mudassir ppt
TRANSCRIPT
HIGH-PERFORMANCECOMPUTING (HPC) FOR
CEMThe Future
By
Syed Mudassir Rehman
ID: 16160367
When Computational Electromagnetics was progressing, Computer Hardware was advancing.
High Performance Computing became an unavoidable research direction adopted by CEM community.
Massive Computational load and significant speedups could be shared and achieved using Distributed CPU and Distributed memories.
Introduction
Major HPC Platforms
1. Computer Clusters
*very economical*cluster with 10-20 nodes*can achieve upto 10-20 times computational acceleration
2. SuperComputers
*Highly optimized Computing Platform*10s of 1000s of nodes can be achieved
3. Cloud Computing
*still growing*emphasizes services and resource sharing*its impact still unknown
They provide great amounts of computing power
Increased difficulty in programming
Simplifies the process of building, testing and tuning an application
Used in Commercial computing for applications such as databases and web servers
Parallel Computer Memory Architectures
Shared Memory System
Interconnect may be a simple bus or a sophisticated switch
Types
Distributed Memory System
It may either be Distributed Shared Memory System or a Simpler Shared-Nothing System, depending on the capabilities of the Network Interface (NIC)
Types
Hybrid Distributed-Shared Memory System
Two or more Symmetric Multiprocessors sharing the same Memory. Thereby forming a Distributing System using networks
Types
Maintaining Scalability when nodes increase.
If there is little communication between the nodes, then the algorithm is embarassingly or pleasingly parallel.
To optimize Scalability and to balance the workload, nodal processes are synchronized and internode communications are reduced.
Challenge
Fast Multipole Algorithms (FMA)
Fast Illinois Solver Code (FISC)
FMA-FFT Method
Finite-Difference Time-Domain Method (FDTD)
Solution
Computational Electromagnetics had been advanced by integration of physics, mathematics and computer technologies.
Parallel Architectures continue to be an active and exciting area of research.
Trend for parallelism at all levels is increasing (2 to 16 processors).
Systems continue to be developed.
Conclusion
Thank You