I have a mongodump that contains a collection with 50 million documents and metadata for 16 compound indexes which are made up of between 4 and 9 fields each. It has been found that doing this full mongorestore with indexes takes roughly 6 hours with no visible resource contention on the server during this period.

However, if I create an empty collection and build the 16 indexes manually on the empty collection and then mongorestore the data only using the –noIndexRestore flag this takes roughly 2 to 3 hours.

I know there are issues relating to index compactness/efficiency from doing the data restore against an empty collection with indexes already built. However, the difference in time taken to do a restore in this way is so much less than using mongorestore that it should merit a new flag in mongorestore which gives users the option to create the empty collection and the build the indexes first.