June 24, 2007 Organized by IBM Research Lab in Haifa, Israel
Abstracts PDF version for printing (280 KB)
Service Engineering: DataBased Science & Teaching in Support of Service Management
Avishai Mandelbaum, Technion
I shall describe examples of complex service operations for which databased simple models have been found useful. The examples cover hospitals, banks, courts and more, but most attention will be given to telephone call centers.
Through the examples, I shall overview an emerging scientific discipline which is now being referred to as "Service Engineering". I focus on empirical findings which motivate or are motivated by (or both) interesting research questions. These findings give rise to modelfeatures that are prerequisite for useful service models, for example customers' (im)patience, timevarying service demand (predictable variability), heterogeneity of customers and servers (skillsbased routing), overdispersion in Poisson arrivals, generallydistributed (as opposed to exponential) service and patiencedurations, and more. Empirical analysis also enables validation of existing models and protocols, either supporting or refuting their relevance and robustness.
Simple Approximation for the TimeVarying M(t)/M/s
Zohar Feldman, IBM Haifa Research Lab
This paper proposes a simple approximation for the transient behavior of queueing systems with timevarying arrival rates. In our approach, the system is analyzed as two separate but closelyintegrating subsystems: the service system and the queue system. We use the ChapmanKolmogorov equations and generating functions to derive two differential equations corresponding to the expected subsystems sizes. These equations are dependent on the instantaneous delay probability and the probability to encounter a nonempty queue, which we approximate in terms of the instantaneous expected size of both subsystems. The immediate results of this approximation are the expected number of busy servers, the expected queue length and the delay probability  at all times. Other performance measures can then be derived or approximated using these values. The main contribution is providing a simple and quick approximation that accommodates the analysis of performance measures in the most general Mt/M/N that may be overloaded for some part of or even the entire timehorizon.
Stable Matching  Assignment of Students to the Dormitories
Uriel Rothblum, Technion
In this paper we report a case study of the assignment of students to dormitories at the Technion  Israel Institute of Technology. Two criteria are applied: first, the privilege of getting oncampus housing is determined by personal socioeconomic parameters, and second, the priority in the assignment of the students who gained oncampus housing privilege to specific dorms is determined by academic seniority and academic excellence. A modification of the classic stable matching model that allows for an "entrance criterion" is developed and analyzed, including the description of an algorithm that produces a stable outcome and a characterization of the output of this algorithm. The algorithm was implemented successfully for the assignment of students to dormitories at the Technion toward the 2004/5 academic year.
Using Stochastic Local Search for Optimizing Vehicle Configuration
Bella Dubrov, IBM Haifa Research Lab
Any given truck is characterized by hundreds of attributes, or variables, such as the type of the engine, wheels, cabin, etc. In addition there are complex rules that determine the valid assignments to these variables. These rules originate from manufacturing, inventory, pricing and so on. The rules are reevaluated and changed periodically.
In the truck configuration problem, we have to find a truck configuration that assigns values to the variables while respecting the needs of a specific customer. We will present our ongoing work on a technology for solving this problem based on a generalpurpose stochastic constraint satisfaction solver that we developed.
Cost Allocation in Supply Chains and Service Systems
Shoshana Annily, TelAviv University
In this talk we discuss the issue of cost allocation in supply chains (SC) and in service management (SM) systems. Both consist of a number of facilities/service providers that may combine forces in order to improve the efficiency of the whole system. A vast amount of literature deals with such optimization models, and good policies have been developed. Once a policy is proposed, a natural question to be posed is how to allocate the cost among the various entities in the system in a fair way. One possible objective is finding a core allocation, meaning an allocation of the total costs such that no subcoalition of entities has an incentive to deviate and act on its own. Core allocation schemes guarantee the stability of the system for the longrun, as they ensure that the formation of the grand coalition pays off to each of the individual entities.
In this talk we present the cost allocation problem for two specific models: In SC we consider the poweroftwo policy for the deterministic, stationary Joint Replenishment Problem (JRP) in the infinite continuous time horizon. In this model each retailer is associated with its demand rate, holding cost rate and (minor) inventory reorder setup cost. In addition, there exists a major setup cost, such that each time a subset of retailers reorder simultaneously, the total setup ordering cost charged is the sum of the major setup and of the individual minor setup costs of the retailers that reorder. The cost of any coalition is the average time total ordering and holding cost of the retailers in the coalition where shortages or backlogging are not allowed. In SM we consider a system that consists of a collection of servers, each characterized by its own capacity and its own stream of customers. A coalition of servers generates a joint entity, which is a single server system with arrival and service rates being the sum of the respective individual rates. The cost of any coalition is the steadystate mean number of customers in the respective system.
Keynote: Robust Solutions of Uncertain Optimization Problems
Aharon BenTal, Technion
We survey the main developments in Robust Optimization (RO), a methodology aimed at solving optimization problems (static and dynamic) affected by uncertainty.
We focus primarily on issues of computational tractability of the robust counterparts emerging from conic optimization problems (linear, conic quadratic and semidefinite programming). We then show how results pertaining to the latter issues can be used to solve difficult chanceconstrained programs under partial stochastic information. Finally, we discuss the synthesis of uncertainty affected discretetime linear control systems by the RO methodology and illustrate the results by treating a supply chain problem.
MultiAgent Systems: The Agent Perspective
Moshe Tennenholtz, Technion
In this talk we consider the challenging and fundamental problem of providing a prescriptive theory of agentcentric decision making in multiagent systems. We aim at offering several complementary approaches to tackling the task of providing the agent with advice about the actions it should choose in multiagent encounters.
Exact and Approximate Solutions of Source Localization Problem
Amir Beck, Technion
We consider the problems of locating a radiating source from range measurements or from rangedifference measurements collected using an array of passive sensors. Both least squares (LS) and maximum likelihood (ML) approaches are studied. Despite the fact that the resulting optimization problems are nonconvex, we provide exact solution procedures for efficiently computing the LS estimates of both problems. Numerical simulations suggest that the exact LS estimate outperforms existing approximations of the LS solution as well as approximations of the ML solution, which are based on a semidefinite relaxation.
Online Ascending Auctions for Gradually Expiring Items
Ron Lavi, Technion
We consider auction mechanisms for the allocation of multiple items in dynamic and unpredictable environments, where buyers frequently enter and leave the markets, and where items may expire. An example for such an environment is the new Internet market places, that became increasingly popular in recent years. Our goal is to design mechanisms that maximize the social welfare, without assuming any knowledge about an underlying distribution. The lack of distributional knowledge, and especially the lack of common priors, seems to be a fundamental property of such dynamic environments, and an essential requirement from any realistic mechanism design in such settings. Under this framework we obtain three results.
Calculating the Business Importance of Entities in a ServiceOriented Enterprise
Segev Wasserkrug, IBM Haifa Research Lab
Component Business Modeling (CBM) serves as a powerful analytical framework for reasoning about the business as a set of business components that collaborate through the provision and consumption of business services. This talk describes a method for calculating the relative importance of the entities that make up a componentized enterprise architecture. The proposed method includes a formal definition of the importance of each entity in the business architecture calculated from the high level business values.
 
