From 6d4845149c9c2f96d038ba41fcad6fedd378ae4c Mon Sep 17 00:00:00 2001
From: erdgeist <>
Date: Wed, 10 Dec 2008 14:19:37 +0000
Subject: Propose a scrape sync protocol.

---
 ot_livesync.h | 72 ++++++++++++++++++++++++++++++++++++++++++++++-------------
 1 file changed, 56 insertions(+), 16 deletions(-)

diff --git a/ot_livesync.h b/ot_livesync.h
index 153dbac..97fe803 100644
--- a/ot_livesync.h
+++ b/ot_livesync.h
@@ -10,32 +10,72 @@
 #include "trackerlogic.h"
 
 /*
-  Syncing is done as udp packets in the multicast domain 224.0.42.N port 9696
+  Syncing is done as udp packets in the multicast domain 224.0.42.5 port 9696
 
   Each tracker should join the multicast group and send its live sync packets
   to that group, using a ttl of 1
 
-  Format of a live sync packet is straight forward and depends on N:
+  Format of all sync packets is straight forward, packet type determines
+  which kind of packet this is:
 
-  For N == 23: (simple tracker2tracker sync)
     0x0000 0x04 id of tracker instance
-  [ 0x0004 0x14 info_hash
-    0x0018 0x04 peer's ipv4 address
-    0x001c 0x02 peer's port
-    0x0020 0x02 peer flags v1 ( SEEDING = 0x80, COMPLETE = 0x40, STOPPED = 0x20 )
+    0x0004 0x04 packet type
+
+                             ########
+ ######## PEER SYNC PROTOCOL ########
+ ########
+
+  Each tracker instance accumulates announce requests until its buffer is
+  full or a timeout is reached. Then it broadcasts its live sync packer:
+
+  packet type SYNC_LIVE
+  [ 0x0008 0x14 info_hash
+    0x001c 0x04 peer's ipv4 address
+    0x0020 0x02 peer's port
+    0x0024 0x02 peer flags v1 ( SEEDING = 0x80, COMPLETE = 0x40, STOPPED = 0x20 )
   ]*
 
-  For N == 24: (aggregator syncs)
-    0x0000 0x04 id of tracker instance
-  [ 0x0004 0x14 info_hash
-    0x0018 0x01 number of peers
-    [ 0x0019 0x04 peer's ipv4 address
-      0x001a 0x02 peer's port
-      0x0021 0x02 peer flags v1 ( SEEDING = 0x80, COMPLETE = 0x40, STOPPED = 0x20 )
-    ]+
+                               ########
+ ######## SCRAPE SYNC PROTOCOL ########
+ ########
+ 
+  Each tracker instance SHOULD broadcast a beacon once in every 5 minutes after
+  running at least 30 minutes:
+
+  packet type SYNC_SCRAPE_BEACON
+  [ 0x0008 0x08 amount of torrents served
+  ]
+
+  If a tracker instance receives a beacon from another instance that has more than
+  twice its torrent count, it asks for a scrape. It must wait for at least 5 + 1
+  minutes in order to inspect beacons from all tracker instances and chose the one
+  with most torrents.
+
+  If it sees a SYNC_SCRAPE_TELL within that time frame, it's likely, that another
+  scrape sync is going on. So one tracker instance MUST NOT react to beacons within
+  5 minutes of last seeing a SYNC_SCRAPE_TELL packet. After a scrape sync all
+  tracker instances have updated their torrents, so an instance in a "want inquire"
+  state should wait for the next round of beacons to chose the tracker with most
+  data again.
+
+  packet type SYNC_SCRAPE_INQUIRE
+  [ 0x0008 0x04 id of tracker instance to inquire
+  ]
+
+  The inquired tracker instance answers with as many scrape tell packets it needs
+  to deliver stats about all its torrents
+ 
+  packet type SYNC_SCRAPE_TELL
+  [ 0x0008 0x14 info_hash
+    0x001c 0x04 base offset (i.e. when was it last announced, in minutes)
+    0x0020 0x08 downloaded count
   ]*
 
- */
+  Each tracker instance that receives a scrape tell, looks up each torrent and
+  compares downloaded count with its own counter. It can send out its own scrape
+  tell packets, if it knows more.
+
+*/
 
 #ifdef WANT_SYNC_LIVE
 
-- 
cgit v1.2.3