Yesterday, I found out that when Mongo was printing:
Wed Aug 18 16:59:33 [initandlisten] waiting for connections on port 27103
Wed Aug 18 16:59:33 [initandlisten] ******
It might not be quite ready. In fact, It is still allocating its oplog and this can take an awful lot of time on slow disk/fs combinations (several minutes in my case).
I opened a bug at Mongo (http://jira.mongodb.org/browse/SERVER-1637) and Eliot pointed me to a fs configuration change that could help.
We have now opted for XFS local disks (instead of 8 RAID-1 EBS disks in Ext3). XFS is supported by the kernel of the instance we use while Ext4 is NOT.
