05/28/2021
By William Moloney
Doctoral Dissertation Proposal Defense Announcement, Computer Science Dept.
Title: Provisioning of Cloud Systems with Minimal Tenant Disruption
Ph.D. Candidate: Pavel Tatashin
Time: Tuesday, June 1, 2021, 11 a.m. EDT
Zoom Meeting: This will be a virtual defense via Zoom. To make it more accessible to cs-msgs recipients, the Zoom link is: https://uml.zoom.us/j/99159409980
Committee Members:
- William Moloney (advisor), Associate Professor, Computer Science Department, University of Massachusetts Lowell
- Jay McCarthy, Associate Professor, Computer Science Department, University of Massachusetts Lowell
- Tom Wilkes, Assistant Teaching Professor, Computer Science Department, University of Massachusetts Lowell
- Steve Sistare, Software Architect, Oracle Corp.
Abstract:
Cloud vendors provide slices of systems (virtual machines) to tenants with various guarantees of availability, performance, and stability. The stability and performance guarantees require the software stack and hardware to be kept updated, while the availability guarantee prohibits long downtimes. Thus, the vendors face the challenge of being able to perform minimally-disruptive (from a tenant's point of view) software and hardware provisioning on their systems.
This work encompasses an analysis and discussion of various techniques that can be used in order to allow cloud vendors to maintain stable and updated server farms with a minimal or possibly no downtime or slowdowns of the virtual machines that
they are hosting. The methods that are tested are quantified based on downtimes and slowdowns from the tenant's point of view when the underlying host software and/or hardware are being updated.
In addition, in this work we design and describe a reference virtualization appliance software stack that we use to analyze cloud hardware. The reference software stack is an image that is composed of a Linux kernel, a virtual machine manager and libraries. The total image size is small (under 20MiB), but has the necessary capabilities to handle real cloud work.