dce24eb1f0
libT: first cut at implementing the internal peers' request queues as pieces rather than blocks, as discussed with erdgeist and denis, to avoid a couple of nasty CPU bottlenecks.
2008-06-09 22:53:45 +00:00
7f8ef25acc
whoops, left some test scaffolding code in r6073
2008-06-07 21:28:47 +00:00
e3e3bee8b5
#800 initial support for GetRight-style fetching of data through http and ftp servers specified in the .torrent's "url-list" tag
2008-06-07 21:26:41 +00:00
041561f232
minor refactoring of tr_bitfield to (a) simplify the tests and (b) make things easier to read
2008-06-07 01:44:54 +00:00
2878d0d32e
bitfield speedups written by denis__. Thanks!
2008-06-05 20:21:56 +00:00
547056b50b
#988 : Optimistic unchoke can unchoke uninterested peers
2008-06-03 04:29:56 +00:00
cdf5cb87d6
another tweak for calculating tr_stat.desiredAvailable faster, suggested by BentMyWookie
2008-05-31 05:22:10 +00:00
e9c7e76c07
#980 optimistic unchoking is broken
2008-05-30 12:58:55 +00:00
70fcc56beb
(libT) (1) shave a few bytes from tr_peer and tr_ratecontrol's footprints. (2) remove some dead code.
2008-05-26 12:14:35 +00:00
2c79888a93
RPC/IPC redesign
2008-05-18 16:44:30 +00:00
7b25d0476c
remove occurrences of "if( 1 )"
2008-05-06 20:31:05 +00:00
86c3ae4fa4
#917 : Another minor speedup suggested by BentMyWookie. He enjoys giving new suggestions right *after* I hit "svn commit" :)
2008-05-06 19:49:25 +00:00
d047c1175d
#917 : small speedup to r5759 suggested by BentMyWookie
2008-05-06 19:39:16 +00:00
0f9ca34848
#917 : when seeding, drop peers that have everything you have (instead of just dropping seeds)
2008-05-06 19:06:46 +00:00
a362df0a14
#905 : update peer_id client lists. add a dozen or so new clients from Azureus (thanks amc1!)
2008-04-29 16:57:16 +00:00
47021f1a12
fold most of the "deep logging" functions into a single function
2008-04-25 02:57:33 +00:00
3445817a80
peer-msgs: faster upload speeds in situations with few peers. this patch needs wider testing for side-effects wrt speed limits.
2008-04-19 19:37:05 +00:00
08b27749ba
#890 : be smarter about noticing seeds from tracker & pex
2008-04-19 15:07:59 +00:00
ea9e0b260f
(1) fix choke/unchoke error. (2) if a peer keeps trying to reconnect to us, hang up during the handshake.
2008-04-17 03:48:56 +00:00
f12ac60d0a
update a couple of the tr_stat fields based on feedback from BMW
2008-04-11 23:05:14 +00:00
5890eaef34
encapsulate the blocklist a little better so that we can add more blocklists in the future if necessary
2008-04-01 19:20:21 +00:00
def9d1a492
better logging of blocked IP addresses.
2008-03-30 00:57:55 +00:00
a20bb8d569
get the blocklist code working & add a unit test for it.
2008-03-29 21:05:51 +00:00
c63b548195
fix bug that caused very large torrents to crash Transmission
2008-03-22 18:10:59 +00:00
5e3dcb5f6f
#751 : more digging to try to figure out where the problem is.
2008-03-06 13:24:44 +00:00
e00d8bf849
fix some gcc 4.2 warnings
2008-02-28 19:33:11 +00:00
f2f24e936d
tear out the "fast peer" code from refillPulse() altogether
2008-02-25 01:35:20 +00:00
b9a9e1f1a8
disable some fastpeer code that's too expensive.
2008-02-25 01:29:30 +00:00
cfe068876e
remove unnecessary #includes
2008-02-19 18:39:49 +00:00
8d14a789cb
#698 : `have' gets too far ahead of `verified'
2008-02-10 04:03:19 +00:00
f72ba4b8d8
#663 : connection limits don't work correctly.
2008-01-28 21:05:50 +00:00
b1cd6ecccc
remove unused "max unchoked peers" feature. fix obsolete TR_FLAG_SAVE comment.
2008-01-23 17:06:36 +00:00
1b7706e101
finish the trunk/1.0x libT merge: router death & portability
2008-01-21 02:11:57 +00:00
de21d77af2
test code to test a "router death" theory in the nightlies
2008-01-20 03:21:51 +00:00
3a47dfdb23
(router death) dial down the max number of new sockets per second even lower
2008-01-19 03:44:44 +00:00
de515ee339
if the torrent's download path doesn't exist, don't create it -- it's might be a removeable disk that got unplugged.
2008-01-18 19:13:32 +00:00
52c3243b4b
more test code to address "router death". Based on maxintech's feedback in the forums, try making the peer manager's reconnect intervals and volume more like 1.00
2008-01-17 03:43:08 +00:00
377c00a826
a test: if we set the reconnect period to any given peer to > 900 seconds, will that make "router death" go away? 900 seconds is the usual router setting for timing out bad sockets.
2008-01-16 20:24:18 +00:00
61fc032dc1
sync peersSendingToUs with isDownloadingFrom, reported by BentMyWookie
2008-01-16 17:52:47 +00:00
a1e20dd7e9
more work on "router death": if we can't connect to a given peer, retry it fewer times and wait much longer between retries.
2008-01-16 15:13:02 +00:00
ca70ef18bb
drive the throttle even lower.. max of 16 connections per second.
2008-01-15 06:02:47 +00:00
2c7a7c4253
throttle the number of new connections per second down even further from 48 to 32 because of the continuing "kills my router" reports
2008-01-13 21:05:38 +00:00
90f5c61327
tweak the conncetions a little bit
2008-01-11 19:44:45 +00:00
876cd4c413
test patch for the "router death" bug: throttle the number of connections made per second
2008-01-11 18:13:09 +00:00
39b1db460d
follow the spec's suggestion that new connections be 3x as likely for optimistic unchoke
2008-01-11 02:09:20 +00:00
6f6adb1d53
try to ramp up fast connections more quickly.
2008-01-11 00:38:26 +00:00
3a6396a3cb
mingw portability fixes
2008-01-10 18:52:46 +00:00
bedfab2ccc
clarify peer upload and download
2008-01-10 00:52:02 +00:00
591f7f8b10
* improve the unchoking algorithm.
...
* when deciding which peers to connect to, take peer's previous speeds into account
* longer delay interval before reconnecting to peers that didn't give us data
* added uTorrent-inspired "flags" column in tr_peer_stat and gtk's torrent details' "peer" tab
2008-01-09 17:33:43 +00:00
4f0f6e150e
fix another OpenBSD alignment bug reported by naddy
2008-01-08 01:57:48 +00:00
35f73c097c
lazy "allowed set" generation. add regression test for "allowed set" generation.
2008-01-06 21:56:30 +00:00
6c3eab0a92
#585 : 1.00 could be sped up by elminating reconnect intervals for some peers
2008-01-05 18:17:56 +00:00
893b0c10a3
to try to prevent router overloads, make (MAX_RECONNECTIONS_PER_PULSE * RECONNECT_PERIOD_MSEC) smaller.
2008-01-03 05:30:45 +00:00
ce6c33b7bf
set copyright info to 2008
2008-01-01 17:20:20 +00:00
71efb5c74b
add per-priority-level endgame mode
2007-12-26 07:25:50 +00:00
819b83d90d
fix another file priority bug.
2007-12-25 06:37:21 +00:00
f8bc11e094
fix a couple of related file priority bugs.
2007-12-25 05:42:33 +00:00
473907d891
a little housekeeping: move tr_torrent stuff into its own header
2007-12-25 05:37:32 +00:00
401cfbdb6b
replace the per-torrent PEX setting with a global one, as per discussion with SoftwareElves and BentMyWookie
2007-12-24 05:03:40 +00:00
5c4f841c7e
added tr_torrentSetPeerLimits(), tr_torrentGetPeerLimits(), tr_setGlobalPeerLimit(), tr_getGlobalPeerLimit()
2007-12-20 21:44:16 +00:00
10b6e8ff8f
lessons learned from 0.82:
...
* if we can't bind a socket to a peer, take that peer out of the retry pool.
* also, try to connect with more peers up-front.
2007-12-15 04:26:31 +00:00
332019593d
speed tweaks based on whitty's feedback
2007-11-29 02:56:31 +00:00
a720545256
fix peer-mgr.c shear between 0.9x and trunk
2007-11-25 17:07:12 +00:00
d735989aec
fix clientIsSnubbedBy() to work correctly when we're a seed
2007-11-25 16:57:08 +00:00
b0f5ff96f3
back to the "router death" issue: throttle how many connections T will try to open at any one time
2007-11-18 17:35:28 +00:00
ed72b1b795
fix choking bug that could penalize good peers.
2007-11-18 06:15:13 +00:00
e371009606
this should improve download speeds. test it!
2007-11-18 03:18:26 +00:00
9a81cd7637
reintroduce the "SWIFT" algorithm
2007-11-18 01:00:49 +00:00
37969406b2
Add "Status" column to tr_peer_stat.
2007-11-17 23:43:33 +00:00
257eaee631
more of the same
2007-11-15 05:47:23 +00:00
ea3a1387f2
more of the same.
2007-11-15 05:20:38 +00:00
8019f8cd2c
raise the number of unchoked peers per torrent
2007-11-15 05:04:21 +00:00
a14fe99e00
another download speed tweak
2007-11-15 04:59:37 +00:00
c4f15263a0
since many people seem to be having a hard time holding onto good peers, be a little more lenient on how quickly we hang up on peers that have transferred piece data with us.
2007-11-14 05:02:03 +00:00
a9e5fff2e6
* fix the `req->length == (uint32_t)( ((block)==((msgs->torrent)->blockCount-1))' bug.
...
* there seems to be a pattern for peers that were (intentionally?) giving incomplete data to trigger the bug above. when a peer does this, give them a strike on its three-strikes-and-you're-banned count
2007-11-13 05:36:43 +00:00
8516860fec
fix a bug in tr_stat's availability numbers, reported by BentMyWookie
2007-11-11 16:33:04 +00:00
adeee3d886
silence some Fast Peer console messages
2007-11-10 22:29:55 +00:00
c32f893038
commit tiennou's fast-peers-final patch from ticket 149
2007-11-10 16:06:00 +00:00
fd430fde90
undoing the r3773-r3774 experiment.
2007-11-09 20:07:52 +00:00
67b719c2de
on Darwin, use NSCParameterAssert() instead of assert().
2007-11-09 19:50:40 +00:00
e168cf3f69
don't let incoming peer connections sneak past MAX_CONNECTED_PEERS_PER_TORRENT.
2007-11-09 15:19:12 +00:00
0d819bd67e
more work on the "router death" issue.
2007-11-08 19:43:48 +00:00
bc3eff899f
* some work on lessening the way we hammer the router.
...
* move strlcpy and strlcat out of their own files and into utils.c
2007-11-08 04:11:09 +00:00
71f4efcbad
update our #includes now that libevent has cleaned up event.h
2007-11-07 18:26:19 +00:00
a6206b036c
"corruption" fix part 2: lots of assertions, and try to complete pieces & ban bad peers sooner
2007-10-31 04:23:51 +00:00
4cde721475
fix oops
2007-10-30 20:17:49 +00:00
aace9469bb
fix compile warning
2007-10-30 20:11:23 +00:00
caa4170271
#426 0.91rc1: if pex is disabled on a stopped torrent, pressing "resume" causes pex to be re-enabled. (reported by pea)
2007-10-28 19:42:47 +00:00
92ac49d9c4
compatability with older C copmilers (ticket #422 , patch by fizz)
2007-10-27 21:29:41 +00:00
8c0e2434f5
fix OpenBSD build error.
2007-10-25 13:59:46 +00:00
52261d88fd
Fix 0.90 socket connection leak.
2007-10-25 13:38:34 +00:00
a5946e396c
tweak the logging a bit.
2007-10-22 18:52:36 +00:00
814a56046d
fix ul/dl speed wart reported by BentMyWookie
2007-10-21 15:47:26 +00:00
25fa56d294
* don't fire the "status changed" event if the status changes while a torrent's being started for the first time.
...
* cleaner setting of the DND flags in the gtk+ client.
* cpStatus cleanup.
2007-10-19 23:23:21 +00:00
00a425a7d3
further tweaks to the peer connection weights after reading the overnight runlog
2007-10-16 15:05:05 +00:00
fa200efd09
more tweaks
2007-10-16 03:14:07 +00:00
c9359d8bde
There are only 16 days left for something to replace this as October's stupidest bug.
2007-10-16 03:00:53 +00:00
ec30180d76
tweak the weights that decide to reconnect to crappy peers less often.
2007-10-16 02:37:34 +00:00
30418e34b8
fix crash reported by John_Clay in http://pastebin.ca/738133
2007-10-16 02:16:57 +00:00
8feeb9ee6b
when deciding which peers to try to connect to, add some weights to punish peers that don't transfer piece data with us. We'll still give them another, but at longer intervals than better peers get.
2007-10-16 00:55:17 +00:00
6eb7edf007
fix choke fibrillation bug in new connections
2007-10-15 16:01:42 +00:00
57eaa27970
Fix bug that could initiate pex in private torrents.
2007-10-14 17:21:24 +00:00
3e878aeba9
tweak the debugging messages a bit.
2007-10-13 14:02:10 +00:00
9211c57ad2
merge in the parts of peerutils' peer pruning code that still make sense
2007-10-13 13:54:05 +00:00
7e0faec4db
more debugging messages
2007-10-11 03:12:48 +00:00
ef4ec75532
Try out some of SoftwareElves' suggestions for faster startup connections
2007-10-11 02:50:05 +00:00
23d2f77778
(1) maybe fix the "peer->msgs != NULL" bug reported by Dhruvk and John_Clay (2) remove dead code
2007-10-11 00:09:58 +00:00
cbaf1b8faa
commit more of tiennou's fastpeers patch
2007-10-10 16:39:12 +00:00
76aed9200d
fix a subtle bug with having two connections to the same peer.
2007-10-10 15:59:59 +00:00
a9849660f6
* fix bug that incorrectly marked some incoming connections for termination
...
* strip unnecessary linefeeds out of some debug messages
2007-10-10 02:59:30 +00:00
602ee120af
fix recent tracker error reported by John_Clay that made it slower to pick up peers
2007-10-09 04:50:10 +00:00
9a79d1e603
the purge flag should take priority over all the other reasons to keep or purge a peer. :p
2007-10-08 01:53:11 +00:00
e1b447304b
fix atom initialization issue with r3320
2007-10-08 01:37:20 +00:00
161ff6a7a2
ban peers that feed us too many bad blocks.
2007-10-08 01:31:27 +00:00
5d8f9d8b13
try to connect to peers faster when a torrent is initially started.
2007-10-07 04:14:58 +00:00
c75f49bc26
* add a per-peer request queue to hold the next 10-15 seconds' worth of requests so that we always have more requests at hand when the current requests start to run low.
...
* increase the tracker `numwant' variable to grow our peer pool
* bugfixes in cancelling requests.
* make the debug log sexy and readable like uTorrent's ;)
2007-10-06 18:20:52 +00:00
b6483cbf3f
don't allow duplicate requests to the same peer, even during endgame. (alus)
2007-10-05 00:16:47 +00:00
80979e0240
a little code simplification -- remove unnecessary branches
2007-10-04 20:49:37 +00:00
0872b4f40a
fix the crash reported by Waldorf and John_Clay
2007-10-04 20:31:19 +00:00
5f2bd1d950
speed improvements... this is getting me 700 KiB/s (which is near to my speed cap) in the ubuntu torrent test
2007-10-03 19:00:51 +00:00
877e2bb86e
some more peer tweaks
2007-10-03 18:19:51 +00:00
a3d1a254d1
minor source code cleanup...
2007-10-03 16:42:43 +00:00
3d2e3cfdfb
add pass subscription info into the peer-msgs constructor so that we can start getting information immediately
2007-10-03 04:04:34 +00:00
0f550043fd
fix peer-mgr.c:1092 asset failure reported by Waldorf_
2007-10-03 02:46:15 +00:00
1f1605986c
tear out some now-exterraneous code.
2007-10-02 20:55:14 +00:00
f4988a3f6e
tweak the peer manager code a little more.
2007-10-02 19:54:14 +00:00
3dbaa7d673
make the peer manager a little greedier w.r.t. keeping connections alive.
2007-10-02 19:25:18 +00:00
cdabb1c0b1
clean up the shutdown code a little bit.
2007-10-02 16:12:44 +00:00
c271cec6a3
work around portability glitch in libevent's headers
2007-10-02 14:58:39 +00:00
f237bf7047
* simplify the the choking intervals
...
* fix peer-mgr destructor bug
* send debugging messagese to TR_DEBUG_FD instead of the console
2007-10-02 14:35:02 +00:00
77d7cf76a1
fix looong-standing potential dangling pointer bug reported by BMW
2007-10-02 03:33:17 +00:00
af686b3297
roll back a bug introduced in r3265
2007-10-02 02:01:57 +00:00
84ab0092fb
some cleanup... remove some dead/obsolete code
2007-10-02 00:05:40 +00:00
b84305cdf3
more agressive purging of connections that give socket errors
2007-10-01 18:09:59 +00:00
4ebe0b563a
throttle back how many concurrent connections we make.
2007-10-01 17:51:07 +00:00
f0562f5947
try to disconnect all the peers when the user hits `pause'
2007-10-01 16:50:51 +00:00
75ff12faba
don't log BT protocol messages in UL/DL speed -- only log piece data transfers.
2007-10-01 16:31:17 +00:00
e4fec9eefd
add tr_stat.peersKnown to indicate how many peers exist in the peer `pool'
2007-10-01 15:51:54 +00:00
cba61802cf
remove newly-unused field tr_peer_stat::isConnected
2007-10-01 15:41:42 +00:00
1a012b8b87
maybe fix the where-are-the-peers-coming-from bug reported by SoftwareElves, John_Clay, and jahpraiseherb
2007-10-01 15:36:31 +00:00
5c11c58113
simplify libT locks now that it's (more-or-less) single-threaded. fix deadlocks. make tr_locks nestable.
2007-10-01 15:17:15 +00:00
f7e3f85e50
follow BT `best practices' more closely w.r.t. choosing which pieces to download
2007-10-01 14:24:22 +00:00
9a3c91cad2
ensure TR_PEER_FROM_INCOMING peers get peer atoms. this should fix BentMyWookie's crash.
2007-10-01 05:32:34 +00:00
56738f065c
fix some memory issues.
2007-10-01 04:12:24 +00:00
b99084e638
more experimenting with how to get good peers
2007-10-01 03:24:52 +00:00
6bef7def52
tweaks
2007-10-01 00:08:12 +00:00
7cc6771318
experimental better peer management.
2007-09-30 23:55:49 +00:00
8a40a15b16
more work on the peer-mgr locks
2007-09-29 14:19:23 +00:00
64239a21e1
prevent a deadlock via the handshake code
2007-09-29 13:47:15 +00:00
0812087a4e
add mutex locks for thread safety.
2007-09-29 06:37:03 +00:00
Charles Kerr