CCID opens in OPEN. CLOSING timer shorter.

dccp/NOTES

@@ -0,0 +1,6 @@
+
+* We don't start CCID in PARTOPEN (rather wait until OPEN), because
+in PARTOPEN the client still needs to send (Ack,Sync) packet pairs,
+but the initial CCID rate is 1 packet per second, which slows down
+the handshake to OPEN, since the second packet of the (Ack,Sync)
+sequence has to wait for a whole second.

dccp/TESTING

@@ -1,10 +1,5 @@
 
-* after 1sec, CCID has not been open on server
-
-* why does packet exchange stop in 6 seconds
-	* because OnIdle is not called regularly, which is the case because SetReadTimeout is not
-	implemented by the HeaderConn.
-	* Add a runtime check in readLoop that SetReadTimeout works roughly correctly?
+* implement SetReadTimeout
 
 * ensure behavior stabilizes if both sides remain silent after connect and connection does not close
 perpetually

dccp/goto.go

@@ -117,7 +117,7 @@ func (c *Conn) gotoPARTOPEN() {
 	c.AssertLocked()
 	c.socket.SetState(PARTOPEN)
 	c.emitSetState()
-	c.openCCID()
+	// c.openCCID()
 	c.inject(nil) // Unblocks the writeLoop select, so it can see the state change
 
 	// Start PARTOPEN timer, according to Section 8.1.5
@@ -181,6 +181,7 @@ func (c *Conn) gotoCLOSING() {
 		c.Lock()
 		rtt := c.socket.GetRTT()
 		c.Unlock()
+		c.Logger.Emit("conn", "Event", nil, "CLOSING RTT=%dns", rtt)
 		b := newBackOff(2*rtt, CLOSING_BACKOFF_MAX, CLOSING_BACKOFF_FREQ)
 		for {
 			err, _ := b.Sleep()

dccp/socket.go

@@ -76,7 +76,7 @@ const (
 	RTT_MIN                = 2e6      // ...
 	MSL                    = 2 * 60e9 // 2 mins in nanoseconds, Maximum Segment Lifetime, Section 3.4
 	CLOSING_BACKOFF_FREQ   = 64e9     // Backoff frequency of CLOSING timer, 64 seconds, Section 8.3
-	CLOSING_BACKOFF_MAX    = MSL      // Maximum amount of time in CLOSING timer
+	CLOSING_BACKOFF_MAX    = MSL/4    // Maximum time in CLOSING (RFC recommends MSL, but seems too long)
 	MAX_OPTIONS_SIZE       = 128
 )