MOBILE COMPUTING

OE-II
UNIT IV: MOBILE TRANSPORT LAYER
• Traditional TCP
• Indirect TCP
• Snooping TCP
• Mobile TCP
• Fast retransmit/fast recovery
• Transmission/ time-out freezing
• Selective retransmission
• Transaction oriented TCP
Mobile transport layer - 1

• Supporting mobility only on lower layers

• up to network layer is not enough
• to provide mobility support for applications

• Most applications rely on transport layer

• such as TCP (transmission control protocol)
• UDP (user datagram protocol) in case of internet

• Two functions of transport layer in internet are

• checksumming over user data
• multiplexing/demultiplexing of data from/to applications
Mobile transport layer - 2

• Network layer only addresses host ports in UDP

• TCP allow dedicated applications to be addressed
• Connectionless UDP does not offer much more than this addressing

• While UDP is connectionless

• does not give certain guarantees about reliable data delivery
• TCP is much more complex
• needs special mechanisms to be useful in mobile environments

• Mobility support in IP such as mobile IP is already enough

• for UDP to work
Mobile transport layer - 3

• Main difference between UDP and TCP

• TCP offers connections between two applications

• TCP can give certain guarantees

• such as in-order delivery or reliable data transmission
• using retransmission techniques
• TCP has built-in mechanisms to behave in ‘network friendly’ manner

• If TCP encounters packet loss

• it assumes network internal congestion and slows down transmission rate
Mobile transport layer - 4

• Key requirement for new developments in internet is

• ‘TCP friendliness’

• UDP requires that applications handle reliability

• in-order delivery etc.

• UDP does not behave in network friendly manner

• i.e., does not pull back in case of congestion
• continues to send packets into already congested network
Traditional TCP

• Mechanisms of transmission control protocol (TCP)

• that influence
• efficiency of TCP in mobile environment

• Congestion control
• Slow start
• Fast retransmit/fast recovery
• Implications on mobility
Congestion control - 1

• Transport layer protocol such as TCP has been designed

• for fixed networks with fixed end-systems

• Data transmission takes place using network adapters

• fiber optics
• copper wires
• special hardware for routers etc.

• This hardware works without introducing transmission errors

Congestion control - 2

• If software is mature enough

• it will not drop packets or flip bits
• if packet on its way from sender to receiver is lost in fixed network
• it is not because of hardware or software errors

• Probable reason for packet loss in fixed network is temporary overload

• some point in transmission path
• i.e., state of congestion at node

• Congestion may appear from time to time

• even in carefully designed networks
Congestion control - 3

• Packet buffers of router are filled

• router cannot forward packets fast enough
• because sum of input rates of packets for output link is higher
• than capacity of output link
• Only thing router can do in this situation is to drop packets

• Dropped packet is lost for transmission

• receiver notices gap in packet stream

• Now receiver does not directly tell sender which packet is missing
• but continues to acknowledge all in-sequence packets up to missing one
Congestion control - 4

• Sender notices missing acknowledgement for lost packet

• assumes packet loss due to congestion

• Retransmitting missing packet

• continuing at full sending rate would now be unwise
• as this might only increase congestion

• It is not guaranteed that all packets of TCP connection

• take same way through network
Congestion control - 5

• To mitigate congestion
• TCP slows down transmission rate dramatically

• All other TCP connections experiencing same congestion

• do exactly same so congestion is soon resolved

• This cooperation of TCP connections in internet is one of main reasons

• for its survival as it is today

• Using UDP is not solution

• because throughput is higher compared to TCP connection just at beginning
Congestion control - 6

• As soon as everyone uses UDP

• this advantage disappears

• After that
• congestion is standard and data transmission quality is unpredictable

• Even under heavy load

• TCP guarantees at least sharing of the bandwidth
Slow start - 1

• TCP’s reaction to missing acknowledgement is

• Slow start

• Sender always calculates congestion window for receiver

• Start size of congestion window is one segment (TCP packet)
• Sender sends one packet and waits for acknowledgement

• If this acknowledgement arrives

• sender increases congestion window by one
• now sending two packets (congestion window = 2)
Slow start - 2

• After arrival of two corresponding acknowledgements

• sender again adds 2 to congestion window
• one for each of acknowledgements
• Now congestion window equals 4

• This scheme doubles congestion window every time

• acknowledgements come back
• which takes one round trip time (RTT)
• this is called exponential growth of congestion window in slow start mechanism
Slow start - 3

• It is too dangerous to double congestion window each time

• because steps might become too large
• exponential growth stops at congestion threshold

• As soon as congestion window reaches congestion threshold

• further increase of transmission rate is only linear
• by adding 1 to congestion window
• each time acknowledgements come back

• Linear increase continues until time-out at sender occurs

• due to missing acknowledgement
• sender sets congestion threshold to half of current congestion window
Slow start - 4

• Congestion window itself is set to one segment

• sender starts sending single segment

• Exponential growth starts once more up to new congestion threshold

• then window grows in linear fashion
Fast retransmit/fast recovery - 1

• Reason for reduction of congestion threshold

• sender receiving continuous acknowledgements for same packet

• This informs sender

• receiver got all packets up to acknowledged packet in sequence

• In TCP
• receiver sends acknowledgements
• only if it receives any packets from sender
Fast retransmit/fast recovery - 2

• Receiving acknowledgements from receiver shows

• receiver continuously receives something from sender

• Gap in packet stream is not due to severe congestion

• but simple packet loss due to transmission error

• Sender can now retransmit missing packet(s) before timer expires

• This behaviour is called fast retransmit

• Receipt of acknowledgements shows that

• there is no congestion to justify slow start
Fast retransmit/fast recovery - 3

• Sender can continue with current congestion window

• sender performs fast recovery from packet loss

• This mechanism can improve efficiency of TCP

• reason for activating slow start is time-out
• due to missing acknowledgement

• TCP using fast retransmit/fast recovery interprets

• congestion in network
• activates slow start mechanism
Implications on mobility - 1

• Slow start is useful mechanism in fixed networks

• decreases efficiency of TCP
• if used together with mobile receivers or senders
• Reason for this is use of slow start under wrong assumptions

• From missing acknowledgement

• TCP concludes congestion situation

• This may also happen in networks

• with mobile and wireless end-systems
• it is not main reason for packet loss
Implications on mobility - 2

• Error rates on wireless links are higher

• compared to fixed fiber or copper links

• Packet loss is much more common

• cannot always be compensated by layer 2 retransmissions
• error correction (FEC)

• Trying to retransmit on layer 2 could

• trigger TCP retransmission if it takes too long
• Layer 2 now faces problem of transmitting same packet twice over bad link
Implications on mobility - 3

• Detecting these duplicates on layer 2 is not option

• because more and more connections use end-to-end encryption
• making it impossible to look at packet

• Mobility itself can cause packet loss

• There are many situations where soft handover
• from one access point to another is not possible for mobile end-system
Implications on mobility - 4

• When using mobile IP

• there could still be some packets in transit to old foreign agent
• while mobile node moves to new foreign agent

• Old foreign agent may not be able to forward those packets

• to new foreign agent or even buffer the packets
• if disconnection of mobile node takes too long

• This packet loss has nothing to do with wireless access

• but is caused by problems of rerouting traffic
 Implications on mobility - 5

• TCP mechanism detecting missing acknowledgements via time-outs

• concluding packet loss due to congestion
• cannot distinguish between different causes

• This is fundamental design problem in TCP

• error control mechanism (missing acknowledgement due to transmission error)
• misused for congestion control (missing acknowledgement due to network overload)
• In both cases packets are lost (either due to invalid checksums or dropping in routers)
• Reasons are completely different
 Implications on mobility - 6

• TCP cannot distinguish between these two different reasons

• Standard TCP reacts with slow start if acknowledgements are missing
• which does not help in case of transmission errors over wireless links
• which does not really help during handover

• This behaviour results in performance degradation of TCP

• if used together with wireless links or mobile nodes
• cannot change TCP completely
• to support mobile users or wireless links
 Classical TCP improvements

• With introduction of WLANs in mid-nineties

• several research projects were started with goal
• to increase TCP’s performance in wireless and mobile environments
• Classical approaches
• Indirect TCP
• Snooping TCP
• Mobile TCP
• Additional optimizations
• Fast retransmit/fast recovery
• Transmission/time-out freezing
• Selective retransmission
• Transaction-oriented TCP
 Indirect TCP - 1

• Two competing insights led to development of indirect TCP (I-TCP)

• TCP performs poorly together with wireless links
• TCP within fixed network cannot be changed

• I-TCP segments
• TCP connection into
• fixed part
• wireless part
 Indirect TCP - 2

• Example with mobile host

• connected via wireless link
• access point to ‘wired’ internet
• where correspondent host resides
 Indirect TCP - 3

• Standard TCP is used between

• fixed computer
• access point

• No computer in internet
• recognizes any changes to TCP

• Instead of mobile host

• access point terminates
• standard TCP connection
• acting as proxy
 Indirect TCP - 4

• Access point is now seen as

• mobile host for fixed host
• as fixed host for mobile host

• Between access point and

mobile host
• special TCP is used
• adapted to wireless links

• Changing TCP for

• wireless link is not requirement
 Indirect TCP - 5

• Even unchanged TCP

• benefit shorter round trip time
• starting retransmission faster

• Good place for segmenting

• connection between mobile host
• correspondent host is at foreign
agent of mobile IP
 Indirect TCP - 6

• Foreign agent controls

• mobility of mobile host
• can also hand over connection
• to next foreign agent
• when mobile host moves on

• Correspondent host in fixed

network does not notice
• wireless link or segmentation of
connection
 Indirect TCP - 7

• Foreign agent acts as proxy

• relays all data in both directions

• If correspondent host sends packet

• foreign agent acknowledges packet
• tries to forward packet to mobile host

• If mobile host receives packet

• it acknowledges packet
• this acknowledgement is only used by
foreign agent
 Indirect TCP - 8

• If packet is lost on wireless link

• due to transmission error
• correspondent host would not
notice this
• foreign agent tries to retransmit
• this packet locally to maintain
• reliable data transport

• If mobile host sends packet

• foreign agent acknowledges this
packet
• tries to forward it to correspondent
host
 Indirect TCP - 9

• If packet is lost on wireless link

• mobile hosts notice this faster
• due to lower round trip time
• can directly retransmit packet
• Packet loss in wired network
• handled by foreign agent

• I-TCP requires several actions

• as soon as handover takes place
 Indirect TCP - 10

• After handover
• old proxy must forward buffered
data to new proxy
• because it has already
acknowledged the data

• After registration with new

foreign agent
• this new foreign agent can
inform old one about its location
• to enable packet forwarding
Advantages of Indirect TCP - 1

• I-TCP does not require any changes in TCP protocol

• as used by hosts in fixed network
• other hosts in wireless network
• that do not use this optimization

• All current optimizations for TCP still work between

• foreign agent and correspondent host
Advantages of Indirect TCP - 2

• Due to strict partitioning into two connections

• transmission errors on wireless link
• i.e., lost packets
• cannot propagate into fixed network

• Without partitioning
• retransmission of lost packets would take place
• between mobile host and correspondent host across whole network
• Now only packets in sequence
• without gaps leave foreign agent
Advantages of Indirect TCP - 3

• It is always dangerous to introduce new mechanisms

• into huge network such as internet
• without knowing exactly how they will behave

• New mechanisms are needed to improve TCP performance

• e.g., disabling slow start under certain circumstances
• but with I-TCP only between mobile host and foreign agent

• Different solutions can be tested or used at same time

• without jeopardizing the stability of internet
• Optimizing of these new mechanisms is quite simple
• because they only cover one single hop
Advantages of Indirect TCP - 4

• Optimized TCP could use precise time-outs

• to guarantee retransmission as fast as possible

• Even standard TCP could benefit from short round trip time
• so recovering faster from packet loss

• Delay is much higher in wide area wireless network

• than in wired networks due to FEC and MAC
Advantages of Indirect TCP - 5

• Partitioning into two connections

• allows use of different transport layer protocol
• between foreign agent and mobile host
• or use of compressed headers etc.

• Foreign agent can act as gateway

• to translate between different protocols
Disadvantages of Indirect TCP - 1

• Loss of end-to-end semantics of TCP might cause problems

• if foreign agent partitioning the TCP connection crashes
• If sender receives acknowledgement
• it assumes that receiver got the packet

• Receiving acknowledgement now only means

• that foreign agent received the packet

• Correspondent node does not know anything about partitioning

• so crashing access node may also crash applications
• running on correspondent node
Disadvantages of Indirect TCP - 2

• Increased handover latency may be much more problematic

• all packets sent by correspondent host are buffered by foreign agent
• besides forwarding them to mobile host
• foreign agent removes packet from buffer
• as soon as appropriate acknowledgement arrives

• If mobile host now performs handover to another foreign agent

• it takes while before old foreign agent forward buffered data to new foreign agent

• During this time more packets may arrive

• all these packets have to be forwarded to new foreign agent first
• before it can start forwarding new packets
Disadvantages of Indirect TCP - 3

• Foreign agent must be trusted entity

• because TCP connections end at this point

• If users apply end-to-end encryption

• foreign agent has to be integrated into all security mechanisms
Snooping TCP - 1

• I-TCP is segmentation of single TCP connection

• into two TCP connections
• loses original end-to-end TCP semantic

• Snooping TCP enhancement works transparently

• leaves TCP end-to-end connection intact

• Main function of enhancement is to buffer data

• close to mobile host to perform fast local retransmission
• in case of packet loss
Snooping TCP - 2

• Good place for enhancement of TCP could be

• foreign agent in Mobile IP context
Snooping TCP - 3

• Foreign agent buffers all packets with destination mobile host

• additionally ‘snoops’ packet flow in both directions
• to recognize acknowledgements
Snooping TCP - 4

• Reason for buffering packets toward mobile node is to enable

• foreign agent to perform local retransmission
• in case of packet loss on wireless link
• Foreign agent buffers every packet until it receives acknowledgement
• from mobile host
Snooping TCP - 5

• If foreign agent does not receive acknowledgement from mobile host

• within certain amount of time
• either packet or acknowledgement has been lost
• foreign agent could receive duplicate ACK
• which also shows loss of packet
Snooping TCP - 6

• Foreign agent retransmits packet directly from buffer

• performing much faster retransmission compared to correspondent host
• time out for acknowledgements can be much shorter
• because it reflects only delay of one hop plus processing time
Snooping TCP - 7

• To remain transparent
• foreign agent must not acknowledge data to correspondent host
• Foreign agent can filter duplicate acknowledgements
• to avoid unnecessary retransmissions of data from correspondent host
Snooping TCP - 8

• Foreign agent may discard duplicates of packets

• already retransmitted locally
• acknowledged by mobile host
• this avoids unnecessary traffic on wireless link
Snooping TCP - 9

• Data transfer from mobile host with destination correspondent host

• foreign agent snoops into packet stream
• to detect gaps in sequence numbers of TCP
Snooping TCP - 10

• As soon as foreign agent detects missing packet

• it returns negative acknowledgement (NACK) to mobile host
• mobile host can now retransmit missing packet immediately
• Reordering of packets is done automatically at correspondent host by TCP
Advantages Snooping TCP - 1

• End-to-end TCP semantic is preserved

• No matter at what time foreign agent crashes

• If this is the location of buffering and snooping mechanisms

• neither correspondent host nor mobile host
• have inconsistent view of TCP connection as is possible with I-TCP

• Approach automatically falls back to standard TCP

• if enhancements stop working
Advantages Snooping TCP - 2

• Correspondent host does not need to be changed

• most of enhancements are in foreign agent

• Supporting packet stream from correspondent host to mobile host

• does not even require changes in mobile host
Advantages Snooping TCP - 3

• It does not need handover of state

• as soon as mobile host moves to another foreign agent

• Assume there might still be data in buffer

• not transferred to next foreign agent

• All that happens is time-out at correspondent host

• retransmission of packets
• possibly already to new care-of address
Advantages Snooping TCP - 4

• It does not matter if next foreign agent uses enhancement or not

• If not approach automatically falls back to the standard solution

• This is one of the problems of I-TCP

• since old foreign agent may have already signaled
• correct receipt of data via acknowledgements to the correspondent host
• now has to transfer these packets to mobile host via new foreign agent
Disadvantages Snooping TCP - 1

• Snooping TCP does not isolate behaviour of wireless link

• as well as I-TCP
• it takes some time until foreign agent can successfully retransmit packet
• from its buffer due to problems on wireless link (congestion, interference)

• Although time-out in foreign agent may be much shorter

• than one of the correspondent host
• after a while time-out in correspondent host triggers retransmission
Disadvantages Snooping TCP - 2

• Problems on wireless link are visible for correspondent host

• not fully isolated

• Quality of isolation
• which snooping TCP offers
• strongly depends on quality of wireless link
• time-out values
• further traffic characteristics

• It is problematic that wireless link exhibits

• very high delays compared to wired link
• due to error correction on layer 2
• This is similar to I-TCP
Disadvantages Snooping TCP - 3

• If this is the case

• timers in foreign agent and correspondent host are almost equal
• approach is almost ineffective

• Using negative acknowledgements between

• foreign agent and mobile host assumes
• additional mechanisms on mobile host

• This approach is no longer transparent

• for arbitrary mobile hosts
Disadvantages Snooping TCP - 4

• All efforts for snooping and buffering data may be useless

• if certain encryption schemes are applied end-to-end between
• correspondent host and mobile host

• Using IP encapsulation security payload

• TCP protocol header will be encrypted
• snooping on sequence numbers will no longer work

• Retransmitting data from foreign agent may not work

• because many security schemes prevent replay attacks
• retransmitting data from foreign agent may be misinterpreted as replay
Disadvantages Snooping TCP - 5

• Encrypting end-to-end is the way many applications work

• so it is not clear how this scheme could be used in future

• If encryption is used above transport layer

• snooping TCP can be used
 Mobile TCP - 1

• Dropping packets due to handover or higher bit error rates

• is not only phenomenon of wireless links and mobility
• occurrence of lengthy and/or frequent disconnections is another problem
• Quite often mobile users cannot connect at all

• One example is islands of wireless LANs inside buildings

• but no coverage of the whole campus

• TCP sender tries to retransmit data controlled by retransmission timer

• that doubles with each unsuccessful retransmission attempt
• up to maximum of one minute
 Mobile TCP - 2

• This means that sender tries to retransmit unacknowledged packet

• every minute and will give up after 12 retransmissions
• No data is successfully transmitted for period of one minute
• Retransmission time-out is still valid and sender has to wait
• Sender also goes into slow-start because it assumes congestion

• Proxy has to buffer more and more data

• so longer the period of disconnection
• more buffer is needed

• If handover follows disconnection even more

• state has to be transferred to new proxy
 Mobile TCP - 3

• Snooping approach also suffers from being disconnected

• Mobile will not be able to send ACKs so
• snooping cannot help in this situation

• M-TCP approach has same goals as I-TCP and snooping TCP:

• to prevent sender window from shrinking
• if bit errors or disconnection
• but not congestion cause current problems

• M-TCP wants to improve overall throughput

• to lower the delay
• to maintain end-to-end semantics of TCP
• to provide more efficient handover
 Mobile TCP - 4

• M-TCP is especially adapted to problems

• arising from lengthy or frequent disconnections

• M-TCP splits TCP connection into two parts as I-TCP does

• unmodified TCP is used on standard host-supervisory host (SH) connection
• while optimized TCP is used on SH-MH connection

• Supervisory host is responsible for exchanging data between

• both parts similar to proxy in I-TCP
• M-TCP approach assumes relatively low bit error rate on wireless link
 Mobile TCP - 5

• It does not perform caching/retransmission of data via SH

• If packet is lost on wireless link
• it has to be retransmitted by original sender
• This maintains TCP end-to-end semantics

• SH monitors all packets sent to MH and ACKs returned from MH

• If SH does not receive an ACK for some time
• it assumes that MH is disconnected
• It then chokes sender by setting sender’s window size to 0
• Setting window size to 0 forces sender to go into persistent mode
• i.e., state of sender will not change no matter how long receiver is disconnected
 Mobile TCP - 6

• This means that sender will not try to retransmit data

• As soon as the SH detects connectivity again
• it reopens window of the sender to old value
• Sender can continue sending at full speed
• This mechanism does not require changes to sender’s TCP

• Wireless side uses adapted TCP

• that can recover from packet loss much faster
• This modified TCP does not use slow start
• M-TCP needs bandwidth manager to implement fair sharing over wireless link
 Advantages of Mobile TCP

• It maintains TCP end-to-end semantics

• SH does not send any ACK itself
• but forwards ACKs from MH

• If MH is disconnected
• it avoids useless retransmissions
• slow starts or breaking connections
• by simply shrinking sender’s window to 0

• Since it does not buffer data in SH as I-TCP does

• it is not necessary to forward buffers to SH
• Lost packets will be automatically retransmitted to SH
 Disadvantages of Mobile TCP

• Lack of buffers and changing TCP on wireless part has disadvantages:

• As SH does not act as proxy as in I-TCP
• packet loss on wireless link due to bit errors is propagated to sender
• M-TCP assumes low bit error rates
• which is not always valid assumption
• Modified TCP on wireless link not only requires modifications to MH protocol
• but also new network elements like bandwidth manager
 Fast retransmit/fast recovery - 1

• Moving to new foreign agent can cause packet loss or time out
• at mobile hosts or corresponding hosts

• TCP concludes congestion and goes into slow start

• although there is no congestion

• Mechanisms of fast recovery/fast retransmit can use

• after receiving duplicate acknowledgements
• thus concluding packet loss without congestion
• as soon as mobile host registers at new foreign agent using mobile IP
• it starts sending duplicated acknowledgements to correspondent hosts
 Fast retransmit/fast recovery - 2

• Proposal is to send three duplicates

• This forces corresponding host to go into fast retransmit mode
• not to start slow start
• correspondent host continues to send with same rate it did
• before the mobile host moved to another foreign agent

• As mobile host may also go into slow start

• after moving to new foreign agent
• this approach additionally puts mobile host into fast retransmit
 Fast retransmit/fast recovery - 3

• Mobile host retransmits all unacknowledged packets

• using current congestion window size without going into slow start
• Advantage of this approach is its simplicity

• Only minor changes in mobile host’s software

• result in performance increase
• No foreign agent or correspondent host has to be changed

• Main disadvantage of this scheme

• insufficient isolation of packet losses
 Fast retransmit/fast recovery - 4

• Forcing fast retransmission increases efficiency

• but retransmitted packets still have to cross whole network between
• correspondent host and mobile host

• Approach focuses on loss due to handover:

• packet loss due to problems on wireless link is not considered

• This approach requires more cooperation between

• mobile IP and TCP layer
• making it harder to change one without influencing other
 Transmission/time-out freezing - 1

• While approaches presented so far can handle

• short interruptions of the connection
• either due to handover or transmission errors on wireless link

• Some were designed for longer interruptions of transmission

• Examples are use of mobile hosts in car driving into tunnel
• which loses its connection to satellite
• user moving into cell with no capacity left over
• mobile phone system will interrupt the connection

• Reaction of TCP
• even with enhancements of above
• would be disconnection after time out
 Transmission/time-out freezing - 2

• MAC layer has already noticed connection problems

• before the connection is actually interrupted from TCP point of view
• MAC layer knows real reason for interruption
• does not assume congestion as TCP would

• MAC layer can inform TCP layer of upcoming loss of connection

• current interruption is not caused by congestion
• TCP can now stop sending
• ‘freezes’ current state of its congestion window and further timers
• If MAC layer notices upcoming interruption early enough
• both mobile and correspondent host can be informed
 Transmission/time-out freezing - 3

• With fast interruption of wireless link

• additional mechanisms in access point are needed
• to inform correspondent host of reason for interruption

• Otherwise correspondent host goes into slow start

• assuming congestion and finally breaks the connection

• As soon as MAC layer detects connectivity again

• it signals TCP that it can resume operation at exactly same point
• where it had been forced to stop

• For TCP time does not advance

• so no timers expire
 Transmission/time-out freezing - 4

• Advantage of this approach is it offers way to resume TCP connections

• even after longer interruptions of connection

• It is independent of any other TCP mechanism

• such as acknowledgements or sequence numbers
• so it can be used together with encrypted data

• This scheme has some severe disadvantages

• Not only does software on mobile host have to be changed
• to be more effective correspondent host cannot remain unchanged
 Transmission/time-out freezing - 5

• All mechanisms rely on capability of MAC layer

• to detect future interruptions

• Freezing state of TCP does not help

• in case of some encryption schemes
• that use time-dependent random numbers

• These schemes need resynchronization

• after interruption
 Selective retransmission - 1

• Very useful extension of TCP is use of selective retransmission

• TCP acknowledgements are cumulative
• they acknowledge in-order receipt of packets up to certain packet

• If single packet is lost

• sender has to retransmit everything starting from lost packet
• go-back-n retransmission

• This obviously wastes bandwidth

• not just in case of mobile network
• but for any network
• TCP can indirectly request selective retransmission of packets
 Selective retransmission - 2

• Receiver can acknowledge single packet

• Sender can now determine precisely
• which packet is needed and can retransmit it

• Advantage of this approach:

• sender retransmits only lost packets

• This lowers bandwidth requirements

• is extremely helpful in slow wireless links
• Gain in efficiency is not restricted to wireless links and mobile environments
• Using selective retransmission is also beneficial in all other networks
 Selective retransmission - 3

• Minor disadvantage of more complex software on receiver side

• because now more buffer is necessary to resequence data

• Memory sizes and CPU performance permanently increase

• bandwidth of air interface remains almost same

• Higher complexity is no real disadvantage any longer

• as it was in early days of TCP
Transaction-oriented TCP - 1

• Assume application running on mobile host

• that sends short request to server from time to time
• which responds with short message
• If application requires reliable transport of packets
• it may use TCP
• Using TCP requires several packets over wireless link
• TCP uses three-way handshake to establish the connection
• At least one additional packet is usually needed for transmission of request
• requires three more packets to close the connection via a three-way handshake
• Assuming connections with lot of traffic or with long duration
• this overhead is minimal
Transaction-oriented TCP - 2

• Figure shows example for overhead introduced

• by using TCP over GPRS in web scenario
Transaction-oriented TCP - 3

• Web services are based on HTTP

• which requires reliable transport system
• In the internet
• TCP is used for this purpose
• Before a HTTP request can be transmitted
• TCP connection has to be established
• This already requires three messages
• If GPRS is used as wide area transport system
• one-way delays of 500 ms and more are quite
common
 Transaction-oriented TCP - 4

• Setup of TCP connection already takes

• far more than second

• This led to
• development of transaction-oriented TCP (T/TCP)

• T/TCP can combine packets for

• connection establishment and connection release
• with user data packets

• This can reduce number of packets down to

• two instead of seven
Transaction-oriented TCP - 5

• Advantage for certain applications is

• reduction in overhead
• which standard TCP has for connection setup
• connection release

• T/TCP is not original TCP anymore

• so it requires changes in mobile host
• all correspondent hosts
• which is major disadvantage

• This solution no longer hides mobility

Classical enhancements to TCP for mobility

• table
Thank you