Survivability-oriented self-tuning of web systems

Bihuan Chen, Xin Peng, Yijun Yu, Wenyun Zhao2011 · DOI

TL;DR

This paper tackles the challenge of keeping complex web systems running smoothly even when faced with unpredictable workloads and component failures. They propose a novel "requirements-driven self-tuning" method that uses a value-based approach to automatically adjust both quality and functional aspects of the system at runtime to maintain a high level of "survivability."

What Problem Does This Paper Address?

Modern web systems are increasingly complex and operate in dynamic, often unpredictable environments. This can lead to situations where high workloads or subsystem failures degrade performance or even cause outages. Simply aiming for absolute reliability is often impractical and expensive. Instead, the authors argue for "survivability assurance" – the ability to maintain critical system services with acceptable quality, even under adverse conditions, by strategically sacrificing less crucial functionalities or quality aspects. The core problem is how to achieve this survivability autonomously and intelligently at runtime, without human intervention, by making informed tradeoffs.

Methods

The paper introduces a "requirements-driven self-tuning" framework built upon the MAPE (Monitor, Analyze, Plan, Execute) control loop. The key innovations lie in:

  • Value-Based Survivability: Survivability is framed as maximizing the "system value proposition," which is quantified by considering the business value generated by different system functionalities. This is achieved through a goal modeling approach where goals are enriched with "relative parent values" representing their contribution to the overall system value.
  • Requirements-Oriented Reasoning: The system uses a goal model that captures both quality and functional requirements. This model is used by a reasoner to make tradeoff decisions.
  • Runtime Tradeoffs: The method makes two types of tradeoffs at runtime:
  • Quality Tradeoffs: Relaxing less critical quality requirements (e.g., response time) to preserve more important ones.
  • Functional Tradeoffs: Unbinding less critical functional services to ensure essential services and their quality are maintained. These unbound services can be rebound later when conditions improve.
  • Feedback Controller: A PID controller is used to decide when to initiate tradeoffs and to rank the preference of goals for reasoning.

Datasets

Not discussed in the paper. The paper uses a conceptual online shopping system as an example for illustrating their goal model.

Key Insights and Ideas

  • Survivability as Value Maximization: Reframing survivability not just as uptime, but as maximizing the business value delivered to users, provides a more practical and economically driven approach to system resilience.
  • Runtime Tradeoffs are Crucial: The ability to dynamically adjust both quality and functional aspects of a system at runtime is essential for handling unpredictable environments.
  • Goal Models as the Brain: Using enriched goal models with value annotations allows for intelligent, requirements-driven decision-making about which parts of the system can be sacrificed or restored.
  • Distinction Between Quality and Functional Tradeoffs: The paper clearly delineates how quality tradeoffs involve adjusting performance/attributes, while functional tradeoffs involve enabling/disabling entire services, highlighting the different impact and complexity of each.

Key Results

The paper presents a conceptual framework and method. Specific quantitative results or performance evaluations are not detailed in this excerpt. The key result is the proposed method itself, which offers a structured approach to achieving survivability through autonomous, value-driven, runtime adjustments of both quality and functional requirements.

Limitations and Future Work

  • Timed Delay for Functional Tuning: The authors acknowledge that functional tradeoffs (unbinding/binding components) can be more radical and introduce uncertainty in their effect. They propose adopting a timed delay for the effect of these actions, as suggested by prior work, to mitigate oscillations.
  • Future Work: The authors intend to investigate the timed delay mechanism more deeply and precisely. They also plan to apply their method to the "new paradigm of cloud computing."

Summary updated

Abstract

Running in a highly uncertain and changing environment, Web systems cannot always provide full set of services with optim¬al quality, especially when the workload is high or failures in subsys-tems occur frequently. It is thus desirable to continuously maintain a high satisfaction level of the system value proposition, hereafter survivability assurance, while relaxing/sacrificing certain quality/functional requirements that are not crucial to the survival of the Web systems. In this paper, we propose a requirements-driven self-tuning method for survivability assurance of Web systems. Using a value-based feedback controller plus a requirements-oriented reasoner, our method makes both quality and functional requirements tradeoffs decisions at runtime.
BibTeX
@misc{chen2011survivability,
  author = {Bihuan Chen and Xin Peng and Yijun Yu and Wenyun Zhao},
  title = {{Survivability-oriented self-tuning of web systems}},
  year = {2011},
  doi = {10.1145/1963192.1963205},
}
Figure