I have a collection with 75 million objects (375 bytes avg size) and one index, _id.

mongodump of this collection takes hours.

Some random info that may help:

• WiredTiger, mongo 3.04
• System has 16gb of ram, and is using provisioned I/O on amazon EC2
• Collection uses a custom _id - a string of 22bytes
• _id index is ~2gb in size (fits into ram)
• CPU is very high (one core appears to be pinned) during mongodump.
• High read disk i/o relative to write.
• There is no other database activity; I've taken this secondary offline to test this problem out.

Interestingly, when I do --forceTableScan, it completes in less than 15 minutes.

Here's a representative log entry from the slow mongodump:

2015-07-24T14:02:17.602+0000 I QUERY [conn5256] getmore LPRS.LPRSObjects cursorid:32178252446 ntoreturn:0 keyUpdates:0 writeConflicts:0 numYields:274 nreturned:12070 reslen:4194601 locks:{ Global: { acquireCount:

}, Database: { acquireCount:

}, Collection: { acquireCount:

} } 5550ms