- From: Sampo Syreeni <decoy@iki.fi>
- Date: Wed, 13 May 2015 22:02:30 +0300 (EEST)
- To: ahogan@dcc.uchile.cl
- cc: semantic-web@w3.org
On 2015-05-13, ahogan@dcc.uchile.cl wrote: > One drawback of the current approach is that at the moment you have to > load the RDF graph into memory, which may be prohibitive for very very > large graphs (of connected blank nodes), but I'm not sure how often > that happens in practice. (It wasn't a problem for any of the graphs > in the BTC-14 dataset for example.) And, once you factor in the presence of named nodes whose names do encode a whole lot of connectivity information as well, in theory you ought to be able to add in buffer tree style thinking over the lexical sort of the named nodes, combined with rough, incremental topological sorting of the blank nodes hanging off the named ones. That way a little bit of thought ought to yield you an algorithm which is both cache local and distributable enough to handle practical graphs with sizes into the current big data territory. > https://github.com/aidhog/blabel/wiki Nice stuff. -- Sampo Syreeni, aka decoy - decoy@iki.fi, http://decoy.iki.fi/front +358-40-3255353, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2
Received on Wednesday, 13 May 2015 19:02:59 UTC