Utility driven optimization of real time data broadcast schedules

Abstract

Data dissemination in wireless environments is often accomplished by on-demand broadcasting. The time critical nature of the data requests plays an important role in scheduling these broadcasts. Most research in on-demand broadcast scheduling has focused on the timely servicing of requests so as to minimize the number of missed deadlines. However, there exists many environments where the utility of the received data is an equally important criterion as its timeliness. Missing the deadline may reduce the utility of the data but does not necessarily make it zero. In this work, we address the problem of scheduling real time data broadcasts with such soft deadlines. We investigate search based optimization techniques to develop broadcast schedulers that make explicit attempts to maximize the utility of data requests as well as service as many requests as possible within an acceptable time limit. Our analysis shows that heuristic driven methods for such problems can be improved by hybridizing them with local search algorithms. We further investigate the option of employing a dynamic optimization technique to facilitate utility gain, thereby eliminating the requirement of a heuristic in the process. An evolution strategy based stochastic hill-climber is investigated in this context.

Introduction

Wireless computing involves a network of portable computing devices thoroughly embedded in our day-to-day work and personal life. The devices interact with each other and with other computing nodes by exchanging rapid and continuous streams of data. To facilitate almost imperceptible human–computer interaction, data access times in such environments must be maintained within a specified quality-of-service (QoS) level. Challenges in doing so arise from the fact that wireless bandwidth is typically a limited resource, and thus it is not always possible to meet the quality requirements of every device. This constraint not only makes wireless data access a challenging problem, but also identifies “optimal resource allocation” as one of the major research problems in this domain.

A wireless environment encompasses both peer-to-peer and client-server modes of data dissemination. For example, a typical pervasive health care system may involve multiple sensor nodes disseminating data on the monitored vital signs of a patient to a personal digital assistant carried by the attending health care personnel. Data communication follows a peer-to-peer architecture in such a setting. On the other hand, a wireless environment designed to serve queries on flight information in an airport is based on a client-server mode of communication. Flight information usually reside in a database server from where data is disseminated based on the incoming queries. For an environment like an airport, it is appropriate to assume that the database will be queried more frequently for certain types of data. Similar situations can be imagined in a stock trading center, a pervasive traffic management system, or a pervasive supermarket. Such scenarios open up possibilities of adopting a broadcast based architecture to distribute data in a way that multiple queries for the same data item get served by a single broadcast. The focus of this paper is directed towards a combinatorial problem that arises in this approach.

Quality of service (QoS) is an important facet in wireless data access. Consider the following example application – a traffic management system for a big city. The city government implements a traffic pricing policy for all vehicles on all roads based on factors such as the distance traveled, the type of road traveled on, the time of day, vehicle category and customer type (for example, special fee paying, traveling salesman, and paramedic on-call) The system gathers real time traffic data via thousands of sensors, such as traffic cameras, magneto-resistive traffic sensors and radars spread throughout the city. To help drivers optimize their travel costs, the traffic management system provides various routing services to drivers to avoid roadblocks, construction delays, congestion, accidents, etc.¹

A driver requests and gets routing services using smart GPS equipped devices as follows. A device periodically uploads the vehicle’s current location (based on GPS information), intended destination, time willing to spend for traveling to destination, a prioritized list of routing restrictions (for example, waypoints that the driver would like to visit if possible, rest stops, and scenic byways), vehicle category and customer type, and current speed. In response, the server replies with a set of route information. Each route information contains, among other things, information about road segments to travel on this route and traffic patterns on these road segments. The GPS equipped device uses this information and uses other parameters set by the driver to compute a good route to take. Note that since the response to many drivers will no doubt contain overlapping route segments and traffic information, it makes sense broadcasting this data.

The requests arrive at the server with various priority levels and soft deadlines. For example, a higher fee paying customer gets a higher priority than a lesser fee paying customer. A VIP’s convoy may get a still higher priority. Emergency responders get the highest priority. A routing request that comes associated with a set of routing restrictions, for example, a list of waypoints that the driver would like to visit en route, automatically gets associated with a set of soft deadlines based on the speed of the driver. The requests from different drivers may also get re-prioritized at the server so as to meet a broader goal of reducing congestion and enabling smoother traffic flows.

In this example, the different data that the server needs to serve is associated with different utility values. Owing to the dynamic nature of the utility of responses to queries, the time criticality factor cannot be ignored altogether when disseminating data. The server would like to satisfy as many queries in a timely manner as possible. However, some queries may be delayed beyond their specified deadlines (for example, the query from the VIP’s convoy). The users, who hardly realize the broader goals of the traffic management system and various bottlenecks in the information infrastructure, would like to have their requests served at the earliest; however, it is reasonable to assume that delayed data still provide some utility even if received after a specified deadline. For example, a delayed route information may prevent a driver from visiting a particular waypoint en route or may require the driver to use the next gas station. Nonetheless, it may still allow the driver to choose a good route to the destination. An assumption of reduced data utility in the face of missed deadline enables data broadcasts to be tailored in such a way that total utility associated with a broadcast is maximized. This helps maintain a certain QoS level in the underlying infrastructure.

Note that the specific problem we are trying to address is by no means restricted only to the example application outlined above. Similar scenarios are witnessed in environments such as stock exchanges. Here stock brokers on the floor seek and receive periodic market data on wireless hand-held devices and notebooks and analyze them to determine which stocks are rewarding. They also buy and sell stocks using such devices. Popularity of stocks changes throughout the day, and it is important to analyze such trends along multiple dimensions in order to buy and sell stocks. Thus, although queries from brokers are implicitly associated with deadlines, these can be considered soft. The overall goal of the stock exchange still remains to satisfy as many requests as possible in a timely manner. To wrap up the scope of the problem domain we would like to point out that in order to make useful decisions, the stock broker may make a request for a group of data from the stock exchange (for example, stock prices of three different oil companies). A typical constraint on such a request is that all these data items must be received (not necessarily in any particular order) before the stock broker can perform the relevant local analysis of the market trends. Such a request can be considered as a transactional request (or simply a transaction). For scheduling in such cases, additional constraints must be placed for ensuring the validity of data received.

Wireless broadcast mechanisms have been extensively investigated earlier. However, very few of the proposed approaches give attention to the effective utility involved in the timely servicing of a request. Time criticality has been earlier addressed in a number of contexts with the assumption of a hard deadline [2], [3], [4]. Broadcast scheduling in these works mostly focus on the timely servicing of a request to minimize the number of missed deadlines. When the assumption of a soft deadline is appropriate, a broadcast schedule should not only try to serve as many requests as possible, but also make a “best effort” in serving them with higher utility values. Often, heuristics are employed in a dynamic setting to determine these schedules. However, their designs do not involve the QoS criteria explicitly. Heuristic based methods make local decisions w.r.t. a request or a broadcast, and often fail to capture the sought global QoS requirement. Much of this is due to the fact that real time decision making cannot span beyond an acceptable time limit, thereby restricting the usage of “full length” optimization techniques to the domain. It does become imperative to design hybrid strategies that can combine the fast real time performance of heuristics and the better solution qualities obtained from search based optimization.

Our contributions in this paper are summarized in the following points:

• We explore data broadcasting as a combinatorial problem defined with an objective of utility maximization. This utility driven optimization of broadcasts adds more value to a wireless system since the energy cycles used by the devices while retrieving a data item will presumably not be wasteful.

• We provide an extensive analysis of the search space involved in the problem and report the performance of multiple heuristics on a set of statistically generated synthetic requests. We argue that traditional heuristics usually generate solutions in a sub-optimal part of this space with respect to a given global utility measurement.

• We empirically demonstrate that “local search” can be used in real-time to boost the performance of naive heuristics such as EDF, with performance levels often at par with other elaborate heuristics. In addition, we propose an evolution strategy based light-weight stochastic hill-climber that explicitly searches the space of schedules to maximize utility. Results demonstrate that, in certain problem types, this explicit search strategy can generate higher objective values than a traditional heuristic based approach.

• We perform the analysis indicated in the above points in the context of two problem types, namely at the data item level and at the transaction level. In the former type, requests involve a single data item, while multiple unordered data items are involved in a request of the latter type. Besides the local search and evolution strategy based approaches, we present the performance of heuristics such as EDF, HUF, R × W, SIN-α, NPRDS and ASETS* on one or more of these problem types. Transaction level scheduling induces an exponentially large search space and has been observed to demonstrate very different dynamics than a data item level problem.

The rest of the paper is organized as follows. Section 2 summarizes the related work in this domain. The broadcast model and the scheduling problem are discussed in Section 3. The explored solution methods are described in Section 4. Section 5 discusses the scheduling problem at the transaction level. The experimental setup followed by results and discussions are summarized in Sections 6 Experimental setup, 7 Results and discussion, respectively. Finally, Section 8 concludes the paper.