Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.programming > #4624 > unrolled thread

Scalability and Parallel Dual Simplex

Back to article view | Back to comp.programming

Contents

  Scalability and Parallel Dual Simplex Ramine <ramine@1.1> - 2014-06-16 20:07 -0700

#4624 — Scalability and Parallel Dual Simplex

Hello,


I have done some projections for my parallel Dual simplex
solver , and since the Parallel part of the Amdahl equation
is using Divide, Multiply and Add asm instructions on two doubles
this will enhance the speed of my parallel Dual Simplex, since on DDR3 
the throughput can go up to 17 MB/s and on my 2.4 Ghz CPU
this will give 8 bytes memory transfers per clock from the memory
to the  CPU, the memory transfers are serial and Multiply, Divide and 
ADD asm instructions on two doubles are parallelizable, and since the 
multiply and add asm instructions will take more CPU clocks and since
8 bytes can be transfered from memory to the CPU in one clock
so the serial part of the memory transfers of the Amdahl equation
is smaller than the parallel part of the asm instructions like
the Divide and Mutiply and ADD asm instruction on two doubles, so this 
will give around 4X scalability on larger problems, but for smaller 
problems it will scale more cause the transfer from cache to cache is 
less expensive.


But 4X scalability on larger linear programming problems is not so good, 
so i have decided to not write my parallel Dual Simplex Solver.

But what do you think ? do you want to me to code it ?
is 4X scalability a good think for you ?



Thank you,
Amine Moulay Ramdane.

[toc] | [standalone]

Back to top | Article view | comp.programming

csiph-web