Improving service in the presence of surge traffic: a K-policy voluntary flushing queueing system
- PDF / 417,461 Bytes
- 13 Pages / 439.37 x 666.142 pts Page_size
- 6 Downloads / 189 Views
Improving service in the presence of surge traffic: a K-policy voluntary flushing queueing system Won Seok Yang1 · Nam K. Kim2 · Sungjune Park3
· Chandrasekar Subramaniam3
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Unexpected surge in online services causes poor performance and makes it challenging for the service provider to differentiate and serve genuine users. In this paper, we propose a novel approach using voluntary flushing to mitigate disruptions caused by surge traffic, such as distributed denial of service (DDoS) attacks on e-commerce websites. Voluntary flushing of customers has never been modeled because flushing was perceived negatively for customer service. We show that it may not be the case in today’s online services. We propose three queueing performance measures to evaluate the system and present optimal policies for flushing under various scenarios. We use DDoS attacks as use-case to present numerical analysis and discuss the implications of the flushing policies. Keywords Queueing systems · Voluntary flushing · Distributed denial of service attack
1 Introduction One of the challenges in managing queuing-based services, such as web services, is poor performance due to unexpected surge or spike in traffic in the queue. For example, some news websites may experience a sudden multi-fold increase in traffic after being mentioned in a popular news feed, and the resulting delay in access to the website causes dissatisfaction with users. A further challenge with surge traffic is the uncertainty in determining if the surge is due to an increase in genuine users (i.e., users who are serious about getting the service) or increase in casual users (e.g., visitors who are driven to the service due to some news or other triggers, but do not produce revenue for the service). In some situations, such as a distributed denial of service (DDoS) attack, which mimics a flash crowd, the objective of the surge traffic is to disrupt the service. Most current approaches and models are based on detecting and discriminating the type of traffic upfront so that service to genuine users continues with
B
Sungjune Park [email protected]
1
Department of Business Administration, Hannam University, Daejeon, Republic of Korea
2
Department of Industrial Engineering, Chonnam National University, Gwangju, Republic of Korea
3
Department of Business Information Systems and Operations Management, The University of North Carolina at Charlotte, Charlotte, NC 28223, USA
123
Annals of Operations Research
minimal disruption. In this paper, we propose a voluntary queue-flushing model in which optimal policies are developed to keep service levels acceptable for genuine traffic without explicitly detecting disruptive traffic. We call this model the K -policy system. Compared to the well-known N -policy system (Yadin and Naor 1963; Heyman 1968), where the decision is to start the service based on the system size (i.e., the number of customers in the system), in the K -policy system, the decision is
Data Loading...
