What Great Research ?s Can RAMP Help Answer? What Are RAMP’s Grand Challenges ?
Grand Challenges A complete software development vehicle for every parallel programmer –Debugger Functionally performance –Co-design with HW
Grand Challenges Enable education and training of (parallel) programming –By providing useful feedbacks –By offering an inexpensive accessible parallel system
Grand Challenges Enable the study of global system optimization so that a RAMP system is more efficient than “traditional computing systems” –For solving applications through parallelism
Research Questions Scalability issue of MP systems –Design issues with systems consist of large number of processors 1K 10K 100K
Research Questions CMP Architecture –Die area allocations tradeoffs Logic, memory and interconnect ratio –Power and thermal study –Interconnect topology evaluation –Performance study
Research Questions Full system simulation –I/O behavior –Flexible performance monitoring of all levels of a system –OS effects –Debugging methodology
Research Questions Modeling of very large systems –Generate workloads emulating a large system “Internet in a box” Input injection into a middleware/backend system
Research Questions Interaction of SW and HW –Acceleration of special functions –Performance tradeoffs of SW/HW –New instructions and their implementation study –Software assertion’s effect observed by HW for software debugging
Research Questions Synchronization of MP system –Lock speculation and transaction memory –Lock behavior modeling and evaluation Lock permutation
Research Questions Fault tolerance designs –Fault injection –Redundancy tradeoffs
Research Questions Others –Validation of new designs against existing designs –Speed up functional simulation –Security