Brainstorm

From GPU

Contents

Brainstorm Page


-initially created by Nanobit, feel free to add whatever you please.


Topic:

GPUSE distributed search engine

The initial search frontend was a little success, with a good relevant portion of the web crawled, o tonload of active visitors, and sometimes excellent search results (and sometimes poor..). One of the reasons of the poor results is, unfortunately, the amount of spam and xxx pages that do anything to show up in search records. In case of xxx not an issue if people search on it, but some 'webmasters' (do they deserve that name) abuse irrelevant search terms to promote their content.

Proposed fix: A system that not only indexes web pages, but also actively 'visit' them using the, for example, Mozzilla webbrowser component. After page is loaded, a screenshot is taken and stored. This process is done for at least 1 page per domain that shows up in the (top-100?) search results.

A website is constructed that asks (registered) visitors to moderate those screenshots. People can apply tags: normal, spam, technical, social, media, adult, etcetera. Those tags are used to filter out irrelevant results, in aid to the existing blacklist, and to improve overall search results. Of course, work can be done on our 'page rank' system as well, but humans can spot a 'spam' page most likely visiually within seconds.

When this database of human moderations grows, a new system can be added that does some kind of image recognition and/or (just) find (visually) simular pages. This AI system can aid page ranking and get improved with human feedback by above method.

PWanDB - creating virtual worlds

  • A script language that focusses screen rendering. An open-GL wrapper that can build 'primitives', data, images, and scripts, can be stored in the pwandb. Multiple users can create their own 3D content, much like secondlife does. This content is shared amongst all users, 'all' we need is a browser that allows an 'avator' to walk in this virtual space to have a distributed secondlife clone.
  • To fix overloaded/oversized databases, pwandb could work with realms so that a database holds only namespace of a particular subject, for example above 3D world. Users can subscribe which realms they want their pwandb to connect to.
  • Avator movement and inter-avatar communications (chat ;p) are distributed in real-time over the gpu network.

PWanDB - finding a script language

  • Good news on a script language. Mozilla has a perfect javascript engine, and my initial research learns that this may be perfectly suited for the pwandb script language. Features: fast expression parsing. User defined functions and objects. So, we can make and object 'gpu' that allows all interactions to a gpu host, for example, accessable from within javascript. Javascript is also a well-known language by many, with proper syntax, flexible variables (arrays etc), and well suited for 'small apps'. The user-functions plugin allows us to call other javascript pieces and use their output. To me, this like a dream come true, javascript been one of my favorite candidates for the pwandb script language. It will be entirely sandboxed by nature.

PWanDB - adding attributes to one variable name

Much like LDAP does, a pwandb variable could consist of multiple attributes. For example, to indicate mimetype, timestamps, etc. New attributes could be added freely, and a 'default' attribute defines the default value of this variable of no particular atrtibute is requested. This could help limiting the amount of variables needed. Drawback: a change in attribute would force nodes to exchange the entire variable content (at current design). Forhere, i stated that in my opininion it was sufficient to use the namespace itslef to define additional attributes when needed. Both approaches are valid, just keep the concept open and see if there is practical need.