spark-gremlin | implement implement the Tinkerpop Gremlin API

OLAP based Gremlin traversals will be much slower than standard OLTP traversals even for small datasets. There is considerable cost just in getting Spark primed up to process your traversal. That overhead alone might easily give your OLAP query a 1 minute handicap over OLTP. In the comments to your question you explained that your query is taking around six minutes. That does seem a bit on the long side but maybe in the realm of normal for OLAP depending on your environment??

Some graph will optimize for an OLAP count() and get you a pretty speedy result but you tagged this question with "JanusGraph" so I don't think that applies here.

You typically don't see the value of OLAP based traversals until you start concerning yourself with large scale graphs. Compare counting 100+ million edges in OLAP versus OLTP and you probably won't mind waiting six minutes for an answer at all (as OLTP might not finish at all).

It's hard to say what you might do to make your current setup faster as you are really just proving things work at this point. Now that you have a working model, I would suggest that the next step would be to generate a significantly larger graph (10 million vertices maybe) and try your count again with a decent sized spark cluster.