modeling and simulation for architecture: breakout arun r. krste a. dave m. doe h.y. derek c. lizy...
TRANSCRIPT
Modeling and Simulation for Architecture: Breakout
Arun R.Krste A.
Dave M. Doe H.Y.Derek C.
Lizy J.Kevin L.Dean K.Karen B.
Shekhar B.
Mike P.Chris C.Jim K. John H.
Sudhakar Y.Luiz C.
Steve R.Wilf P.Jim K.
.
Participants
Moderator and Scribe
Darren K. and Arun R.
High level• Lots of discussion about the SW Issues & Apps• Major concern on interaction & interfaces
– between communities– “common language”
• Good at iterative development– Need good starting points– Need an informed path
• Lots of discussion of HW Design space– Caution over multiple metrics– Know performance well– Weaker on power– don’t know about reliability– Cost very hard, but very important
• “Cultural” Barriers & IP Issues
1. Object of Modeling: What to Model
• Goal: “Make Design Decisions”– Gain Insight– “Common Language” “Shared Testbed”
• Need to Support Co-design Loop– Arch. want Application models– Apps. want Architecture models
• Abstraction is needed to deal with Complexity• Different goals for different times/audiences
– Design exploration in terms of power, performance, reliability, and cost– Design validation– SW development
• Lots of parameters in the design space
2. Can we do it now
• Good at iterative refinement• Lots of discussion on need for abstract model w/ parameters
& trend lines (abstract machine model)– Iteratively approaches– Iterative series of definitions (multiple scopes, levels of
abstraction)– Good at evolutionary approaches– captures key parameters– capture disruptive changes– Correct abstractions are CRITICAL
• Caution: radical changes make things difficult, uncertain– Hard to jump 10+ years out
3. Gaps• Some research gaps, some interactional gaps
• “Cultural Barriers”– IP: what is precompetitive? what can we share?– App. vs. HW vs. compiler vs. OS etc...
• Skeleton Applications & Skeleton Architectures– Multiple levels– Correct abstractions– Not assume a single execution model
• Need way to interact between HW & SW• Fidelity vs. Time vs. Scale tradeoffs (decomposability)• Methodology
– What scale? what abstractions? – Need community consensus of what is needed
• Roadmap of Modeling capabilities
Gaps (cont.)
• Infrastructure– Technology models– Lots of discussion for tech. models for disruptive
technologies and how we trust them• Modeling Testbed– Bring together architecture & application skeletons @
Multiple levels of abstraction w/ Technology Modeling• Resources– Access to state of art machines for simulation– Access to state of art machines for validation
4. Community of Interaction
• Application Teams (Co-Design Centers)– Also, OS, Runtime, Compiler
• Different facets within architecture teams– Memory– Processor– Network– Device-level & High level
• Need lots of cross-cutting interaction
5. Measure for Success
• #1: No Surprises in Deployed Machine• Infrastructure is available, used, and broadly
accepted• Influence is demonstrated– Impact of methodology is demonstrated– We know which gaps we filled