Open Collaborative Research OCR on Multicore Exploitation

1
Open Collaborative Research (OCR) on Multicore
Exploitation

• Evelyn Duesterwald
• IBM T.J. Watson Research Center

2
Open Collaboration Research (OCR) Program

• An IBM funded program to support open
  collaborative research between IBM and
  universities in Computer Science (including
  related disciplines in Electrical Engineering and
  Math) and its applications in accordance with the
  Open Collaboration Principles
• IP openly published or available in royalty free
  public commons
• software available as open source
• Focusing on a limited number of strategically
  defined topics where open collaborative
  innovation would benefit IBM, the university
  collaborators, and the world at large
• Piloted in 2006
• Research topics Software Quality, Privacy
  Security Policy Management, Clinical Decision
  Support, Mathematical Optimization, Multicore
  Exploitation
• What makes it work?
• Multi-year, so that faculty can take on new
  students and obligations
• Collaborative, allowing IBM and university
  participants to forge deep relationships
• Open, providing maximum opportunity for others to
  build on the results
• Challenging, research requiring considerable
  innovation
• Well-funded, large enough to make a difference

3
OCR Program on Multicore Exploitation

• Late in 2007, Charles Lickel (Research VP
  Software) and Tilak Agrawala (Research VP
  Systems) approved to jointly fund 6 OCRs on
  Multicore Exploitation

4
Research on Multicore ExploitationA Taxonomy
Revolutionary Rethinking the Stack
Evolutionary Optimizing the Stack
Integration Blurring the Stack

• Mainstream workloads
• Incremental address software inertia by steady
  advances
• Enhance integration through new software
  interfaces for increased performance

• Special Domain
• Break-away radically application-driven
  large-scale parallelism should be effortless
• Extreme heterogeneity alternative system design
  point to enable new workloads

Operating System
5
Where the top 6 proposals are in the taxonomy
Revolutionary
Evolutionary
Revolutionary Rethinking the Stack
Evolutionary Optimizing the Stack
Integration Blurring the Stack
Optimistic Concurrency Workloads (UT Austin)
Concurrency Libs SUNY Oswego) XMPLAR
(Rice) Scheduling (CMU)
Liquid Metal (Berkeley)
Virtualization (Georgia Tech)
6
OCR Process

• What is needed to approve second year payment?
• Evaluation close to end of your first year
  (Oct/Nov) with the sponsoring VPs Charles Lickel
  and Tilak Agrawala
• In-person review meeting at Watson
• Presentation of research results to VPs

7
OCR Workshop Goals

• Technical advances
• Foster collaboration across OCR projects
• Exploit Synergies
• Please post your presentations in the Wiki (or
  send them to me so I can post them)!