Backlog Estimation and Management for Real-Time Data Services

Abstract

Real-time data services can benefit data-intensive real-time applications, e.g., e-commerce, via timely transaction processing using fresh data, e.g., the current stock prices. To enhance the real-time data service quality, we present several novel techniques for (1) database backlog estimation, (2) fine-grained closedloop admission control based on the backlog model, and (3) hint-based incoming load smoothing. Our backlog estimation and feedback control aim to support the desired service delay bound without degrading the data freshness critical for real-time data services. Workload smoothing, under overload, help the database admit and process more transactions in a timely manner by probabilistically reducing the burstiness of incoming data service requests. In terms of the data service delay and throughput, our feedback-based admission control and probabilistic load smoothing considerably outperform the baselines, which represent the current state of the art, in the experiments performed in a stock trading database testbed.

Integrating Proactive and Reactive Approaches for Robust Real-Time Data Services

“…• As a result, the integrated approach is more adaptive to dynamic workloads compared to feedback-only approaches for real-time data services such as [4], [3], [5], [6], [7]. This is a key feature needed to improve the quality of real-time data services in dynamic environment.…”

“…We extend the Berkeley DB [8] (an open source database from Oracle) to support real-time data services via the feed-forward and feedback schemes. The tested workloads are not only within but also beyond the operating range to explore the system behaviors extensively, unlike most existing work on feedback control of real-time data services [4], [3], [5], [6], [7] that only considers workloads within the operating range. The open-loop approach, i.e., the unmodified Berkeley DB, and feed-forward-only approach based on the CLT fail to support the desired data service delay bound even for relatively light workloads.…”

“…Therefore, our integrated proactive and reactive approach directly considers database semantics. The main contributions of our work are threefold: control scheme for database backlog management [7] to support the desired average and transient delay for data services. Our feed-forward and feedback control approaches adjust the backlog bound via admission control, if necessary, to support the desired average and transient delay for data services in both proactive and reactive ways that are complementary to each other.…”

“…To further enhance the performance of real-time data services, we integrate the feed-forward approach with a feedback-based admission control scheme for real-time data services [7]. By integrating them, we create the following benefits for real-time data services:…”

“…Based on the derived system model, a feedback controller is designed to adjust the system behavior, if necessary, to achieve the desired performance. In our previous work [6], [7], our approaches based on formal feedback control theory implemented in a real database system [8] significantly enhance the data service delay and throughput compared to the tested open-loop and ad hoc feedback approaches.…”