1. Wireless TCP Prasun Dewan Department of Computer Science University of North Carolina dewan@unc.edu

2. 2 Problem n TCP layered on top of of IP n IP interface provided to TCP is independent of physical layer u Implementation dependent on physical layer u Wireless just another physical Layer n Problem?

3. 3 Tuning problem n Working correctly not an issue n Working efficiently is n On wire links losses (normally) due to congestion n On wireless losses can be due to u Unreliable physical medium u Intermittent connectivity u Handoff losses F Can be reduced With old base station buffering messages With adjacent base stations joining a multicast group and buffering messages –Reduces delay n TCP/IP policies for wired links will mistake wireless losses and delays for router congestion

4. 4 TCP/IP Response to Losses n Assume losses due to congestion n Drops (halves) transmission window size u Window size determines how many packets can be sent before waiting for ack u Determines throughput n If losses due to unreliable physical layer needless reduction of throughput

5. 5 Solution n Distinguish between congestion and other losses n Do not reduce window in response to non congestion losses

6. 6 Issues n Where in the path from sender to receiver to solve problem? n How to distinguish between the two reasons for losses?

7. 7 Possible Adaptations n Where in the path from sender to receiver to solve problem? u End-to-end F Sender and receiver together addresses problem F They address congestion loss F They should also address medium loss u Link layer addresses problem F Problem occurred in the link F Should be solved there too u Split connection F One TCP connection from wired end to base, another from base to wireless end F Problem solved locally F But solved at TCP layer (more semantics)

8. 8 End-to-end n How to distinguish between the two reasons for losses?

9. 9 Traditional TCP Loss Detection n Timeout u average round trip time + 4*mean deviation n Duplicate acks from receiver u Ack indicates sequence number of next expected message u If message m i gets lost then acks of subsequent messages will have sequence # i

10. 10 Explicit Loss Notification n When packet dropped over wireless u Subsequent acks indicate loss occurred n How loss detected in Wireless LAN u If corrupted packet F Receiver detects CRC errors F Passes to transport layer u If entire packet is lost F Base station observes duplicate acks F Attaches ELN to them n What if wireless link is sending? u Congestion vs. loss

11. 11 Problem with End to End n Un-necessary duplicate acks sent all the way to source n Un-necessary retransmissions from source to destination n Does not address wireless sender

12. 12 Link Level n Handle the problem at the link level,that is where the loss occurred. n Local retransmission instead of end to end retransmission n Link-level timer much smaller (~20ms) n TCP timers larger (multiples of 500 ms) u Depends on end to end delay n Several possibilities u Window of size 1 u Larger window F Cumulative acks F SMART scheme Cumulative ack mentions sequence # of received packet Previous packets presumed to be lost rather than reordered

13. 13 Link Level Problems n Both end to end and link level retransmissions are working n “Incompatible” timers cause retransmission by both parties u Unless > 10% loss retransmissions occur u End to end timeout should not occur n Link-level does not try to deliver in order n Duplicate retransmissions can reach receiver because of out of order delivery n If messages in order, dup acks and end to end retransmissions do not occur, unlike end to end n Can add TCP awareness to link level so dup acks do not occur n Layering broken

14. 14 Split Connection Algorithm n Break a single TCP connection from wired end to wireless end into u TCP connection from wired end to base station u TCP connection from base station to wireless end F SMART scheme better than RENO since sender does not wait for timeout or duplicate acks F Out of order messages do not occur on single link n End to end out of order delivery does not occur n Sender never gets duplicate acks n Two TCP stacks encountered u Sharing of pointers between stacks at base station helps n End to end semantics violated u Sender can get ack before receiver gets message n Buffer space at base station bounded u Does not ack wired end when this happens? n Original TCP sender can stall n Link level with TCP awareness best