Hacker News Reader

Comment by maxdemarzi | Hacker News Reader

This is over a year old, and using ancient versions of software. It would be nice if they showed their testing code not just some horrific and unreadable gremlin queries. Taking 1.6 seconds to create a new node in neo4j and 2.5 to update it? Yeah, I don't think that's anywhere close to legit.

Never met a vendor benchmark that wasn't faulty in some way... http://maxdemarzi.com/2015/10/16/benchmarks-and-supercharger...

View on HN

To be fair, the SIGMOD'15 conference was in June 2015 so the paper is only ~6 mos old. While it is based on Gremlin 2 rather than Gremlin 3, most all the code in the paper is SQL CET statements -- there's very little Gremlin. In Table 8 on the last page they show their code for the Gremlin->SQL CET translations, and only a few tweaks are needed to update the step names in column one to Gremlin 3.

If the paper is only 6 months old, why would the authors use an ancient version of Neo4j?

Titan 0.4 [1] and Neo4j 1.9.4 [2] were the versions released in the Fall of 2013 so maybe that's when they did the work.

[1] https://github.com/thinkaurelius/titan/releases/tag/0.4.0

[2] https://github.com/neo4j/neo4j/releases/tag/1.9.4

The conference was in June 2015. The paper would have been submitted for peer review in Summer/Fall 2014. Which means that it could have been written in Winter/Spring 2014.

SIGMOD is not like a dev conference where you can change the paper the day before your talk.

Reply on news.ycombinator.com