Other notes/ideas from 5/11/12 meeting:
Track counters by source (mongos or direct, user initiated, system – repl/shard-cmd --, localhost/ip, user)
Ind. cmd (types) by coll/db – like count, aggregate, mapReduce, drop/etc
Histograms/buckets by costs/latency and variance for things
Need nano/microsecs timing with reporting in fractions of ms/micros
Track extent stuff, freelist/etc., create/delete, per file
Data size w/without padding.
Maybe track possible faults per ns.
Track network related stats: create/destroy, socket exceptions, outgoing bytes, buffer bytes, sockets stats (ops, bytes, stack usage)
BSON* operations: iteration, append, realloc, copy
Update operations: ex $addToSet/$pull/$etc.
How to answer if fragment exists and compact is needed.