Committed to Code

A Gnutella/Pastella client to share CPU-time and a framework for distributed computing on P2P grids. We support peaceful, free and open research and hope to build an internet supercomputer. Currently, we can render movies on a cluster of 10 computers.

Current applications which run on the GPU network are a distributed search engine, a terragen landscape generator, a P2P distributed wan storage, a climate simulation frontend and many experimental plugins.

Code Analysis


News

GPU II preview for developers (0.5.0)

GPU is currently undergoing its 4th rewrite, codenamed freedom (3th was named solaris which reached version 0.963, 2th polaris, 1th unnamed). The preview package has version 0.5.0. This package is solely intended for developers as there is still no ... [More] client working, just a bunch of unit tests.

The package is intended to sketch the whole idea of the next generation GPU. Some people might be disappointed that GPU moves away from peer-to-peer to become a client-server architecture, but we believe this is the only drawback. The highlights of the coming release (probably 6 months from now) are:
- client/server architecture should allow scaling of nodes up to several thousands of clients
- client is in Freepascal and can be compiled on Windows, Linux and Mac OSX
- Server is in PHP/mySQL and runs everywhere where LAMP stack runs
- All software used to develop GPU II is Open Source and available freely
- The plugin architecture of GPU I is retained, although the plugin layer has been rewritten (available in preview) and plugins need to be ported as well
- Communication is entirely over HTTP, workunits can be downloaded, and resulting workunits can be uploaded via HTTP back to the server (available on GPU preview package)
- HTTP makes GPU transparent to most proxies, so that clients can run also in companies and not only at home
- GPU II retains features like channels (to run whiteboard, chat and sensor networks), job management, clustering services, all rewritten in light of the experiences
done with GPU I (available on GPU preview package)
- GPU II will stay a multipurpose distributed computing architecture, so that any client and any server will support any computation (unlike the BOINC architecture)
- Each server should be able to support 40-100 clients according to some initial computations, the network should scale with number of servers
- One superserver will manage the whole network of servers

The package contains most of the server, communication layer client and server side, plugin layer clientside.
we hope the release will fuel some interest into the GPU project.
Have fun! [Less]


0.960: volcano Eyjafjallajoekull in Island

0.960: climate simulation simulates vulcano Eyjafjallajoekull

GPU 0.960 features a simple climate simulation. With the latest update (0.9607) which can be retrieved with the menu "Tools->Autoupdate to latest version" on the ... [More] main GPU application, the climate simulation
also shows the path of ashes spread by the vulcano Eyjafjallajökull and the CO2 production of the world population.

However, it is difficult to calibrate the parameters of the model. Anyone can try simple tasks like: having a world with stable temperature at 15.5 Celsius, start a nuclear war, move Earth down to Mercury orbit or up to Iupiter, stop earth rotation or revolution, etc.

A screenshot of the climate simulation can be seen at https://sourceforge.net/project/screenshots.php?group_id=58134&ssid=132912
Source code of this frontend can be found under /lib/libclimate and /src/frontend/thedayaftertomorrow. [Less]


Super C.R.A.P.

This is a preview of what is being built right now. Super C.R.A.P is an acronym for Super Cluster Ready At Processing, core of the system is a Rollin KVM switch with 8 outputs (VGA, mouse, keyboard) bought for 218 CHF (=140 €). Additionally, 8 ... [More] cables with VGA, mouse and keyboard are needed (each one costs about 15 €). An old 10 MBit Ethernet Switch will provide connectivity to the nodes of the cluster.

The cluster which now features 4 computers (2 XP, 1 Gentoo and 1 Sun) will be extended to 8 computers, all computers are coming from people who substituted their one and therefore are a little bit old, but very cheap (got them for free). The Rollin KVM Switch does not support the Sun Ultrasparc, so it will be put up separately.

The cluster will be both a showcase and a sounding board for GPU. Particular care will be needed when operating the 8 computers simultaneously, the necessary power will be taken from different phases. Estimated consumption is about 8x500W + 200 W (LCD screen) = 4200 W which is quite a big oven :-)

The cluster should be fully operational for beginning of 2010. (1 comments) [Less]


Super C.R.A.P.

This is a preview of what is being built right now. Super C.R.A.P is an acronym for Super Cluster Ready At Processing, core of the system is a Rollin KVM switch with 8 outputs (VGA, mouse, keyboard) bought for 218 CHF (=140 €). Additionally, 8 ... [More] cables with VGA, mouse and keyboard are needed (each one costs about 15 €). An old 10 MBit Ethernet Switch will provide connectivity to the nodes of the cluster.

The cluster which now features 4 computers (2 XP, 1 Gentoo and 1 Sun) will be extended to 8 computers, all computers are coming from people who substituted their one and therefore are a little bit old, but very cheap (got them for free). The Rollin KVM Switch does not support the Sun Ultrasparc, so it will be put up separately.

The cluster will be both a showcase and a sounding board for GPU. Particular care will be needed when operating the 8 computers simultaneously, the necessary power will be taken from different phases. Estimated consumption is about 8x500W + 200 W (LCD screen) = 4200 W which is quite a big oven :-)

The cluster should be fully operational for beginning of 2010. (1 comments) [Less]


Two breathtaking ideas / 0.952 released

The first one is from nanobit who fortunately rejoined the project: to use a script engine (e.g the one of Thexa4 released with 0.9256) as basis for distributed stored procedures. Stored procedures are a well-known thing for people working with ... [More] databases, they are most of the time SQL procedures stored for convenience in what a database system can administer at best: a table with a column for the row number and another for the lines of the script. Using the distributed WAN database firstly introduced by nanobit in 0.943 joined with the stored procedure concept would allow another way of doing distributed computing with one major advance: it is possible for the user to distribute new code in almost real time.

The second one is to have sensors which broadcast collected data via GPU to the nodes. A special frontend collects data and filters it by sensor types (e.g earthquake sensors, temperature, insulation) and by streamers (the nodes which have a sensor attached and which broadcast its data). There is a prototype on CVS and we have two earthquake sensors, one in Holland and one in Switzerland. It will take a fairly amount of time and development effort until GPU people will be able to look at both signals and to compare them and average them. The averaging effect should in principle remove noise from the signal (like bus, or neighbour walking around).

0.9253 introduces an important bugfix in the milestoning mechanism of Eternity II plugin. Duncan White kindly reported it, and helped to test it. And we still are working on the duplicate packets issue which plagues the GPU network since 2004. Maybe this time we will find a solution, at least we identified once for all in source code where the problem is. Last but not least, this news is written with the 100$ laptop, a dream who is becoming true,see http://www.olpcnews.com

0.953 will be release at mid March. Keep an eye on the ChangeLog on CVS for the latest details.

0.952 released
--------------
0.952 is released on sourceforge.net. It contains important fixes to the connection layer done by Alex Fihman. This should be the most stable GPU ever.

Additionally, the command:

'add', whichplugin

will tell that basic.dll is the plugin that implements the function add. This will help users in deciding which plugins to unload, if some kind of jobs creates problems. (0 comments) [Less]


Read all GPU, a Global Processing Unit articles…

Edit RSS feeds.