- * hints, data for non-existing ranges
- * opens multiple channels to the same address
+ * min_owd TINT_NEVER is logged
+
+ v hints, data for non-existing ranges
+ v opens multiple channels to the same address
v hints do not expire
- * survive 10% loss
- * unlimited ping pong
- * git
- * check hints agains ack_out?_
- * check data against ack_in
- * channel suspend/wake. 3 cong modes state machine - ???
+ v RTT calculations need improvement (test)
+ v google-log is unnecessary
+ * reduce template use (peer queue)
+ v hints do not expire
+ v survive 10% loss
+ v unlimited ping pong
+ v git sha-1
+ v check hints agains ack_out?_
+ v check data against ack_in
+ v channel suspend/wake. 3 cong modes state machine - ???
+ * release hints for a dormant channel
* minimize the number of template instantiations
- * Channel thinks how much it HINTs a second,
+ v Channel thinks how much it HINTs a second,
picker thinks which HINTs are snubbed
+ * files <1sec download : how HINTs are sent?
+ v dead Channels are not killed => cannot open a new one
+ (have a channel already)
+ v peers don't cooperate
+ * RecoverProgress fails sometime
+ v leecher can't see file is done already
+ v why leecher waits 1sec?
+ * hint queue buildup
+ * file operations are not 64-bit ready
+ http://mail.python.org/pipermail/patches/2000-June/000848.html
+ * recovery: last packet
+ v no-HINT sending to a dead peer
+ * what if rtt>1sec
+ v unHINTed repeated sending
+ v 1259859412.out#8,9 connection breaks, #8 rtt 1000, #9 hint -
+ mudachestvo, cwnd => send int 0.5sec
+ 0_11_10_075_698 #9 sendctrl may send 0 < 0.000000 & 1732919509_-49_-45_-200_-111 (rtt 59661)
+ 0_11_10_075_698 #9 +data (0,194)
+ 0_11_10_575_703 #9 sendctrl loss detected
+ 0_11_10_575_703 #9 Tdata (0,194)
+ 0_11_10_575_703 #9 sendctrl may send 0 < 0.000000 & 1732919509_-49_-44_-700_-110 (rtt 59661)
+ v complete peer reconnects 1259967418.out.gz
+ * underhinting causes repetition causes interarr underest causes underhinting
+ * misterious initiating handshake bursts
+ v whether sending is limited by cwnd or app
+ * actually: whether packets are ACKed faster than sent
+ * uproot DATA NONE: complicates and deceives
+ v r735 goes to github; r741
+ * receiver is swapping => strange behavior
+ v on high losses cwnd goes to silly fractions => slows down recovery
+ v code the pingpong<->keepalive<->slowstart transition
+ v empty datagram hammering (see at linode)
+ * make a testkit!!!
+ * never back from keepalive syndrome (because of underhashing)
+ * HTTP daemon, combined select() loop
+ * range requests, priorities
+ v LEDBAT
+ * CUBIC
+ v misterious mass packet losses (!data)