Hacker News Reader

Comment by nostrademons | Hacker News Reader

It looks like a Googler's 20% project.

However, I'll point out that most of the actually-used open-source projects coming out of Google aren't actually "Google projects", they are "projects by Googlers released under the OSPO process". LevelDB (Jeff Dean & Sanjay Ghemawat), Protocol Buffers (Kenton Varda), Guice (Bob Lee, Jesse Wilson, and Kevin Bourillion), Gumbo (myself), and angular.js (a team within DoubleClick) all started out as small internal projects built to scratch an individual's itch that were then released externally because hey, why not. The "corporate" open-source projects have been things like Android, Chrome, GWT, Closure, Polymer, etc.

View on HN

Exactly this. And I'm the Googler. Hi!

Hi. Former googler who does work that Graph Databases would be useful for. If you don't mind I'd like to ask:

Does it have bulk import and if so what is it's speed for bulk import rougly speaking?

It does have bulk import; I've been loading largish subsets of the Freebase dumps.

Load speed is pretty good (into persistent storage, I assume) and can be improved with some of the database parameters. A rough estimate is that a million triples or so takes about 5 minutes, but that slows down as it gets bigger. 134m triples took me 6-8hrs, so I slept on it.

Since no one's mentioned it yet, another alternative to Neo4J is AllegroGraph (though you need to pay for it; the free version supports 5 million tuples).

It _does_ support bulk loading, with over 500K triples per second. According to http://franz.com/agraph/allegrograph/agraph_benchmarks.lhtml, given enough RAM, it can load over a billion tuples in just over half an hour.

Virtuoso 7.1. and OWLIM 5.5 have similar loading speeds. In this case the decompression algorithm is often the bottleneck needing multiple files to be read in parallel to go faster. Oracle 12c Semantic Network and Yarcdata uRiKA can also go faster in loading if correctly set up.

What kind of usage is this seeing inside Google?

Reply on news.ycombinator.com