PARALLEL DATA LAB 

PDL People

George Amvrosiadis


Contact:
www |
Office:
RMCIC 2311
Mailing Address: Electrical & Computer Engineering
Carnegie Mellon University
5000 Forbes Avenue
Pittsburgh, PA 15213-3891
Position:
Associate Research Professor
Projects: Delta FS, Mimir, Zoned Storage

 

Research Interests:

My research interests lie in the areas of:
  • Scalable storage performance and reliability;
  • Operating and distributed systems;
  • Data analysis and modeling
My current primary research objectives are:
  • Reimagine file systems as transient, light-weight services that applications can instantiate, scale, and optimize on-demand. The goal is to propose a (timely!) alternative to parallel file systems with fully synchronous namespaces, which will remain performant as we transition to exascale systems and beyond.

  • Develop efficient mechanisms that turn the OS into a gray box. A step in that direction is Duet [4], a framework that notifies processes when data of interest becomes cache-resident. This allows groups of processes with overlapping data accesses to adjust their workflows in order to improve their collective efficiency. Duet has been used to improve the efficiency of maintenance tasks [4], job schedulers [5], and file system monitoring tools [1].

  • Discover and fix pain points in systems through data analysis and modeling of field data. Some examples: improving the reliability of enterprise backup systems [2, 5], characterizing the effect of temperature on the reliability of datacenter hardware [6, 8], and mitigating the impact of maintenance work by predicting when a storage device is expected to be idle [7].