Doctoral Thesis Proposal - Hojin Park

— 4:30pm

Location:
In Person - Reddy Conference Room, Gates Hillman 4405

Speaker:
HOJIN PARK, Ph.D. Student, Computer Science Department, Carnegie Mellon University
https://hojinp.github.io/

Public clouds are widely adopted for their scalability, flexibility, and reduced operational overhead, but optimizing costs while maintaining performance remains a significant challenge. This thesis addresses the problems of reducing storage cluster costs in public clouds and minimizing cross-cloud/region data access costs by exploiting the elasticity and diversity of public cloud resources combined with real-time workload monitoring. 

First, this work addresses the challenge of reducing storage cluster costs in public clouds through Mimir, an automated system that leverages heterogeneous storage types to determine storage cluster configurations that reduce costs while meeting workload performance requirements. Next, this thesis introduces Macaron, an auto-configuring caching system designed for cross-cloud/region data access. By dynamically adjusting cache capacity based on workload characteristics, Macaron minimizes remote data access costs and ensures low latency. Utilizing the object storage for caching storage type, Macaron demonstrates substantial cost savings compared to existing cross-cloud/region data access approaches. 

Building on Macaron, I propose a novel prefetching design for cross-region data access in public clouds that includes an object grouping-based prefetching algorithm and dynamic cache space allocation for demand-requested and prefetched data. This design aims to mitigate cache pollution, reduce wasted prefetches, and improve data access latency while minimizing additional costs. 

By addressing cost and performance optimization in public cloud storage and cross-cloud/region data access systems, this thesis enables reducing costs of utilizing public cloud resources for storage and caching. 

Thesis Committee

George Amvrosiadis (Co-Chair)
Gregory R. Ganger (Co-Chair)
Jignesh M. Patel
Carlo Curino (Microsoft Research)

Add event to Google
Add event to iCal