Adapting Key-Value Storage Systems to Minimize Cost in the Public Cloud
Open Access
- Author:
- Alfares, Nader
- Graduate Program:
- Computer Science and Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- March 24, 2020
- Committee Members:
- Bhuvan Urgaonkar, Thesis Advisor/Co-Advisor
George Kesidis, Thesis Advisor/Co-Advisor
Chitaranjan Das, Program Head/Chair
Viveck Ramesh Cadambe, Thesis Advisor/Co-Advisor - Keywords:
- Public Cloud
Strong Consistency
AWS
Memcached
Object Sharing
Storage Systems
Systems - Abstract:
- The growth of big data analytics and the volatility/diversity of pricing across public cloud services have presented many opportunities for optimization tailored to the need of an application. We focus our optimization on key-value storage systems in the public cloud. Such systems are widely popular due to their simplified semantics that allows applications to scale rapidly in multi-user environments, producing challenges on resource management and fairness guarantees. Since public cloud providers grant tenants to place their resources in specific regions (or datacenters), we divide our work into two settings of optimizations. First, we consider an optimization of a system within a single datacenter. Specifically, we deal with the caching layer that is often placed along with the database. We conduct our work by considering an object sharing framework to minimize storage cost. Second, we consider an optimization of a geographically distributed storage system that guarantees linearizability while meeting Service-Level Agreement (SLA). We rely on collected data for public cloud pricing models and performance measurements in our optimization formulation. In addition, we consider an optimistic approach and heuristic based optimizations.