Abstract: The Paderborn University BSP Web Computing (PUB-Web) Library (formerly known as PUBWCL) is a system for executing parallel algorithms according to the Bulk-Synchronous Parallel (BSP) model. Its main difference from the Paderborn University BSP Library (PUB) is that it is not designed for the use on parallel computers but on PCs and workstations distributed all over the whole of the internet.
A characteristic of PUB-Web is, in contrast to other well-known web-computing projects like e.g. SETI@home or distributed.net, that the clients do not request small independend subproblems to solve from a central server but do work together on the problem by communicating with each other and synchronizing themselves according to the BSP model.
Recently, there have been many approaches to use the idle-times on the many computers distributed all over the whole world for computation-intensive tasks; all these idle-times together represent a really gigantic computation power. These approaches have been realized by providing so-called clients which computer owners or administrators can download and install in order to donate their unused computation power to large-scale projects on a voluntary basis.
Among the well-known projects of that kind, there are e.g. the following:
- distributed.net
- Great Internet Mersenne Prime Search (GIMPS)
- Search for Extraterrestrial Intelligence (SETI@home)
- Folding@home Distributed Computing
All these projects have in common that the problem to work on has to be split into many small subproblems by the central server; the clients (they are often designed as a screen saver doing the work in the background) download a subproblem, solve it, send the result back to the server, and continue with the next subproblem.
With these approaches, computation actually takes place in parallel, but only independent subproblems can be solved because there is no communication between the clients.
PUB-Web removes this restriction: the idea is to execute massively parrallel algorithms on computers distributed all over the internet, communicating with each other and synchronizing themselves according to the Bulk-Synchronous Parallel (BSP) model.
Like the participants of the other projects mentioned above, all the participants of PUB-Web have to download and install a client. In contrast to these other projects which only need a connection to the internet when uploading the results or downloading new subproblems, PUB-Web requires a permanent internet connection due to the continuous communication; beside a direct connection to the internet, also dial-up connections are an option because PUB-Web tolerates short network errors (e.g. interruption due to a reconnect) and even IP address changes.
Its main difference from the Paderborn University BSP Library (PUB) is the target group (for example, PUB is optimized for parallel computers and clusters) and the programming language (Java instead of C) as well as the consequences resulting from this fact: on the one hand there is be a drawback in execution speed in comparison to PUB due to the use of Java and the latency periods of the internet; but on the other hand Java leads to more security (apropos Java Sandbox) and many simplifications with regard to platform independend, object oriented, and network pogramming; last but not least, one can develope more failsafe applications with lower costs.