Dipyaman Banerjee

Simulation-based evaluation of dispatching policies in service systems

A service system is an organization of the resources and processes, which interacts with the customer and produces service outcomes. Since a majority of the service systems are labor-intensive, the main re-sources are the service workers. Designing such service systems is nontrivial due to a large number of parameters and variations, but crucial for business decisions such as labor staffing. The most important design point of a service system is how and when service requests are assigned to service workers a.k.a. dispatching policy. This paper presents a framework for evaluation of dispatching policies in service systems. A discrete event simulation model of a service system in the data-center management domain is presented. We evaluate four dispatching policies on five real-life service systems. We observe that the simulation-based approach incorporates intricacies of service systems and allows comparative analysis of dispatching policies leading to more accurate decisions on labor staffing.

A Framework for Distributed Monitoring and Root Cause Analysis for Large IP Networks

As the size of a centrally managed IP network increases, the cost of monitoring network devices and the number of reported events increase super-linearly. This in turn degrades the performance of the event correlation engine that is responsible for suppressing dependent events and escalating root cause events to a network administrator. To solve this scalability problem, we propose a distributed framework that partitions the network into smaller management domains and enables concurrent monitoring and event correlation in those domains. The gain in performance, however, comes with the challenge of correlating cross-domain events which occurs when failure in one domain induces events in other domain(s). In this paper, we investigate such situations and show in the worst case it would be impossible to determine the root cause. We propose a two step approach to solve this problem. First, we define a property called route-closure, which if satisfied by every partition not only minimizes the number of cross-domain events but also eliminates cases wherein root cause analysis may be inconclusive. We also describe a technology-centric partitioning mechanism that constructs partitions satisfying the route-closure property. Next, we propose a distributed architecture to efficiently identify and correlate cross-domain events. We use a commercial network management system to implement our distributed framework and run experiments by injecting synthetic events on large, real network topologies. Our experimental results show that our approach can manage over 200,000 managed entities and handle event bursts of size 15,000 in under five minutes without compromising the efficacy of event correlation.

Accommodating user diversity for in-store shopping behavior recognition

This paper explores the possibility of using mobile sensing data to detect certain in-store shopping intentions or behaviours of shoppers. We propose a person-independent activity recognition technique called CROSDAC, which captures the diversity in the manifestation of such intentions or behaviours in a heterogeneous set of users in a data-driven manner via a 2-stage clustering-cum-classification technique. Using smartphone based sensor data (accelerometer, compass and Wi-Fi) from a directed, but real-life study involving 86 shopping episodes from 30 users in a malls food court, we show that CROSDACs mobile sensing-based approach can offer reasonably high accuracy (77:6% for a 2-class identification problem) and outperforms the traditional community-driven approaches that unquestioningly segment users on the basis of underlying demographic or lifestyle attributes.

KARMA: Improving WiFi-based indoor localization with dynamic causality calibration

WiFi-based indoor localization solutions are actively in commercial use today. WiFi radio maps are typically created in an offline process, and location estimation is performed in realtime (online) using the maps. Both the radio map and the signal strengths provided as input during the online phase are affected by several dynamic factors resident in the environment. We call these “causality factors”. Hence, the online location accuracy is far from the quality achieved in laboratory tests with training data, since the causality factors have varied in between. This impacts the quality expected by heterogeneous applications in real deployments. To address this issue, we investigate a novel online dynamic calibration methodology called KARMA. KARMA utilizes a one-time fingerprint of the space and systematically applies a set of causality calibration functions in real-time to compensate for the change in the factors, at test time. As a result, location providers can now significantly reduce the costly re-learning of the models for different factors, and improve real-time location prediction accuracy. Experimental studies demonstrate that KARMAs strategy, while keeping the fingerprinting task contained, can improve localization quality by a factor of 2x, compared to a typical one-state fingerprinting approach employed in many commercial deployments today. It also compares very favorably with an exhaustive all-state fingerprinting approach.

Improving floor localization accuracy in 3D spaces using barometer

Technologies such as Wifi and BLE have been proven to be effective for indoor localization in two dimensional spaces with sufficiently good accuracy but the same techniques have large margin of errors when it comes to three dimensional spaces. Popular 3D spaces such as malls or airports are marked by distinct structural features - atrium/hollow space and large corridors which reduces spatial variability of WiFi and BLE signal strengths leading to erroneous location prediction. A large fraction of these errors can be attributed to vertical jumps where the predicted location has same horizontal coordinate as the actual location but differs in the vertical coordinate. Smartphones now come equipped with barometer sensor which could be used to solve this problem and create 3D localization solution having better accuracy. Research shows that the barometer can be used to determine relative vertical movement and its direction with nearly 100% accuracy. However exact floor prediction requires repeated calibration of the barometer measurements as pressure values vary significantly across device, time and locations. In this paper we present a method of automatically calibrating smartphone embedded barometers to provide accurate 3D localization. Our method combines a probabilistic learning method with a pressure drift elimination algorithm. We also show that when the floor value is accurately predicted, Wifi localization accuracy improves by 25% for 3D spaces. We validate our techniques in a real shopping mall and provide valuable insights from practical experiences.

Improving Productivity in Design and Development of Information Technology (IT) Service Delivery Simulation Models

The unprecedented scale of Information Technology (IT) service delivery requires careful analysis and optimization of service systems. The simulation is an efficient way to handle the complexity of modeling and optimization of real-world service delivery systems. However, typically developed custom simulation models lack standard architectures and limit the reuse of design and implementation artifacts across multiple models. In this work, following the design science research methodology, based on a formal model of service delivery systems and applying an adapted software product line (SPL) approach, we create a design artifact for building product lines of IT service delivery simulation models, which vastly simplify and reduce the cost of simulation model design and development. We evaluate the design artifact by constructing a product line of simulation models for a set of IBM’s IT service delivery systems. We validate the proposed approach by comparing the simulation results obtained using our models with the results from the corresponding custom simulation models. The case study demonstrates that the proposed approach leads to 5–8 times reductions in the time required to design and develop related simulation models. The potential implications of the application of the proposed approach within an organization are quicker responses to changes in the business environment, more information to assist in managerial decisions, and reduced workload on the process reengineering specialists.

BluePark: tracking parking and un-parking events in indoor garages

Finding a parking spot in a busy indoor parking lot is a daunting task. Retracing a parked vehicle can be equally frustrating. We present BluePark, a collaborative sensing mechanism using smartphone sensors to solve these problems in real-time, without any input from user. We propose a novel technique of combining accelerometer and WiFi data to detect and localize parking and un-parking events in indoor parking lot. We validate our approach at the basement parking of a popular shopping mall. The proposed method outperforms Google Activity Recognition API by 20% in detecting drive state in indoor parking lot. Our experiments show 100% precision and recall for parking and un-parking detection events at low accelerometer sampling rate of 15Hz, irrespective of phone?s position. It has a low detection latency of 20s with probability of 0.9 and good location accuracy of 10m.

Promoting Carpooling with Distributed Schedule Coordination and Incentive Alignment of Contacts

Carpooling is recognized as a viable alternative by city managers to reduce traffic demand and consequent congestion and pollution on city roads. Members of a carpool face the daily challenge of deciding whether to go alone or with some or all members of their carpool, and further, to take their own vehicle or ride with others. These decisions may seem trivial in isolation but if not taken judiciously, over time, may disrupt the carpool. In this paper, we have designed a system which models daily carpool scheduling as a repeated stochastic game played among willing carpoolers which allows for individual choices and promotes carpooling as an equilibrium outcome through equitable payoff distribution using social credits. We empirically show the efficacy of our system through simulation and also discuss how game-theoretic design inherently helps to deploy the system as a mobile application.

Columbus: Configuration Discovery for Clouds

Low-cost, accurate and scalable software configuration discovery is the key to simplifying many cloud management tasks. However, the lack of standardization across software configuration techniques has prevented the development of a fully automated and application independent configuration discovery solution. In this work, we present Columbus, an application-agnostic system to automatically discover environmental configuration parameters or Points of Variability (PoV) in clustered applications with high accuracy. Columbus uses the insight that even though configuration mechanisms and files vary across different software, the PoVs are encoded using a few common patterns. It uses a novel rule framework to annotate file content with PoVs and a Bayesian network to estimate confidence for annotated PoVs. Our experiments confirm that Columbus can accurately discover configuration for a diverse set of enterprise and cloud applications. It has subsequently been integrated in three real-world systems that analyze this information for discovery of distributed application dependencies, enterprise IT migration and virtual application configuration.

How’s My Driving? A Spatio-Semantic Analysis of Driving Behavior with Smartphone Sensors

Road accident is one of the major reasons for loss of human lives, especially in developing nations with poor road infrastructure and a driver needs to constantly negotiate with several adverse conditions to ensure safety. In this paper, we study several such adverse conditions that are relevant to safe driving and propose a novel method for identifying them as well as characterizing driving behavior for such conditions. Experimental results reveal that our proposed methodology is promising and more flexible than prior work in this area. In particular, our prediction results reveal that our methodology is an aggressive one where most of the bad driving behaviors are determined at the cost of a few instances of good behavior being falsely characterized as bad ones.