Computer Science and Engineering, Department of


First Advisor

Ying Lu

Date of this Version


Document Type



Puttaswamy, Ravi Kiran. Scale-Out Algorithm For Apache Storm In SaaS Environment, THESIS, December 2018


A THESIS Presented to the Faculty of The Graduate College at the University of Nebraska In Partial Fulfilment of Requirements For the Degree of Master of Science, Major: Computer Science, Under the Supervision of Ying Lu. Lincoln, Nebraska: December, 2018.

Copyright (c) 2018 Ravi Kiran Puttaswamy


The main appeal of the Cloud is in its cost effective and flexible access to computing power. Apache Storm is a data processing framework used to process streaming data. In our work we explore the possibility of offering Apache Storm as a software service. Further, we take advantage of the cgroups feature in Storm to divide the computing power of worker machine into smaller units to be offered to users. We predict that the compute bounds placed on the cgroups could be used to approximate the state of the workflow. We discuss the limitations of the current schedulers in facilitating this type of approximation as the resources are distributed in arbitrary ways. We implement a new custom scheduler that allows the user with more explicit control over the way resources are distributed to components in the workflow. We further build a simple model to approximate the current state and also predict the future state of the workflow due to changes in resource allocation. We propose a scale-out algorithm to increase the throughput of the workflow. We use the predictive model to measure the effects of many candidate allocations before choosing it. Our approach analyzes the strengths and drawbacks of Stela algorithm and design a complementary algorithm. We show that the combined algorithm complement each others strengths and drawbacks and provides allocations to maximize throughput for much larger set of scenarios. We implement the algorithm as a stand alone scheduler and evaluate the strategy through physical simulation on the Apache Storm Cluster and on software simulations for a set of workflows.

Adviser: Ying Lu