[Colloquium] Tapan Srivastava MS Presentation/May 24, 2023

[meganwoodward at uchicago.edu]

This is an announcement of Tapan Srivastava's MS Presentation
===============================================
Candidate: Tapan Srivastava

Date: Wednesday, May 24, 2023

Time: 10 am CST

Location: JCL 298

M.S. Paper Title: Saving Money for Analytical Workloads in the Cloud

Abstract: As users migrate their analytical workloads to cloud databases, it is becoming just as important to reduce monetary cost as it is to improve query performance.  In the cloud, users choose to pay based on either the compute time or the amount of data a query reads.  We observe that analytical queries are compute- or IO-intensive and each query type executes cheaper in a different pricing model.  We exploit this opportunity and propose methods to build cheaper execution plans across pricing models. We implement these methods, observe how SQL syntax and query optimizers impact cost, and consider a transparent deployment of DuckDB on infrastructure-as-a-service to avoid the cost premiums of platform-as-a-service without placing the burden of deployment or maintenance of software on users.  We reduce workload costs by as much as 80% and produce execution plans spanning multiple pricing models that save as much as 47%. We also reduce individual query costs by as much as 90%. We simulate the effect of different cloud prices on real query performance and observe that there are still significant multi-cloud opportunities if cloud vendors change their prices.  These results indicate the massive opportunity to save money by building execution plans across multiple pricing models. 

Advisors: Raul Castro Fernandez

Committee Members: Aaron Elmore, Michael Franklin, and Raul Castro Fernandez