Presentation is loading. Please wait.

Presentation is loading. Please wait.

TCP performance in Wireless Networks Ehsan Hamadani July 2004.

Similar presentations


Presentation on theme: "TCP performance in Wireless Networks Ehsan Hamadani July 2004."— Presentation transcript:

1 TCP performance in Wireless Networks Ehsan Hamadani E.hamadani@city.ac.uk July 2004

2 Presentation Outlines In this presentation, I have the following goals: An overview on TCP functionality and its problems in wireless links. Review the main idea of proposed solutions to TCP when they are employed in wireless links.

3 TCP Reliable Delivery Receiver ACKs No ACK for packet 4 =>Timeout 7 6 5 4 3 21 Data 2 Sender 3 Buffer

4 halved Packet loss TCP Congestion Control CWND t Congestion window size (CWND) limits the rate at which sender can send data traffic, in order to avoid congestion situation in network. congestion window is smaller when congestion is larger and vice versa Slow Start Congestion Avoidance

5 TCP Problems in wireless links In Wire network: Packet Loss => Congestion In Mobile Networks: Packet Loss => disconnections, handoffs, handoffs, high bit error rates high bit error rates TCP was designed to work well in networks with low channel error rates, however TCP was designed to work well in networks with low channel error rates, however Wireless links can exhibit high & variable error rates due to attenuation, fading, or interfering active radiation sources. TCP has no means of detecting the nature of the error, so losses due to channel errors also get treated as congestion losses

6 Proposed Modifications to current TCP Split connection Link level approach Network assisted approach End-to-End approach

7 Split connection Link layer approach Network assisted approach End-to-End approach

8 Split Connection The main idea => To isolate wireless related problems from the existing network protocols. Splitting the TCP connection into 2 independent flow control between the mobile host and the fixed host into two separate connections: 1)Wired connection between the fixed host and the base station 2)Wireless connection between the base station and the mobile host. MH BS FH Connection1 Connection2

9 Split connection-cont Advantages 1)Separate the flow control and congestion control functionality on the wireless link from wire network 2)Allows the base station to manage much of the traffic overhead 3)Allows the use of varying but optimal MTU over both networks 4)Faster reaction to mobility and wireless related events 5)No modifications are needed at the fixed hosts Disadvantages 1)Violation of TCP end to end semantics (An acknowledgment may be delivered to the sender even before the data actually reaches the destination) 2)Failure of base station may will lead to disruption of connection while normal TCP may survive 3)The packets have to incur of overhead of going through the TCP stack protocol twice at the base station as against the zero times for non-split connection 4)Experiencing latency in exchanging state information during hand-off 5)It needs large buffer size at the base station 6)Less effective than other approaches

10 Split connection Link level approach Network assisted approach End-to-End approach

11 Link Level approach  Main idea => The problem is local, and hence should be solved locally.  It hides the characteristics of the wireless link from the transport layer and tries to solve the problem at the link layer.  They use techniques like forward error correction (FEC) and automatic repeat request (ARQ).  Advantages: 1)It retains end-to-end semantics of TCP while achieving a very high throughput 2)Entities on the link are likely to be able to respond more quickly to changes in the error environment. 3)End-to-end retransmission demands bandwidth on every link, while local retransmission requires extra bandwidth only where it’s truly needed.  Disadvantages: 1)could result in packet reordering or in large fluctuations of the roundtrip time, either of which could trigger TCP timeouts and retransmissions. 2)duplicate retransmissions generated by the local recovery protocol and by TCP. 3)Sometimes it is better to drop a late packet than to retransmit it, since retransmission may make the next packet late as well (TCP unaware). 4)They result in higher RTT estimation resulting in wrong value of RTO for TCP. 5)link layer protocols do not help in the case of long or frequent disconnections. Determining link level timeout value & number of retransmissions Determining link level timeout value & number of retransmissions Critical factors Interaction between link layer and TCP timeout Interaction between link layer and TCP timeout

12 Snoop Protocol In the snoop protocol the base station monitors communication across the connection in either direction and caches packets. A packet loss is detected by the arrival of a small number of duplicate acknowledgements from the receiver or by local timeout at the base station. Basic operation at BS: Basic operation at BS: Data from MH -> FH Data from MH -> FH Detect missing packets Perform negative acknowledgements Data from FH -> MH Data from FH -> MH Cache unacknowledged TCP data Perform local retransmissions MHFH BS

13 Split connection Link level approach Network assisted approach End-to-End approach

14 TCP Feedback Main approach => When there is no route available, there is no need to retransmit packets that will anyway not reach the destination In TCP-F the TCP sender can distinguish between route failure and network congestion by receiving Route Failure Notification (RFN) from intermediate nodes. The idea is to push the TCP into a “snooze state” when such messages are received. In this state: i.TCP stops sending further packets (new or retransmission) ii.It freezes all its variables such as timers and CWND size, Upon receipt of a Route Re-establishment Notification (RRN), via routing protocol, indicating that there is again an available path to the destination, the sender leaves the frozen state and resumes transmission using the same variables values prior to the interruption. Upon receipt of a Route Re-establishment Notification (RRN), via routing protocol, indicating that there is again an available path to the destination, the sender leaves the frozen state and resumes transmission using the same variables values prior to the interruption. In addition, a route failure timer is used to prevent infinite wait for RRN messages. It is triggered whenever a RFN is received, and in case it expires the frozen timers are reset allowing the TCP congestion control to be invoked normally.

15 Explicit Bad State Notification (ESBN) If the wireless link is in bad state, the base station sends an EBSN message to the source for every retransmission of a segment to the mobile host in order to prevent the source from timing out. The EBSN message causes the source to reinitialize the timer. If the base station sends the EBSN messages before the source timer expires, then there will be no timeouts at the source. Disadvantage 1) Requires TCP code modification at the source to be able to interpret EBSN messages. 2) Security issues

16 Split connection Link level approach Network assisted approach End-to-End approach

17 End to End Schemes Main idea =>TCP sender should handle all types of losses. encrypted TCP headers or if TCP data and acks do not go through same base station Other solutions would not work if we use encrypted TCP headers or if TCP data and acks do not go through same base station Advantages It keeps the end-to-end semantics of TCP. Only require revision in the TCP algorithms at the sender and minimal or no processing at the intermediate nodes Disadvantages It is not as precise and quick as link level approach It adds more complexity and process overhead to sender side.

18 Wireless TCP (W-TCP) Use inter-packet delay (as opposed to losses) as indication of congestion Specifically, the relative differences between the average receiver inter packet delay and the average sender inter packet delay is used to detect congestion Specifically, the relative differences between the average receiver inter packet delay and the average sender inter packet delay is used to detect congestion Let DS be the interpacket delay at the sender Let DR be the interpacket delay at the receiver Let DR be the interpacket delay at the receiver i. DR > DS: congestion! ii. DR = DS: no congestion iii. DR < DS: not possible on a consistent basis

19 Conclusion 1) What steps we should take to overcome problems in lossy wireless medium?  This depends on the nature and location of the error. For instance: Adjust the timeout and apply graduated adjustments after retransmission (e.g., during congestion); Adjust the timeout and apply graduated adjustments after retransmission (e.g., during congestion); Suspend data transmission for periods of “blackout” and retransmit aggressively when conditions clear (e.g., during hand-offs); Suspend data transmission for periods of “blackout” and retransmit aggressively when conditions clear (e.g., during hand-offs); Reduce transmission effort for some period of time (e.g., during fading channels); Reduce transmission effort for some period of time (e.g., during fading channels); Do nothing (e.g., after a transient random loss). Do nothing (e.g., after a transient random loss). 2) where is the right place to add the required functionality?   This question does not have a clear answer. It depends on trade-off between cost, applicability, protocol approach, application, …

20 Questions? Thanks


Download ppt "TCP performance in Wireless Networks Ehsan Hamadani July 2004."

Similar presentations


Ads by Google