Darwinism

History

谢桂纲 5ae9946d3b update source file license headers		8 years ago
..
Demo	update source file license headers	8 years ago
images	improvements on the namespace	10 years ago
Demo.sln	not sure, needs a interval media object	10 years ago
GA_distribute.pptx	add image	10 years ago
README.md	improvements on the namespace	10 years ago

README.md

Distribute Computing Demo: Genetics Algorithm

The speed limitation step in GA estimates

One of the speed limitation step in GA eastimates is the fitness sorts for selects the best population that can evolved. For calculate the fitness of a parameter set, ODEs will be computed and the Euclidean distance will be calculated between the estimates function and observation data. This calculation operation will takes a long time to run if the data resolution is very high and population is very large.

For a GA estimation that running on a single machine, the only way to boost this operation is using parallel linq, but a single machine is too power limited, so that a server cluster will required for solving this large scale data analysis problem. For solving a system parameter estimates problem using GA, in a server cluster, then we needs distribute computing.

For running GA on a single machine, makes it run on a server cluster, we just needs devide its population into sevral parts, and then push the fitness calculation task onto each single machine in our server cluster, then we can makes this GA analysis running in distribute method.

A basic calculation in GA

Distribute Linq & Parallel Linq

Here is a example of Parallel Linq:

Dim LQuery = From x As <type> In source.AsParallel Select <.ctor>

And here is a example of Distribute Linq, its usage is very similar to parallel linq:

Dim LQuery = From x As <type2> In source.AsDistribute(task) Select <.ctor>

How it works?

Using Distribute Linq in GA

VirtualBox demo

Config Server

This demo was running on a latest Ubuntu server, after complete the installation of the Ubuntu server, then you should install the mono first by using:

sudo apt-get install mono-complete

And then deploy the server side runtime of this distribute environment just by upload the program to a directory and then start by command

mono ./dlinq.d /start

Here I using the virtual machine clone in the VirtualBox to simulate a local server cluster for this demo test:

Running the demo in host,

./dga_demo