on TCP behavior and study them with the help of a numerical model. Finally, we present a detailed investigation of our policies using actual modulation schemes and real channel traces collected on a commercial 1xEV-DO network. The results show that compared to the existing scheme, our policies save typically about 20% to 30% power with marginal or no reduction in throughput.

This paper explores the use of rate adaptation in cellular networks to maximize throughput of long-lived TCP sessions. We focus on the problem of maximizing the throughput of TCP connections and propose a joint optimization of MAC and physical layer parameters with respect to TCP sending rate. In particular, we propose a simple TCP-aware channel scheduler that adapts the wireless channel rate to changes in the TCP sending rate and explore its performance for both single and multiple concurrent sessions. In the case of a single TCP session, we develop a fluid model of its steady-state behavior in such a system that adapts between two channel rates. Our results indicate that a two-rate scheme improves TCP throughput by 15% to 20% over a system that does not exploit rate adaptation and that little additional benefit accrues from the addition of a third channel rate. Finally, we extend the framework to scenarios where bandwidth is shared by multiple TCP sessions. We propose two channel allocation algorithms and explore their performance through simulation. Our results indicate that TCP throughput is relatively insensitive to either channel allocation algorithm and adaptive rate variation is the dominant factor in performance.

Proportional Fair (PF) is a frequently used channel- aware scheduling algorithm in 3G wireless networks. However, recent work by us and others has shown that, in practice, PF suffers from significant robustness issues that can unnecessarily starve "well-behaved" users. In this paper, we analyze these issues with the goal of developing an alternative scheduling algorithm more robust than PF. We start by identifying two scenarios in which PF can cause starvation. We analyze both scenarios and develop mechanisms that prevent such starvation. Then, we combine these mechanisms to propose our Parallel Adaptive Quantile-based (PAQ) scheduling algorithm. We use simulation experiments with synthetic and measurement-based traces of wireless channel conditions to show that PAQ is not only robust but also achieves comparable or better throughput and fairness than PF.

We present an experimental characterization of the physical and MAC layers in CDMA 1xEV-DO and their impact on transport layer performance. The 1xEV-DO network is currently the fastest mobile broadband cellular network, offering data rates of up to 3.1 Mbps for both stationary and mobile users. These rates are achieved by using novel capacity enhancement techniques at the lower layers. Specifically, 1xEV-DO incorporates rapid channel rate adaptation in response to signal conditions, and opportunistic scheduling to exploit channel fluctuations. Although shown to perform well in isolation, there is no comprehensive literature that examines the impact of these features on transport layer and application performance in real networks.

Locating mobile users and devices efficiently is a critical operation in cellular networks. This is done using a combination of location update(by the mobile) and paging (by the network). The paging scheme determines how and where to search for a mobile user given the latestlocation update information from that user. In this paper, we considerhow to increase the efficiency of the paging scheme.

Sampling techniques are widely used for traffic measurements at high link speed to conserve router resources. Traditionally, sampled traffic data is used for network management tasks such as traffic matrix estimations, but recently it has also been used in numerous anomaly detection algorithms, as security analysis becomes increasingly critical for network providers. While the impact of sampling on traffic engineering metrics such as flow size and mean rate is well studied, its impact on anomaly detection remains an open question. This paper presents a comprehensive study on whether existing sampling techniques distort traffic features critical for effective anomaly detection. We sampled packet traces captured from a Tier-1 IP-backbone using four popular methods: random packet sampling, random flow sampling, smart sampling, and sample-and-hold. The sampled data is then used as input to detect two common classes of anomalies: volume anomalies and port scans. Since it is infeasible to enumerate all existing solutions, we study three representative algorithms: a wavelet-based volume anomaly detection and two portscan detection algorithms based on hypotheses testing. Our results show that all the four sampling methods introduce fundamental bias that degrades the performance of the three detection schemes, however the degradation curves are very different. We also identify the traffic features critical for anomaly detection and analyze how they are affected by sampling. Our work demonstrates the need for better measurement techniques, since anomaly detection operates on a drastically different information region, which is often overlooked by existing traffic accounting methods that target heavy-hitters.

TCP is the dominant transport protocol over both wired and wireless links. It is however, well known that TCP is not suitable for wireless networks and several solutions have been proposed to rectify this shortcoming. In this work, we explore cross-layer optimization of the rate adaptation feature of cellular networks to optimize throughput of a single long-lived TCP session. Modern cellular networks rate RF technology that allows them to dynamically vary the wireless channel rate in response to user demand and channel conditions. However, the set of data rates as well as the scheduler's rate adaptation policy are typically chosen to optimize throughput for inelastic applications. In order to optimize such a system for TCP, we propose a two state TCP-aware scheduler that switches between two chanrates as a function of the TCP sending rate. We develop a fluid model of the steady-state behavior of a TCP session in such a system and derive analytical expressions for TCP throughput that explicitly account for rate variability as well as the dependency between the scheduler and TCP. Using the model we choose RF layer parameters that, in conjunction with the TCP-aware scheduler, improve term throughput of a single TCP flow by 15.25%. We also compare our analytical results against those obtained from ns-2 simulations and confirm that our model indeed closely approximates TCP behavior in such an environment.

Packet sampling is commonly deployed in high-speed backbone routers to minimize resources used for network monitoring. It is known that packet sampling distorts traffic statistics and its impact has been extensively studied for traffic engineering metrics such as flow size and mean rate. However, it is unclear how packet sampling impacts anomaly detection, which has become increasingly critical to network providers. This paper is the first attempt to address this question by focusing on one common class of nonvolume-based anomalies, portscans , which are associated with worm/virus propagation. Existing portscan detection algorithms fall into two general approaches: target-specific and traffic profiling. We evaluated representative algorithms for each class, namely: 1) TRWSYN that performs stateful traffic analysis; 2) TAPS that tracks connection pattern of scanners; and 3) entropy-based traffic profiling. We applied these algorithms to detect portscans in both the original and sampled packet traces from a Tier-1 provider's backbone network. Our results demonstrate that sampling introduces fundamental bias that degrades the effectiveness of these detection algorithms and dramatically increases false positives. Through both experiments and analysis, we identify the traffic features critical for anomaly detection that are affected by sampling. Finally, using insight gained from this study, we show how portscan algorithms can be enhanced to be more robust to sampling.

This paper investigates survivable lightpath provisioning and fast protection switching for generic mesh-based optical networks employing wavelength-division multiplexing (WDM). We propose subpath protection, which is a generalization of shared-path protection. The main ideas of subpath protection are: 1) to partition a large optical network into smaller domains and 2) to apply shared-path protection to the optical network such that an intradomain lightpath does not use resources of other domains and the primary/backup paths of an interdomain lightpath exit a domain (and enter another domain) through a common domain-border node. We mathematically formulate the routing and wavelength-assignment (RWA) problem under subpath protection for a given set of lightpath requests, prove that the problem is NP-complete, and develop a heuristic to find efficient solutions. Comparisons between subpath protection and shared-path protection on a nationwide network with dozens of wavelengths per fiber show that, for a modest sacrifice in resource utilization, subpath protection achieves improved survivability, much higher scalability, and significantly reduced fault-recovery time.

In this paper, we present a connection-provisioning framework to satisfy customers' availability requirements using appropriate protection schemes. The framework contains two pats: (a) WDM mesh network service availability analysis; and (b) a connection-provisioning approach using the analysis. We present the availability analysis for connections with different protection schemes (i.e., unprotected, dedicated, or shared protected) and propose an integer linear program (LIP) based provisioning approach for static traffic. We verify the theoretical availability analysis through simulations, and demonstrate the effectiveness s of our provisioning approach using numerical examples.

This paper investigates the problem of dynamic shared-path-protected lightpath provisioning in optical mesh networks employing wavelength-division multiplexing (WDM). We prove that the problem of finding an eligible pair of working and backup paths for a new lightpath request requiring shared-path protection under the current network state is NP-complete. We develop a heuristic, called CAFES, to compute a feasible solution and an algorithm, called OPT, to optimize resource consumption for a given solution. The merits of our approaches are that they capture the essence of shared-path protection and approach to optimal solutions without enumerating paths. We evaluate the effectiveness of our heuristics and the results are found to be promising.

We investigate the survivable traffic-grooming problem for optical mesh networks employing wavelength-division multiplexing (WDM). In the dynamic provisioning context, a typical connection request may require bandwidth less than that of a wavelength channel, and it may also require protection from network failures, typically fiber cuts. Based on a generic grooming-node architecture, we propose three approaches for grooming a connection request with shared protection: protection-at-lightpath level (PAL); mixed protection-at-connection level (MPAC); separate protection-at-connection level (SPAC). In shared-mesh protection, backup paths can share resources as long as their corresponding working paths are unlikely to fail simultaneously. These three schemes explore different ways of backup sharing, and they trade-off between wavelengths and grooming ports. Since the existing version of the problem for provisioning one connection request with shared protection is NP-complete, we propose effective heuristics. Under today's typical connection-bandwidth distribution where lower bandwidth connections outnumber higher bandwidth connections, we find the following: 1) it is beneficial to groom working paths and backup paths separately, as in PAL and SPAC; 2) separately protecting each individual connection, i.e., SPAC, yields the best performance when the number of grooming ports is sufficient; 3) protecting each specific lightpath, i.e., PAL, achieves the best performance when the number of grooming ports is moderate or small.

This paper investigates the characteristics and performance of different optical grooming switches, i.e., optical cross-connects (OXCs) capable of traffic grooming, under a dynamic traffic environment. We present four optical grooming-OXC architectures, namely, single-hop grooming OXC, multihop partial-grooming OXC, multihop full-grooming OXC, and light-tree-based source-node grooming OXC. After exploring their grooming capabilities, we propose three grooming schemes and two corresponding algorithms, grooming using auxiliary graph and grooming using light-tree. Through the algorithms, we evaluate the performance of different optical grooming OXCs in a dynamic traffic environment under different connection bandwidth-granularity distributions. Our investigation uncovers the following results: 1) the multihop full-grooming OXC can achieve the best network performance, but it may encounter cost and scalability constraints; 2) by using significantly less low-granularity electronic processing and intelligent traffic-grooming algorithms, the multihop partial-grooming OXC shows reasonable network performance and, hence, can be viewed as a cost-effective alternative when a network node does not require full-grooming capability; 3) the single-hop grooming OXC may cause a large amount of capacity waste and lead to poor network performance; and 4) through its multicast capability, a light-tree-based source-node grooming OXC can significantly out-perform the performance of a single-hop grooming OXC in terms of network throughput and network resource efficiency. From our results, we also observe that the connection bandwidth-granularity distribution has a large impact on network throughput and network resource efficiency and, therefore, should be carefully considered for network design and traffic provisioning.

We investigate the design of a WDM backbone network with optical cross-connects (OXCs) of different switching granularities to reduce the network-wide OXC port cost. We enhance our proposed graph model (Zhu, H. et al., IEEE/ACM Trans. Networking, vol.11, p.285-99, 2003), and the extended graph model can represent different node architectures in which a node may have multiple OXCs with different switching granularities simultaneously. Based on this model, we propose a provisioning algorithm for a single connection and a framework for network design, which can intelligently determine the type of OXCs at each node according to the traffic so that the benefit of different types of OXCs can be utilized. Numerical examples are presented showing that granularity-heterogeneous networks are more cost-effective than granularity-homogeneous networks.

This study investigates the problem of fault management in a wavelength-division multiplexing (WDM)-based optical mesh network in which failures occur due to fiber cuts. In reality, bundles of fibers often get cut at the same time due to construction or destructive natural events, such as earthquakes. Fibers laid down in the same duct have a significant probability to fail at the same time. When path protection is employed, we require the primary path and the backup path to be duct-disjoint, so that the network is survivable under single-duct failures. Moreover, if two primary paths go through any common duct, their backup paths cannot share wavelengths on common links. This study addresses the routing and wavelength-assignment problem in a network with path protection under duct-layer constraints. Off-line algorithms for static traffic is developed to combat single-duct failures. The objective is to minimize total number of wavelengths used on all the links in the network. Both integer linear programs and a heuristic algorithm are presented and their performance is compared through numerical examples.

As the operation of our fiber-optic backbone networks migrates from interconnected SONET rings to arbitrary mesh topology, traffic grooming on wavelength-division multiplexing (WDM) mesh networks becomes an extremely important research problem. To address this problem, we propose a new generic graph model for traffic grooming in heterogeneous WDM mesh networks. The novelty of our model is that, by only manipulating the edges of the auxiliary graph created by our model and the weights of these edges, our model can achieve various objectives using different grooming policies, while taking into account various constraints such as transceivers, wavelengths, wavelength-conversion capabilities, and grooming capabilities. Based on the auxiliary graph, we develop an integrated traffic-grooming algorithm (IGABAG) and an integrated grooming procedure (INGPROC) which jointly solve several traffic-grooming subproblems by simply applying the shortest-path computation method. Different grooming policies can be represented by different weight-assignment functions, and the performance of these grooming policies are compared under both nonblocking scenario and blocking scenario. The IGABAG can be applied to both static and dynamic traffic grooming. In static grooming, the traffic-selection scheme is key to achieving good network performance. We propose several traffic-selection schemes based on this model and we evaluate their performance for different network topologies.

In a WDM optical network, the bandwidth requirement of a customer's connection can vary over a wide range, and many of these connections could have a capacity that is much lower than the capacity of a wavelength channel. Efficiently grooming low-speed connections onto high-capacity wavelength channels can significantly improve the bandwidth utilization and minimize the network cost. Our research shows that it is not necessary to have traffic-grooming capability at every network node. We call a network which has only a few grooming nodes to be a sparse-grooming network. Through proper network design and traffic engineering, it is possible for a sparse-grooming network to achieve similar network performance as a network which has grooming capability at every node. We investigate the problem of designing such a sparse-grooming WDM mesh network. The problem is mathematically formulated and several design schemes are proposed. Illustrative numerical results from the mathematical formulation as well as heuristics show that, by properly choosing the grooming nodes, a network with sparse-grooming capability can achieve good network performance and the network cost can be significantly reduced.

We employ a new, generic graph model for dynamic traffic grooming in WDM mesh networks. The novelty of this model is that, by only manipulating the edges of an auxiliary graph created by the model and the weights of these edges, the model can achieve various objectives using different grooming policies, while taking into account various constraints. Based on the auxiliary graph, we develop a dynamic traffic-grooming algorithm. Different grooming policies can be implemented by different weight functions assigned to the edges in the auxiliary graph. We propose four fixed grooming policies and an adaptive grooming policy (AGP), and our results show that AGP outperforms the fixed grooming policies.