can we beat them?? alias the need for speed!

[dangermaus]

first sorry for crossposting: the message is displayed in gnutella.com, too

we are a team of programmers that try to build up a supercomputer using the Gnutella network as transport layer.
Our project is hosted here:
http://sourceforge.net/projects/gpu

There is a whole team at Berkeley and Princenton university
that tries to build up a supercomputer over a peer to peer network.
This is quoted from the article "The group hopes its research can be used to distribute massive amounts of computing power for scientific applications in the form of "distributed supercomputers"".

http://www.vnunet.com/News/1133307

A strange thing is that they think to use another system for routing: packets travel through the network
using random walks! The heat equation as routing system :-#

and last but not least they will probably move all Seti@home users to their client:
http://setiathome.ssl.berkeley.edu/b...n=viewthread&a mp;num=1161

Do you think we can beat them??

[dangermaus]

version 0.66f released with transparency effects

[dangermaus]

oops. we are trying to get the maximum attention on the project...
hopefully the powers out there won't disagree.

[dangermaus]

only for peaceful, open and free research, as you can read from the project site :-) examples are the existing Search for Extraterrestrial Life and the Cancer research project. (In our docs that can be downloaded from http://sourceforge.net/projects/gpu both projects are presented)

if the project become succesfull an indipendent international gremium (e.g. the development team has people from about 10 different nations) will decide what plugins to sign. Signed plugins can then be installed to GPU.

a note on the random walk system for developers: it is very easy to implement it.

current Gnutella clients get a packet from one connection and
forward it to all other connections. The routing system is called
flooding and has lot of disadvantages.

future gnutella clients (if they get a packet from one connection)
they will choose randomly only one connection and forward it over that connection.

Packets will then travel randomly through the network and follow a pattern given from the heat equation (a partial differential
equation that can be often seen in diffusion processes). If you plot the density of packets in the network (where only one node sends out random walkers), you'll get a normal distribution.

Note that in the random walker system, the TTL constraint is needed.
However the routing table that keeps duplicate packets is not
necessary anymore. This was used to prevent packets travelling in circles.

I think if the team at Berkeley and Princenton say the new system is more efficient, we can believe it. To implement that change in clients is very easy; a small change to the routing method is enough (don't forward over all connections, but choose only one at random).

[dangermaus]

well, it can be, but I don't think this has to be a rule.

evil research can be performed by powerful governments, so they have enough money to buy a supercomputer. E.g. you can't build a nuclear weapon only with a supercomputer: you need skilled people (they are normally well paid), you need uranium, you need a whole nuclear plant to enrich the uranium. You need rocket scientists. You need rockets. You need a region where you can perform tests.


the Seti@home project on the contrary was started because the university was low at funds and couldn't buy another fast supercomputer. Distributing tasks is done by two older supercomputers. They can't compete with the fastest supercomputer on Earth, the nipponic Earth simulator.

Fortunately Seti@home computes at a speed of 40 Teraflops / s. eh, eh, the Earth simulator reaches only 35 teraflops / s.

[Paradog]

Well,
Its opensource which means that I could establish my own network for use of cracking encrypted passwords and so on.

I'm not sure...
I wouldnt like my cpu to be used for things like that.
Just my five bucks

[dangermaus]

I think even if all computers in Internet would install GPU,
we could crack something with about 80-bit key length,
but not more.
The old DES lies in this range. Triple DES perhaps at the limit. RSA with 1024 bit key length and Diffie-Helllmann are clearly outside, although RSA has some lacks that could be exploited (they use a 32-bit random generator and in principle GPU could generate all factors used in RSA).

Therefore a local network will not have enough power to crack a code.
However, what could happen is that cryptography could use a spread supercomputer to guaranteee more security.

I think the point is the following: a supercomputer over Internet is just a tool that can be used for doing well or doing bad. The choice is to the people who will use it.

Like a hammer: I can use it to build a nice house that will protect people from bad weather, or to smash it on the head of my neighbour.

[dangermaus]

ok.
soon or later I hope to get approval from the Swiss Federal Institute of Technology, so that we have about the same legitimation Seti@home has (University of Berkeley).