Titan-Cassadra: backend exception when searching for node

[kuzeko]

I'm searching for nodes in a graph with ~21M nodes (single machine, 24 cores, 120GB RAM), the query is the following

 g.V().has(propName, propValue)

there are no indexes involved.

I get the following exception

 javax.script.ScriptException: com.thinkaurelius.titan.core.TitanException: 
       Could not execute operation due to backend exception

but with no stack trace or any other info.
I do not find anything in the Cassandra log either.

How do I investigate this?
Should I tune some configuration parameter?

[kuzeko]

I've found that the error arises 99% of the times due to problems with the garbage collection.
Apparently cassandra has problem in handling timeouts derived from GC pauses, and queries like this have generates some issues.

The best way to solve this is to tune the GC parameters when running with cassandra in embedded mode.

Here what worked for me (pay attention to the -Xms option as well!):

-Xms20G -Xmx120G -XX:+UseParNewGC -XX:NewSize=2G -XX:+UseConcMarkSweepGC -XX:+CMSParallelRemarkEnabled \
-XX:SurvivorRatio=8 -XX:MaxTenuringThreshold=1 -XX:CMSInitiatingOccupancyFraction=75 \
-XX:+UseCMSInitiatingOccupancyOnly -XX:+DisableExplicitGC

I think that either better handling is supported for cassandra embedded, or t would be best to remove support to it.

Despite what written on the properties file, the docs publicly advertise this setup!