Self-* Overload Control for Distributed Web Systems

Abstract

Unexpected increases in demand and most of all flash crowds are considered the bane of every web application as they may cause intolerable delays or even service unavailability.Proper quality of service policies must guarantee rapid reactivity and responsiveness even in such critical situations. Previous solutions fail to meet common performance requirements when the system has to face sudden and unpredictable surges of traffic. Indeed they often rely on a proper setting of key parameters which requires laborious manual tuning, preventing a fast adaptation of the control policies.We contribute an original Self-* Overload Control (SOC) policy. This allows the system to self-configure a dynamic constraint on the rate of admitted sessions in order to respect service level agreements and maximize the resource utilization at the same time. Our policy does not require any prior information on the incoming traffic or manual configuration of key parameters.We ran extensive simulations under a wide range of operating conditions, showing that SOC rapidly adapts to time varying traffic and self-optimizes the resource utilization. It admits as many new sessions as possible in observance of the agreements, even under intense workload variations. We compared our algorithm to previously proposed approaches highlighting a more stable behavior and a better performance.

Towards Autonomic Service Provisioning Systems

“…The idea of using windows for self-optimizing computing systems is not new, but while SPIRE uses event-based windows, other scientists propose time-based windows [7] or complex algorithms to detect traffic surges [8]. The use of events, instead, provides a simpler way to implement adaptive windows: under heavy traffic conditions the allocation and admission algorithms are executed more often than when the load is light.…”

AWAIT: Efficient Overload Management for Busy Multi-tier Web Services under Bursty Workloads

“…Connection requests are forwarded into two different request queues, and admission control is performed using two metrics: the accept queue length and measurement-based predictions of arrival and service rates from that class. Bartolini et al, in their recent work [6,7], introduce a quite elaborate session admission algorithm, called AACA, that self-configures a dynamic constraint on the rate of incoming new sessions to satisfy guarantees of the Service Level Agreements (SLA). However, the rate limitation for the next iteration interval is based on a relatively straightforward prediction of the session arrival rate from the previous interval measurements.…”

“…As Internet services become indispensable both for businesses and personal productivity, the efficient management of Internet services under periods where the system is overloaded or simply highly variable, is of critical importance. There is a host of solutions to maintain user-perceived performance levels in the form of service-level objectives (SLOs) that focus mainly on admission control and/or techniques for service differentiation that are threshold based [13,6,7,19] but their effectiveness can be compromised if the workload is bursty, i.e., it is characterized by sudden temporal "surges" in the intensity of user arrivals [23] and user demands [22]. While capacity planning of systems under bursty workload conditions has been recently demonstrated as critical for business success [22,23], the problem of efficient admission control and service differentiation under temporal workload bursts remains largely unexplored.…”

Self-∗ through self-learning: Overload control for distributed web systems

“…The design of SOC is inspired by the policies we introduced in our previous articles [7,29]. Here we have added a change detection and a decision rate adaptation mechanism to manage flash crowds, as well as the scaling procedure to handle capacity variations.…”