TRIAL TODO

STATE MACHINE
* imposed HINTs are terribly broken, resent for the data in flight 
* check ACK/HAVE redundancy
* HAVE overuses find_filtered
* set priorities on ranges
* small-progress update problem (aka peer nap)
  guarantee size of updates < x% of data, on both ends
* pex is affected by peer nap
* how will tracker aggregate pexes?
* SWIFT_MSGTYPE_RCVD
* HAVE ALL / HAVE NONE
* aggregate ACKS (schedule for +x ms)
* channel close msg (hs 0)
* connection rotation / pex / pex_del
* misterious bug: Rdata (NONE)
* ?amend MAX_REORDER depending on rtt_dev
* Tdata repetitions bug

PERFORMANCE
* move to the.zett's binmaps
* optimize redundant HASH messages
* move to rolling HAVE queue
* 32 bit time field
* ?empty/full binmaps
* initiate RTT with prev RTT to host:port
* fractional cwnd

CACHING/FILES
* connection rotation
* file rotation
* real LRU/LFU
* file/hash-file re-open in read-only mode
* no cache recheck, failure-resistant
* completion mark
* unified events/callbacks
* move to 64-bit IO
* Transfer(fd) constructor
* think of sliding window(s)
* the ability to sniff file without downloading

MANIFOLD
* all-swarm performance stats
* run chained setups (cmd line protocol subsetting)
* implement: multiple swift instances per server
* run thousand-daemon caching tests (use httpgw)
* use a dedicated tracker
* add NATs to the setup
* recover mfold.libswift.org
* integrate Windowses

API
* pluggable storage

NAT
* NAT type detection => need peer identifiers (x100 amplification)

MFOLD
* integrate multi-peer changes by Jori
* do global swarm stats

OTHER
* httpgw or nginx?
* Sha1Hash constructor ambiguity
* don't #include .cpp
* think of using HTTP (?) as a fallback
* add header/footer, better abstract to the draft
* Gertjan: separate peer from channel? cng ctrl per peer ?
* packing hashes into a single datagram (tracking 1000s)
* partial channels / lightweight channels

THOUGHTS
* 6 degrees of sep = 3-hop TorrentSmell
* 60% immediately not connectable
* support traffic