To content
Department of Computer Science

Model-based Validation of SLAS For Distributed Software Systems

Project start: 2010

This DFG-funded project aims to develop methods for validating SLAs for service-oriented architectures using simulative and analytical models.

  • Creation of models of SOAs to verify the performance of these systems.
  • Specification of performance constraints through formalization in the modeling language
  • Creation of an analytical model with performance bounds for SOA based on Network Calculus
  • Improvement of statistical models for evaluation of simulation results of SOA models

Today, large distributed software systems are implemented as service-oriented architectures (SOA). They are based on the idea that individual software components provide their services via networks. Services can thus be composed from simpler services. In the course of the spread of cloud computing, the boundaries between local applications, basic or composed services from the network are thus blurred.

When planning SOAs, it is not only correct functionality that is important, but also non-functional properties such as response time, reliability and throughput. If performance bottlenecks exist in individual components, this inevitably leads to the overall system not being able to provide the required performance. In this respect, a detailed specification of the performance and reliability to be provided as well as its validation is necessary as part of the design and operation of SOA.

For this reason, functional and non-functional properties of services are recorded in service level agreements (SLAs). These serve as a kind of contract between service provider and customer. The service provider can give guarantees for the performance of its service if the customer adheres to defined limits for the utilization.

The goal of this project is to improve the early performance evaluation of SOA by including SLAs with non-functional propertieson. This is to be done with the help of models. Thus, systems can be validated against SLAs in early design phases.
For modeling it is necessary to have a formal specification of non-functional properties in SLAs. In the project, the specification is to be developed starting from predefined basic patterns.

The next step is to examine the performance of the existing system in the model. Two different approaches are to be taken in the project: Analytical methods and simulation.

For the simulative evaluation of models of service-oriented architectures, the modeling environment ProC/B developed in SFB559 is complemented by statistical methods for SLA validation. This part of the project can build on extensive previous work (links here).

A method used for the analysis of data networks is the Network Calculus. It describes bounds for load and service rate for network elements and concludes their maximum response time. The bounds in Network Calculus are similar to the specifications of SLAs for service-oriented architectures, which is why we consider this approach of using Network Calculus for modeling SOAs to be appropriate.
The challenge for system modeling with Network Calculus considering SLAs is that the real performance of a service is hidden behind the SLA description. In particular, the processing speed of requests is unknown. Only the agreed upper load limit and the guaranteed response time are available for system modeling, the actual service rate is missing. This makes validation of models by Network Calculus difficult.

Our solution is to swap the input to Network Calculus. Instead of using bounds for the arrival rate and the service rate to determine the waiting time, the inputs to the SLA Calculus are bounds for arrival rate and waiting time from the SLA specification. The new calculus now provides the required service rate of the system. In further research, we also consider the combination of SLAs within composed systems.

Publications

  • Falko Bause, Peter Buchholz, Jan Kriege, Sebastian Vastag:
    Simulation Based Validation of Quantitative Requirements in Service Oriented Architectures.
    Winter Simulation Conference 2009 (WSC 2009), Dec. 13-16, Austin (Texas, USA), 2009
    [Abstract]  [PDF]  [BibTeX]
  • Sebastian Vastag:
    Modeling quantitative requirements in SLAs with Network Calculus
    In Proceedings of the 5th International ICST Conference on Performance Evaluation Methologies and Tools (ValueTools), ENS Cachan, France, May 17th-20th 2011, ICST.

    [Abstract]  [PDF]  [BibTeX]

  • Sebastian Vastag:
    A Calculus for SLA Delay properties
    accepted for the 16th International GI/ITG Conference on Measurement, Modelling and Evaluation of Computing Systems and Dependability and Fault Tolerance (MMB 2012)

    [Abstract]  [PDF]  [BibTeX]

CONTACT

Dipl.-Inform. Sebastian Vastag