ConvertToCapped 29TB - "can't map file memory" when replication is off

I'm trying to convert an existing 4TB collection to capped 29TB. I
started on the replica set secondary, bringing it up without the
replSet argument because I'm not sure if the table lock might cause
replication problems if it's actively replicating (I did enlarge the
oplog to a size that should be able to handle the roughly 20 hours I
anticipate will be needed for the conversion). However, when I bring
up the (normally) SECONDARY without the replSet argument, and attempt
the convertToCapped, I get

db.runCommand({"convertToCapped": "log", size: 29000000000000});
{
        "assertion" : "can't map file memory",
        "assertionCode" : 10085,
        "errmsg" : "db assertion failure",
        "ok" : 0
}

 and in the logfile I see

couldn't open /data/db/lfs/lfs.1013 errno:24 Too many open files

I see the same error even if just looking at collection stats
(db.log.stats() ) . Another strange thing is that the VM size in top
is only 181M instead of the expected 8TB. This is 2.0.2 running on
Linux 64bit. I'm not sure why the absence of replication would cause
this. I verified that all datafiles and oplog files are owned by
mongod. I am starting the process with "numactl --interleave=all". The
system has 48G RAM.

I tried the conversion on the SECONDARY with replication running, but
I get "not master". I was hoping to run it on the SECONDARY so I could
get the replica set re-synced before running it on the PRIMARY.