weird splits in sharded cluster

[Sean Fannan]

We are experiencing an odd issue with a new sharded cluster we are setting up. There are 5 shards, setup before any data is loaded in.  The shard key is the _id and the format of the key is:

XXXXXXX_MongoID where XXXXXXX is a random number.

The weird issue we are having is that when we insert a bunch of documents into the shard and splits happen, the splits do not seem to follow the chunk size (which has not been altered and should be 64MB).  So we will have some chunks that have a big range and others with a very very small range.  So for example:

An example of a big range:

min: 1887300_50a6e0309c5ddd4c8fc30584

max: 1906300_50a6cc769c5ddd49b451e6db

items in this chunk:  254199

An example of a small range:

min: 4077100_50a6deaf9c5ddd4cd9de1f32

max: 4077100_50a6e01c9c5ddd4ba67bbc16

items in this chunk: 109

Then, if you examine the items in the small range example, there are no items that are abnormally huge or anything that would make them 64MB.  So it really feels like the mechanism to split chunks is running away every once in a while and is splitting them more than they should.  The problem that results from this is that because the chunks are not the same size, even when the number of chunks are balanced between the shards, the amount of data between each shard is completely unbalanced.

We are running 2.2.1 on every box in the system, mongod, config servers and mongos servers.

Any ideas what could be at play here that I'm missing?