The Demand for Computational Speed reasonable time period grand challenge problems 1 weather forecas

1
Lecture 1

• The Demand for Computational Speed- reasonable
  time period? - grand challenge problems 1)
  weather forecasting 2) modeling motion of
  astronomical bodies- speedup
• Types of Parallel Computers- shared memory
  multiprocessor system- message-passing
  multicomputer- distributed shared memory-
  MIMD/SIMD classifications
• Cluster Computing

2
Modeling and Simulation of DNA Structures
3
Global Weather Forecasting
4
(No Transcript)
5
(No Transcript)
6
All pictures are from http//antwrp.gsfc.nasa.gov/
apod/
7
1 Earth Simulator

• The ES is based on
• 5,120 (640 8-way nodes) 500 MHz NEC CPUs
• 8 GFLOPS per CPU (41 TFLOPS total)
• 2 GB (4 512 MB FPLRAM modules) per CPU (10 TB
  total)
• 640 640 crossbar switch between the nodes
• 16 GB/s inter-node bandwidth

Rmax 35.86TFlops
8
2 ASCI QLOS ALAMOS HP ALPHASERVER SC
Rmax 13.88TFlops

• 3072 AlphaServer ES45s from Hewlett Packard
  (formerly Compaq)
• 12,288 EV-68 1.25-GHz CPUs with 16-MB cache
• 33 Terabytes (TB) memory
• Gigabit fiber-channel disk drives providing 664
  TB of global storage
• Dual controller accessible 72 GB drives arranged
  in 1536 51 RAID5 storage arrays, interconnected
  through fiber-channel switches to 384 file server
  nodes.

9
http//www.top500.org/lists/2003/06/top5.php
10
Myrinet
http//www.myricom.com/myrinet/overview/index.html
11
MIMD Model
This model is
equivalent to a modern day assembly plant and is
the fastest and most efficient model. In this
example, our plant is required to build 100 cars.
Here, multiple assembly lines are given
instructions (multiple instructions) to build
automobiles all contributing to the goal of
building 100 cars. Each assembly line is assigned
a number of cars to assemble (multiple data) and
works independently of every other line. The
speed with which one line is able to complete a
car does not affect the speed of the other lines.
Instructions for what each line will do comes
from management who oversees and coordinates the
work.
12
SIMD Model
The SIMD Model (Single Instruction, Multiple
Data) lends itself well to the automobile
assembly line analogy. In this scenario, one
worker executes the same task (single
instruction) on multiple cars (multiple data) as
they come down the assembly line. This worker
does not have a choice of tasks and is dependent
on the arrival of data (automobile) to perform
the task.
13
Speedup Factor

• Ts 100sec, Tp 20sec ( nodes 7)
• S(n) 100/20 5Maximum speedup 7 (linear
  speedup)
• Super-linear speedup- more memory space-
  problem dependant factors

14
Amdahls Law
150 sec
10 sec
5 sec
10 5 15 sec
15
Amdahls Law (2/2)
150 sec
10 sec
5 sec
150/15 10
Maximum speedup 150/10 15 ( processors gt 28)