Enhancing Rules For Cloud Resource Provisioning Via Learned Software Performance Models

- ASE 2015 Online Appendix

This web page is a companion to our ASE 2015 submission entitled "Enhancing Rules For Cloud Resource Provisioning Via Learned Software Performance Models".

Cloud computing is a system service model in which stakeholders deploy and run their software applications on a sophisticated infrastructure that is owned and managed by third-party providers. The ability of a given cloud infrastructure to effectively re- allocate resources to applications is referred to as elasticity. Of course, in practice, clouds are not perfectly elastic. Since cloud providers must provide elastic cloud services to a wide range of customers, their cloud platforms do not provision their resources precisely and automatically for specific applications. At the same time it is currently infeasible for cloud providers to allow customers to guide the cloud on how best to elastically provision their applications.

To significantly reduce the cost of deploying software applications in the cloud, we solve a fundamental problem at the intersection of cloud computing and software performance testing. Our core idea is to automatically learn behavioral models of software applications during performance testing to synthesize provisioning strategies that are automatically tailored for these applications. With our idea, the problem of precise cloud elasticity is translated into enabling a feedback-directed loop between software development and cloud deployment. We implemented our approach and applied it to two software applications in the cloud environment, namely Cloudstack. Our experiments demonstrate that with our approach the cloud is able to provision resources more efficiently, so that the applications improve their throughput by up to over 40%.

1. Learning performance rules



2. Subject Applications


JPet Storewebsite
Dell DVD Storewebsite;  databases


3. Results

3.1 Throughputs for JPetStore using Cloudstack and PRESTO




3.2 Throughputs for DellDVDStore using Cloudstack and PRESTO




4. Tools & Databases


5. Authors

  • Mark Grechanik - University of Illinois, Chicago, USA.
    Email: drmark at uic dot edu
  • Qi Luo - The College of William and Mary, VA, USA.
    E-mail: qluo at cs dot wm dot edu
  • Denys Poshyvanyk - The College of William and Mary.
    E-mail: denys at cs dot wm dot edu
  • Adam Porter - University of Maryland at College Park.
    E-mail: aporter at cs dot umd dot edu