Serialization performance

[Chat-Wane]

When running passage-server, querying ?s ?p ?o, half the time is spend on serializing in JSON format (writeRow):

Few tracks for improving execution time:

• Change the serializer by Blazegraph's, might be faster, might be not. It might at least remove the need for converting into Jena's Node (i.e., parse and getString in the middle of the flamegraph);
• Implement a thread dedicated to serialization.

[Chat-Wane]

Test on ?s ?p ?o timeout 60:

• Jena's serializer ~3.9M elements
• Blazegraph's serializer ~3.9M elements

Course of action:

• Parallel still,
• Update blazegraph's serializer from 2.17 -> rdf4j-queryresultio-sparqljson 5.0.2. This requires to update all interfaces to rdf4j ones.

[Chat-Wane]

On parallelization: the serializer creates a child thread that is dedicated to hasnext + next + converting to Node.
On ?s ?p ?o, a 2M blockingqueue, 60s timeout, we get 4.7M elements.

Considering that both threads are full of productive computation (i.e. not waiting (put is small and take does not show)), it would be difficult to get better performance without further splitting.