*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Scalable Main-Memory Object Management
Alexander Merritt
School of Computer Science
College of Computing
Georgia Institute of Technology
Date: Tuesday, May 31, 2016
Time: 10AM to 12PM EST / 7AM to 9AM PDT
Location: KACB 3100
Committee:
---------------
Dr. Karsten Schwan (Advisor, School of Computer Science, Georgia Tech) Dr. Ada Gavrilovska (Committee Chair, School of Computer Science, Georgia Tech) Dr. Taesoo Kim (School of Computer Science, Georgia Tech) Dr. Kishore Ramachandran (School of Computer Science, Georgia Tech) Dr. Moinuddin Qureshi (School of Electrical and Computer Engineering, Georgia Tech) Dr. Dejan Milojicic (Hewlett Packard Labs, Hewlett Packard Enterprise)
Abstract:
-----------
New and emerging memory technologies are giving rise to servers with massive pools of main memory, but these systems are difficult to program efficiently: terabytes of memory, disaggregated bandwidth, and hundreds of cores pose scalability challenges for all layers in the software stack. Transparent, granular operating system interfaces make it difficult and inefficient for applications to express semantic relationships with their data. Library allocators are subjected to much larger scales they were not designed for, as well as increasingly complex allocation behaviors, creating high memory fragmentation.
To navigate these challenges, this thesis proposes (1) new memory-centric operating system abstractions to more effectively manage and share both virtual and physical memory without interfacing with the filesystem or network APIs, and (2) a log-structured memory object allocator that leverages these new abstractions to more effectively make informed decisions about where data is placed and how it is accessed, scaling up to hundreds of cores by partitioning and decentralizing its components across large shared-memory platforms. We demonstrate the feasibility of this approach with data-intensive applications and workloads, including a tightly-coupled image analytics pipeline we developed, to stress the real-time capabilities of our solution.
