prof. younghee lee 1 computer networks u lecture 10: tcp: wireless network prof. younghee lee

27
1 Prof. Younghee Lee Computer networks Lecture 10: TCP: wireless network Prof. Younghee Lee

Upload: harriet-lawson

Post on 17-Jan-2018

227 views

Category:

Documents


0 download

DESCRIPTION

Prof. Younghee Lee 3 What is the problem? u Reducing cwnd make sense? –On a CDMA channel, errors occur due to interference from other user, and due to noise »Interference due to other users is an indication of congestion. If such interference causes transmission errors, it is appropriate to reduce cwnd »If noise causes errors, it is not appropriate to reduce cwnd. –When a channel is in a bad state for a long duration »It might be better to let TCP backoff, so that it does not unnecessarily attempt retransmissions while the channel remains in the bad state u Solution: –“wireless-aware TCP” (I-TCP, ProxyTCP, Snoop-TCP, split connections...) –New TCP version?: Reactive, Proactive

TRANSCRIPT

Page 1: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

1Prof. Younghee Lee

Computer networks Lecture 10: TCP: wireless network

Prof. Younghee Lee

Page 2: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

2Prof. Younghee Lee

What is the problem? TCP assumes that packet loss is due to congestio

n. Wireless Network:

– Higher transmission error rate– Frequent handoff– Burst error => time out => ss

Page 3: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

3Prof. Younghee Lee

What is the problem? Reducing cwnd make sense?

– On a CDMA channel, errors occur due to interference from other user, and due to noise

» Interference due to other users is an indication of congestion. If such interference causes transmission errors, it is appropriate to reduce cwnd

» If noise causes errors, it is not appropriate to reduce cwnd.

– When a channel is in a bad state for a long duration» It might be better to let TCP backoff, so that it does not unnecessarily attempt retransmissions whil

e the channel remains in the bad state

Solution: – “wireless-aware TCP” (I-TCP, ProxyTCP, Snoop-TCP, split connections...)– New TCP version?: Reactive, Proactive

Page 4: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

4Prof. Younghee Lee

Various solution• IEEE database has 599 papers on Wireless TCP as of Fall 2004…

Page 5: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

5Prof. Younghee Lee

Various Solution Link level mechanisms Split connection approach TCP-Aware link layer: snoop TCP-Unaware approximation of TCP-aware link lay

er: Delayed Dupacks Explicit notification: no cwnd control on packet loss Receiver-based discrimination Sender-based discrimination

Page 6: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

6Prof. Younghee Lee

Link level retransmission? Advantages

– No changes at TCP Disadvantages

– Possible retransmission same packet at both link layer and TCP– Reordering– Highly variable delay

Retx on ARQ» CDMA, TDMA use ARQ» AIRMAIL - ARQ and FEC» Why?

FEC?– Correctable errors hidden from the TCP sender– Incurs overhead => adaptive FEC [Eckhardt98]; choosing appropriate F

EC dynamically Possible choice

– Low error rate => FEC– High / bursty error => link level retransmission

Page 7: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

7Prof. Younghee Lee

Link level retransmission? Retransmissions can cause head-of-the-line

blocking– One BS to multiple receivers– BS uses FIFO – A receiver in a bad state => multiple retransmission =>

block forwarding the messages to receivers in good state Retransmission => increase RTT therefore RTO

Page 8: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

8Prof. Younghee Lee

Link level retransmission? Large TCP Retransmission Timeout Intervals

– Good for reducing interference with link level retransmits» How many times to retransmit at the link layer before giving up?

No bound: reliable but large RTT => TCP level retransmission?

– Bad for recovery from congestion losses– Need a timeout mechanism that responds appropriately

for both types of losses: congestion loss? Link error? When is that beneficial to TCP performance?

– in-order delivery is beneficial to TCP: selective retransmission, Go back N

– Large TCP retransmission timeout

Page 9: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

9Prof. Younghee Lee

Split connection approach: Indirect TCP(I-TCP)

2 TCP connections– SACK for BS <-> MH

Advantages– Optimize performance to wireless TCP– No changes to TCP for FH

Disadvantages– Violate end to end semantic: Ack..– High overhead due to dual stack at BS, high delay due to queueing delay at BS, extra copying of data at BS– Overhead is heavy when handoff is frequent– BS retains hard state: BS fails? => hand-off latency increases due to state transfer…

FH

BS

BS MH

cell2

cell1TCP TCP Tailored

for wireless

Page 10: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

10Prof. Younghee Lee

Indirect TCP(I-TCP)(2) Implementation: proxy

I-TCP proxyUser space

KernelSocket layer

TCP layer

IP layer

To FHTo MH

FH socketMH socket

Page 11: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

11Prof. Younghee Lee

Indirect TCP(I-TCP)(3)

Wireless link: 2Mbps

Ethernet link: 10Mbps

Local area throughput

Wide area throughput

TCP

I-TCP

I-TCP(M->F)

TCP(F->M)

I-TCP(F->M)TCP(M->F)

Page 12: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

12Prof. Younghee Lee

Split connection approach: Variations

1. Selective Repeat Protocol(SRP) over UDP for tuning to wireless behavior– Mobile-End Transport Protocol

» Terminate the TCP connection at BS TCP connection runs only between BS and FH

» BS pretends to be MH (MH’s IP functionality moved to BS): (metaphor: NI <-> Host)» BS guarantees reliable ordered delivery of packets to MH

BS-MH link can use any arbitrary protocol optimized for wireless link– Performance better than I-TCP

» Why?

2. Asymmetric transport protocol ( Mobile-TCP)– Lower overhead protocol at MH, higher overhead protocol at BS & FH

» Smaller header (header compression) , simpler on/off flow control, only error detection, no congestion control at wireless link ; BS does error correction

3. Mobile End Transport Protocol: BS pretends to be MH(MH’s IP function also)

FH

BS

BS MH

cell2

cell1TCP SRP over

UDP

Page 13: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

13Prof. Younghee Lee

TCP aware link layer : snoop UC Berkeley for wireless last-hop network: Improves on split connection

– end-to-end semantics retained– soft state at base station, instead of hard state

Snoop agent present at BS: FH to MH– Lost segments are detected and retransmitted locally– Last-hop round times are estimated– Suppression of duplicated ACKs corresponding to wireless losses avoids unnecessary invocation of fast retransmission.

» retransmit on wireless link, if packet present in buffer– For MH to FH: Explicit Loss Notification(ELN): decoupling of retransmission from congestion control– NACK implementation similar to TCP SACK– Requires modification to only BS– Not useful if TCP headers are encrypted (IPsec)

FH

BS

BS MH

cell2cell1

TCP

Packet retransmission

Page 14: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

14Prof. Younghee Lee

WTCP Motivation

– Snoop hides wireless losses from the sender– But sender’s RTT estimates may be larger in presence of errors– Larger RTO results in slower response for congestion losses

WTCP protocol– WTCP performs local recovery, similar to Snoop– uses the timestamp option to estimate RTT– The base station adds base station residence time to the timestam

p when processing an ack received from the wireless host– Sender’s RTT estimate: not affected by retransmissions on wireles

s link

Page 15: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

15Prof. Younghee Lee

Implementation comparisonCodechange

Hand-offhandle

End-to-endsemantics

Symmetricdemand

Wirelessbottleneckhandle

snoop ++ no yes yes no

I-TCP ++ yes no yes yesWTCP +++ yes yes no yes

Performance: no enough information to quantify which one has the best performance. I-TCP is reported by different group to have decent improvement. Snoop degrades when wireless transmission becomes the bottleneck.

Page 16: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

16Prof. Younghee Lee

TCP-Unaware Approximation of TCP-Aware Link Layer

Delayed Dupacks Protocol [Mehta98,Vaidya99]– Attempts to imitate Snoop, without making the base station TCP-aware– Snoop implements two features at the base station

» link layer retransmission» reducing interference between TCP and link layer retransmissions (by dropping dupacks)

– Delayed Dupacks implements the same two features» at BS : link layer retransmission» at MH : reducing interference between TCP and link layer retransmissions (by delaying dupacks)

Page 17: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

17Prof. Younghee Lee

TCP-Unaware Approximation of TCP-Aware Link Layer

Link layer delivers packets out-of-order when transmission errors occur– Why may a link layer deliver packets out-of-order?

» Only an issue when the link layer does not use stop-and-go protocol– With OOO link layer delivery, loss of a packet from one flow does not block delivery of packets from another flow– If in-order delivery is enforced, when retransmission for a packet is being performed, packets from other flows may also be blocked from being delive

red to the upper layer– TCP receiver delays dupacks (third and subsequent) for interval D, when out-of-order packets received– Dupack delay intended to give link level retransmit time to succeed– Benefit: Delayed dupacks can result in recovery from a transmission loss without triggering a response from the TCP sender

» Can be used even if TCP headers are encrypted– Disadvantage: Recovery from congestion losses delayed

Page 18: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

18Prof. Younghee Lee

TCP variants TCP SACK

– Selective acknowledgement – The receiver sends back ACKs to inform the sender that

the data has been successfully received so that the sender can then retransmit only the missing packet

TCP Westwood

Page 19: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

19Prof. Younghee Lee

Explicit Notification Approximate Ideal TCP behavior: Ideally, the TCP sender

should simply retransmit a packet lost due to transmission errors, without taking any congestion control actions

A wireless node somehow determines that packets are lost due to errors and informs the sender using an explicit notification

Sender, on receiving the notification, does not reduce congestion window, but retransmits lost packet

Motivated by the Explicit Congestion Notification (ECN) proposals [Floyd94]

Page 20: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

20Prof. Younghee Lee

Explicit Notification Explicit Bad State Notification [Bakshi97] when MH

is TCP receiver– Base station attempts to deliver packets to the MH using

a link layer retransmission scheme– If packet cannot be delivered using a small number of

retransmissions, BS sends a Explicit Bad State Notification (EBSN) message to TCP sender

– When TCP sender receives EBSN, it resets its timer» Why? » timeout delayed, when wireless channel in bad state

There are various Explicit Loss Notification Schemes

Page 21: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

21Prof. Younghee Lee

Receiver-Based Scheme TCP Receiver(MH) uses a heuristic to guess cause of packet loss

– When receiver believes that packet loss is due to errors, it sends a notification to the TCP sender

– TCP sender, on receiving the notification, retransmits the lost packet, but does not reduce congestion window

– Receiver uses the inter-arrival time between consecutively received packets to guess the cause of a packet loss: from error? Or congestion?

On determining a packet loss as being due to errors, the receiver may – tag corresponding dupacks with an ELN bit, or send an explicit notification to s

ender Advantages

– Can be implemented without modifying the base station – May be used despite encryption, or if data & acks traverse different paths

Disadvantages– Limited applicability– The slowest link on the path must be the last wireless hop

» to ensure some queuing will occur at the base station– The queueing delays for all packets (at the base station) should be somewhat

uniform: multiple connections on the link will make inter-packet delays variable

Page 22: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

22Prof. Younghee Lee

Sender based Discrimination Scheme Sender can attempt to determine cause of a packet loss

– For example;» TCP Vegas [Brakmo94]

expected throughput ET = W(i) / RTTmin actual throughput AT = W(i) / RTT(i) Condition C = ( ET-AT > beta)

» When a packet loss is detected if last evaluation of C is TRUE, assume packet loss is due to congestion else assume that packet loss is due to transmission errors

If packet loss determined to be due to errors, do not reduce congestion window

Sender can only use statistics based on round-trip times, window sizes, and loss pattern– unless network provides more information (example: explicit loss notifi

cation) Does not work quite well enough as yet !! Only sender needs to be modified

Page 23: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

23Prof. Younghee Lee

TCPs over Satellite Long latency Larger congestion window or larger initial window SACK: allow multiple packet recovery per RTT Space Communication Protocol Standard Transport Protocol (SCPS-TP) Satellite Transport Protocol (STP): split connection approach Early Acks:

– Spoofing: Early acks from ground station result in faster congestion window growth

TCP-Peach [1] – Applications such as HTTP are based on the transfer of small files

» the entire transfer occurs within the slow start phase and the connection is not able to fully utilize the available network bandwidth.

– sudden start and rapid recovery with the traditional TCP’s congestion avoidance and fast retransmit algorithms

» Sudden start introduces the use of dummy packets that are copies of the last data packet the sender has sent.

» The sender sends multiple dummy packets between two consecutive data packets» the reception of ACKs for the dummy packets indicates the availability of the unused network r

esource, and the sender is triggered to increase its sending window quickly.[1] I. F. Akyildiz, G. Morabito, and S. Palazzo, “TCP-Peach: A New Congestion Control Scheme for Satellite IP Networks,” IEEE/ACM Trans. Net., vol. 9, no. 3, June 2001, pp. 307–21.

Page 24: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

24Prof. Younghee Lee

Yet another TCP variants Impact of mobility on TCP Performance

– Hand-off, Mobile IP– Rendezvous Delay ; 1 sec after cell boundary crossed– M-TCP, – Freeze TCP

» It imposes no restrictions on routers and only requires code modifications at the mobile unit or receiver side.

» Mobile handoff or temporary blockage of radio signals by obstacles. » The mobile unit predicts the impending disconnections ; signal strength» The freeze-TCP receiver on the mobile unit proactively sets the advertised window

size to zero in the ACK packets in the presence of impending disconnection. force the sender into persist mode, where it ceases sending more packets whi

le keeping its sending window unchanged.» To implement, cross-layer information must be exchanged,

TCP exposed to handoff algorithms implemented by network interface card(NIC) vendors on the interface devices

Page 25: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

25Prof. Younghee Lee

Yet another TCP variants Issues in multihop wireless networks

– TCP for MANET– ATCP

» a thin layer inserted between the standard TCP and IP layers» explicit congestion notification (ECN) to detect congestion and distingui

sh congestion loss from error loss» ICMP Destination Unreachable message to detect a change of route or

temporary partition» According to these feedbacks, the ATCP layer puts the TCP sender int

o either the persist, congestion control, or retransmit state accordingly» The scheme does not modify the TCP code» TCP semantic is maintained.

Page 26: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

26Prof. Younghee Lee

Wireless TCP Reactive approach

– TCP new Reno, TCP SACK Proactive approach

– TCP Vegas, TCP Westwood– TCP Veno

» same methodology as Vegas» It further suggests a way to differentiate the cause of packet loss

If the number of backlogged packets is below a threshold– Not congested => the loss is considered to be random. – It increases the congestion window in a conservative man

ner Otherwise congestive => Reno scheme

Page 27: Prof. Younghee Lee 1 Computer networks u Lecture 10: TCP: wireless network Prof. Younghee Lee

27Prof. Younghee Lee

Wireless TCP Proactive approach

– TCP Jersey» the available bandwidth estimation (ABE) algorithm

TCP sender side addition continuously estimates the bandwidth available to the connection adjust its transmission rate when the network becomes congested

» the congestion warning (CW) router configuration Routers alert end stations by marking all packets when there is a

sign of incipient congestion. sender of the TCP connection to differentiate packet losses

» calculates the optimum congestion window size using ABE, CW» ECN at the routers to implement CW