Analyzing the Impact of Compute Server Selection for Function Placement with Distributed Storage

Restricted (Penn State Only)
Author:
Deshpande, Madhav Venkatesh
Graduate Program:
Computer Science and Engineering (MS)
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
May 14, 2021
Committee Members:
  • Dong Xie, Thesis Advisor/Co-Advisor
  • Bhuvan Urgaonkar, Thesis Advisor/Co-Advisor
  • Chitaranjan Das, Program Head/Chair
Keywords:
  • Function Placement
  • Serverless
Abstract:
Distributed systems and cloud computing are becoming more prevalent due to an increase in the demand for large compute resources with access to storage. This has led to the natural separation of machines for storage related functions and compute related functions. High speed networking has also made moving data across the network seem like a feasible solution. However we leave performance on the table when we could be executing code on storage servers and avoiding the process of sending data unnecessarily over the network. We look into realistic scenarios, where all the data used is not stored in a single machine but is stored across multiple systems. We have created a framework for just this scenario and have built it so that various customizable storage configurations can be tested against custom benchmarks. This work tests various storage configurations using the framework and presents the results. We analyze these results and provide our rationale for the observed behavior.