Академический Документы
Профессиональный Документы
Культура Документы
Mohammad Malkawi
Dean of Engineering,
Jadara University
mmalkawi@aimws.com
Outline
Complex architecture
High performance
Easier to program
Easier to use
HPCS Goals
Productivity
Processing power
Memory
Interconnection
HPCS Challenges
High Effective Bandwidth
Robustness
HPCS Challenges
Performance Measurement and
Prediction
Scalability
Productivity Challenges
Application codes
Workflow
Productivity Dilemma - 1
Coding
Debugging
Optimizing
Modifying
Over-provisioning hardware
Productivity Dilemma - 2
Productivity Dilemma - 3
Language Challenges
SUN Fortress
IBM - X10
Cray Chapel
Open MP
Trade-Offs
X10 Features
Cascade Approach
Scalable performance
Improving programmability and breadth
of applicability.
Balanced systems also require fewer
processors to scale to a given level of
performance, reducing failure rates and
administrative overhead.
Data layouts
Work decomposition and communication.
Simplifies the creation of the basic algorithms
Allows these structural components to be
gradually tuned over time.
Clusters
Ultrascale
Rich bandwidth
Low latencies
Very high levels of fault tolerance
Highly integrated toolset to scale the
program and not the programmers
Multithreading technologies ( > 100
concurrent threads)
Proximity Communication -2
Proximity Communication -3
Proximity Communication
Capacitive coupling enables high-speed
data communication between
neighboring chips without the need for
wires of any kind
reduced power
communication power
Proximity Communication -4
Proximity Communication -5
Cloud Computing
Service computing
The net is the computer
More than 100 vendors
Growing fast
Programming environment
BACKUP SLIDES
HPCS Technologies
Some Publicly Announced Projects
Virtualization
Security
Resource management
Affinity control
Resource limits
MPI Paradigm
Productivity Challenges
Productivity Challenges