forked from libswift/libswift
-
Notifications
You must be signed in to change notification settings - Fork 0
/
BUGS
65 lines (62 loc) · 2.72 KB
/
BUGS
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
* 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
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
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)
// Ric:
* check why the last HAVE msgs r not sent
* data is sent even if the client has stopped
* when initialized the piece picker might select hints from hint_out without knowing if the peer actually has it!
* IMPORTANT: trace(bin, range) reports bins out of range! Check if bug!!