- Allow at least 3 packets and up to half the window to be active resends
instead of just 1, to reduce stall time after a packet drop
- Increase fast retransmit threshold back to 3 to reduce retransmissions
- Don't fast retransmit if we recently retransmitted it already
- Allow double the window as long as gaps are less than the window
- Don't set the MSS in a resent packet (saves 2 bytes)
- Remove redundant calls to updateAcks()
- Update activity timer when resending a packet
- Reset unackedPacketsReceived counter at all places where acks are sent
so it isn't wrong
- Fix some places where the activeResends count could become wrong
- Prevent storm of CLOSE packets
- Never resend the whole packet in ackImmediately(), just send an ack
- Cancel flusher timer in MessageOutputStream when closed
- Move some createRateStats to ConnectionManager to reduce repeated calls
- Cleanups, javadocs, logging, volatile, finals
- More client options cleanups
- Options changes now propagate to running
socket managers and sessions, and through to the router
* SocketManager:
- Simplify factory, use 4-arg constructor,
make fields final, deprecate 0-arg constructor
- Improve how options are updated
- Javadocs
- Make flush() block less, by waiting only for "accept" into the
streaming queue rather than "completion" (i.e. ACK from the far end).
This prevents complete window stalls when flushing, and should help performance
of apps that use flush(), like i2psnark (and SAM?).
close() still does a flush that waits for completion, as i2ptunnel
doesn't like a fast return from close().
- flush/close javadocs and comments
* i2ptunnel:
- Now that streaming flush() is fixed, use it in IRCClient, and
for initial data in I2PTunnelRunner, to avoid the 250 ms
passive flush delay