Investigate Magma 5X Slower than Couchstore performance

Raising this ticket as a tracker for the investigation started by CBSE. 

The current behaviour seen is when importing documents using pillowfight of size 1MB, magma performance is 5x slower than couchstore.

Backups and Restores were taken during pillowfight running to simulate a live cluster and the performance based on that.

The setup was:

3 Nodes:

1. Data
2. Data + Backup
3. Data

The nodes were of this size:

The test and the results can be found in this excel sheet:
https://docs.google.com/spreadsheets/d/1gd54QM64Rxr8GrxXqaeP1ZikaEVXmCapzobR-IfJzCg/edit?gid=346957412#gid=346957412

This sheet contains two sides: Left hand side is the test run by the client in the CBSE and the right hand side is my testing.

The pillowfight and backup commands can be found in the excel sheet as well.

The supportal link containing the logs for my test case:
https://supportal.couchbase.com/snapshot/0ee7c250a180bbd2625a5abfd20d489b::0