solution
KangurOS Solution
The graph, which I have in mind has directed edges which always connect three vertices. In such a connection, each vertex corresponds to either a subject, the predicate, or the object (like in the semantic web). In another relation, this vertex could also have a different role. This means we have a universe U of objects and a relation R={(x,y,z)|x,y,z el. of U} We decided to use this:
w
P
|
|
x S---------O z
|
|
W
y
S: subject side
P: predicate side
O: object side
W: way id side
w,x,y,z are vertices, containing a file
The following is obsolet:
x S---------O z
|
|
P
y
S: subject side
P: predicate side
O: object side
x,y,z are vertices, containing a file
In a first implementation, a vertex will be stored in four files:
the file itself with its content
a tuple-list with links to the pairs of predicates and objects, where this vertex is subject
a tuple-list with links to the pairs of subjects and objects, where this vertex is predicate
a tuple-list with links to the pairs of subjects and predicates, where this vertex is object
In a database, we would have the following two tables: Objects-database
id: ID |
value: BLOB |
id is indexed, value not (id might be primary key) Link database
subject:ID |
predicate:ID |
object:ID |
All are indexed. The type ID will need to have to be something which could uniquely address all possible objects we could have. Let's for a beginning assume, that id is a 256 bit integer. Then we could even give random numbers to it and we would most likely (at least at is way more likely that our computer is destroyed by a meteor) a double hit, if we go up to 1000 Peta = 10**60 (the hole internetarchive is 2 Petabyte nowadays) objects. if an Object would need at least 256 bit + data to store, this would mean 32000 Petabyte storage. Ideas about clustering objects: To accelerate access time it would make sense, if data which belongs together is near to each other. This gets especially interesting if data is on different servers. Graph visualizers:
http://code.google.com/p/jrfonseca/wiki/XDot - interactive visualization engine using the pydot/graphviz (http://code.google.com/p/pydot/) format
http://ubietylab.net/ubigraph/index.html - not open/free (only as in free beer), no interaction with nodes
http://graphexploration.cond.org/ - nice project based on jython, but no labels
Pure 3D engines for python: