CLUSTOPIUM PROJECT

 

-

Distributive Computing and Clustered Systems Client & Server

Created by Leon Yuhanov (C)Mean Smile Productions 2004, Groovinator Sowftare Company 2004 and Communet 2004

LATEST News

*A Unix/Linux verision is currently in develpment that takes Java out of the eqation, we have been playing around with TCP/IP in C++ and are about a month away from releasing a version of Clustopium that has a 3-fold performance boost. We are also in development of PIT, a Pie(3.141...) to the N generator which will impliment the Clustopium Clustering system.

*Version 2-002 In Development, ETA(End of April 2004). 2-002 will attempt to distribute a Learning Neural Net System using the Clustopium Machine, once again this is a test application to demonstrate the power of the Clustopium Machine. Currently doing beta tests of the JPerceptron Class

*Version 2-001 Completed, 2-001 the DNA Patern Analyzer. Will read DNA data and attempt to read reacurring patterns in record times. 2-001 is a test application to demonstrate the power of the Clustopium Machine. DOWNLOAD IT NOW, Full instructions included

*WOW Clustopium beats single CPU test run by 35.692 Seconds see the complete bench mark results HERE click HERE to see PRELIMINARY test results using large integer arrays. Test runs on a 1-CPU machine vs a 7-CPU (Clustopium)machine, show promising results, during 20 test runs, each starting at 98k ending at 100k, using 100wide incriments in aray sizes Clustopium beats single cpu machine EVERY TIME! See the cluster click here
One mans rubish is anothers treasure If you had a few old pcs lying around what would you do with them, take them appart and sell the parts? You would make more money selling bits then the systems as a whole. But what if you could take these machines and link them up to do usefull tasks such as sorting, seraching, calculations etc.. well thats exactly what i did!.
The Nodes(thrown out machines)

I have in my posession 7 identinal MINI-ITX PC's(my army of nodes grows larger every day!):

AMD K6v2(450mhz->OC to 500Mhz)64Meg Ram, DVD drive, 2 Fast Ethernet LAN cards, 20gig Harddisk, onboard everythign else.
The Server

The server im using is my desktop PC, as im developing on this mahcine i really couldnt be bothered adding another machine to my already overheated powersocket(20 Plugs from 1 conections):

AMD K6v2-3D(500Mhz) 380Meg Ram, 20Gig, 40gig, 120meg Harddisks, CDRW drive, FastEthernet LAN card, and everything else i could fit into this thing.
Current Configuration(HardWare) Thanks to a kind donation from my work of a very good Cisco 2900XL 24port Fast ethernet switch, and a nice old laptop from a friend who works in Swinburne Uni, i have the system set up i nmy room like a wall of computers.

Current Configuration(SoftWare)

 Now this is the tricky part, Current projects such as the Beowulf cluster are great but the software requirements are beyond me. So i wrote the Clustopium Load Balancing Server and Client.
How it works

Image you have a program which needs to calculate a cosmic number of problems, and fast. Now image that instead of loading a single processor with this execution, why not balance the proccesing power accross several nodes, in this case 7. Give each node a critical proccess and clone the nodes.

The CLBServer runs on the server machine whislt the Client runs on each individual node. given the base set of classes that enable you to interconnect each node to the server and send/recieved data back and forth, you can write aplicatiosn which REALY take advantage of multiple CPUs without having to deal in Distributive computing and incredibly costly Cluster Server Farms
Current Version Complete Internetwokring implementation is complete, source is available for sale for unlimited use within company/organization/bussines/group , please contact me directly by Email. Current Version allows up to 100 Nodes but is completely flexible to handle N Nodes.
     
The screen dump is of the first version, it has a fully interactive GUI whilst the client is current solely terminal based(dos/linux terminal). The current screen shot demostrates an active connection to a single node(node 2), node numbering starts at 0 and whilst the program is currently designed to accept 100 node conections, this is completely changable to accept N connections.This version is tested with 7 nodes because I have 7 nodes at my disposal and I could run benchmarks easily. As you can see all the nodes are represented on the left list box. As node 0is the only one connected its the only one available in the list. The debug screen at the top left had corner dumps debug information. The "Disengage Node" button will disconnect a selected node. The "Execute 2-001" button will start the algorithm. The "Exit" button will close all connections and quit the server.

Version 2-001 Reads a DNA data file and requets each node to search for sequencial sequences. 2-001 uses the "jdna" class which i whipped up for quick testing, at the end of each execution the Clustopium Machine runnign 2-001 will return every possible pattern count with in a set of data(in this case a 500Byte set of 'GATCAAGTC...'
 
Download

You can download the benchmark test and try it youreself. You can dl both server and client by following this link ---> Clustopium.zip

  
About

Clustopium is (C)2004 Leon Yuhanov, Mean Smile Productions(Melbourne, Australia), Groovinator Software Company(Melbourne, Australia), Communet(Melbourne, Australia)

Project Credits:

Leon Yuhanov - Clustopium Concept, Source Author, Hardware, Hardware and software configuration, Testing, benchmarking, Page design, Hosting(by IPrimus)

Project Contributors:

Jody Sacks - Idea redefinition, Concept Introduction of Communet, GPR and GPRnet, Node Testing

Daniel Aronson - Terminal Laptop for switch configuration

Jeremy Golberg - Load Balancing Idea redefiintion, Neural Nets Introduction, PIT Development(Pie Number Generator)

Rowan Cox- TCP/IP Streamlining Idea Redefinition

WWW Contributors/Sources:

Potato DNA Sequence Thanks to - www.ncbi.nlm.nih.gov

Neural Net Tutorial thanks to Prof. Jürgen Sauer - Neural Network Tutorial

Best WWW search engine - www.google.com