Wednesday, 13 December 2017

Session 2: Cloud Management

On-line Bayesian Context Change Detection in Web Service Systems

Authors:

Jakub M. Tomczak (Institute of Computer Science)
Maciej Zieba (Institute of Computer Science)

Abstract:

In real-life situations characteristics of Web service systems evolve in time. Therefore, change detection techniques become substantial elements of adaptive procedures for Web service systems management, such as resource allocation and anomaly detection methods. In this paper, we propose an on-line change detector which uses the Bayesian inference. We define two models which describe situations with one change and no change within data. Next we apply Bayesian model comparison for change detection. In order to obtain analytical expressions of model evidences used in the model comparison we provide a coherent framework of change detection which focuses on an approximation of the Bayes factor. The proposed solution, contrary to state-of-the-art methods, works in an on-line fashion and the algorithm’s computational complexity is proportional to the constant size of the shifting window. Low computational complexity of the change detector enables its application in complex computer networks. At the end of the research paper, the quality of the proposed algorithm is examined using simulated Web service system.

DOI: 10.1145/2462307.2462311

Full text: PDF

[#][]

Addressing Self-Management in Cloud Platforms: A Semantic Sensor Web Approach

Authors:

Rustem Dautov (The University of Sheffield)
Dimitrios Kourtesis (The University of Sheffield)
Iraklis Paraskakis (The University of Sheffield)
Mike Stannett (The University of Sheffield)

Abstract:

As computing systems evolve and mature, they are also expected to grow in size and complexity. With the continuing paradigm shift towards cloud computing, these systems have already reached the stage where the human effort required to maintain them at an operational level is unsupportable. Therefore, the development of appropriate mechanisms for run-time monitoring and adaptation is essential to prevent cloud platforms from quickly dissolving into a non-reliable environment. In this paper we present our approach to enable cloud application platforms with self-managing capabilities. The approach is based on a novel view of cloud platforms as networks of distributed data sources - sensors. Accordingly, we propose utilising techniques from the Sensor Web research community to address the challenge of monitoring and analysing continuously flowing data within cloud platforms in a timely manner.

DOI: 10.1145/2462307.2462312

Full text: PDF

[#][]

Behavioral Model for Cloud-Aware Load and Power Management

Authors:

Kiril Schröder (University of Oldenburg)
Wolfgang Nebel (University of Oldenburg)

Abstract:

Within the last few years, the development of data centers has been moving into high-grade flexible architectures that adapt to the needs (by means of virtualization). This flexibility can be used by load management methods to minimize the energy demand. Depending on quality of service and the hardware used, the application of a load and power management (LPM) results in a big dynamic range of the number of servers currently required. Previous energy models for data centers did not take into account this dynamic sufficiently and thus are not suitable for cloud data centers. Therefore, we present two contributions in this paper. First, we enhance an existing LPM for virtual machines, which has been designed for single data centers, enabling it to interact in flexible environments, for example in inter cloud LPM systems. Second, we develop a model which abstracts the behavior of the LPM concerning the server allocation. This model can be consulted for forecasts and obtains an average precision of 93%.

DOI: 10.1145/2462307.2462313

Full text: PDF

[#][]