Transport layer protocols for ad hoc networks

Lecturer: Dmitri A. Moltchanov

E-mail: [email protected]

http://www.cs.tut.fi/kurssit/TLT-2756/
Ad hoc networks D.Moltchanov, TUT, Fall 2009
• Which transport layer protocol?
• Classification of transport layer protocols;
• Traditional TCP;
• Why does not TCP work well in ad-hoc networks?
• TCP enhancements;
• Other transport layer protocols.

Lecture: Transport layer protocols for ad hoc networks 2

Ad hoc networks D.Moltchanov, TUT, Fall 2009

1. Which transport layer protocol?

TCP dominates in wired Internet with the traffic share in between 80%–90%.
It is characterized by the following properties:
• TCP is reliable;
• TCP incorporates congestion control mechanism;
• TCP incorporates end-to-end flow control mechanism.

The following observations can be made concerning ad-hoc networks:

• it is preferable to seamlessly integrate TCP in ad-hoc networks:
– to enable seamless operation of higher layer protocols such as FTP, SMTP, HTTP.
• if not, to make as less modifications to TCP as possible:
– to make ”wireless” and ”wired” TCPs understand each other seamlessly.
• if not, to split the TCP into wireless and wired part:
– to concentrate internetworking functions in gateways.

Lecture: Transport layer protocols for ad hoc networks 3

Ad hoc networks D.Moltchanov, TUT, Fall 2009

2. Classification of the transport layer protocols

Transport layer protocols for ad-hoc wireless
networks

TCP modifications Other protocols

- ACTP;
End-to-end - ATP.
Split approaches
approaches

- Split-TCP. - TCP-ELFN;
- TCP-F;
- TCP-BuS;
- ATCP.

Figure 1: Classification of transport layer protocols for ad-hoc networks.

Lecture: Transport layer protocols for ad hoc networks 4

Ad hoc networks D.Moltchanov, TUT, Fall 2009

3. Traditional TCP
The major responsibilities of TCP in an active session are to:
• provide reliable in-order transport of data:
– to not allow losses of data.
• control congestions in the networks:
– to not allow degradation of the network performance.
• control a packet flow between the transmitter and the receiver:
– to not exceed the receiver’s capacity.
In general, we distinguish between the following operational phases in TCP:
• slow-start phase (also known as exponential start);
• congestion avoidance phase;
• congestion control phase;
• fast retransmit phase;
• fast recovery.

Lecture: Transport layer protocols for ad hoc networks 5

Ad hoc networks D.Moltchanov, TUT, Fall 2009
1. TCP SEGMENT STRUCTURE
32 bits

Source port number Destination port number

Sequence number

Acknowledgement number
Header U A P R S F
Unused R C S S Y I Window size
length G K H T N N

Checksum Pointer to urgent data

Options

Data

Figure 2: TCP segment structure.

Lecture: Transport layer protocols for ad hoc networks 6

Ad hoc networks D.Moltchanov, TUT, Fall 2009
2. TCP CONNECTION ESTABLISHMENT AND TEAR DOWN
Sender Receiver
Conn. req. SYN segm
. (SY N=1, seqNum=x, noData, )

ACK (SY N=1, seqNum =y, ack=x+1, noData, )

Conn. OK. SYN
ACK (SYN=0, seqNum x+1, ack
= =y+1, noData, )

FIN (FIN=1, noData, )

ACK (noData, )

FIN (FIN=1, noData, )

ACK (noData, )

Figure 3: TCP connection establishment and tear down.

Lecture: Transport layer protocols for ad hoc networks 7

Ad hoc networks D.Moltchanov, TUT, Fall 2009
3. ROUND TRIP TIME AND TIMEOUT
Every time TCP sends a segment, it starts a timeout:
• it must be larger than RTT: to avoid unnecessary retransmission;
• it should not be much larger than RTT: to not introduce big delays.

The following dynamic expression is used for setting a timeout:

• Timeout = Estimate of RTT + 4×Deviation, where
– Sample RTT = t(the ACK is received) − t(the segment is sent to IP);
– Estimate of RTT = (1 − x) Estimate of RTT + x Sample RTT;
– x = 0, 125 is mostly chosen given more wait to recent Sample RTT;
– Deviation = (1 − x)Deviation + |Sample RTT − Estimate of RTT|;
– Note 1: small deviations of Sample RTT: Timeout > Estimate of RTT;
– Note 2: big deviation of Sample RTT: Timeout ≈ Estimate of RTT.

Lecture: Transport layer protocols for ad hoc networks 8

Ad hoc networks D.Moltchanov, TUT, Fall 2009

3.1. Reliable in-order data delivery

Reliable in-order data transfer of TCP ensures the following:
• data stream in the receiver’s buffer is uncorrupted;
• data stream in the receiver’s buffer is in order (no gaps, no duplications).
TCP at the client:
• frames application layer data into segments;
• each time TCP releases a segment to IP layer a timer starts;
• the following three cases are possible here:
– if the timer expires and no ACK is received, the TCP retransmits a segment;
– if ACK is received the TCP should check:
∗ this is a first-time ACK:
all data up to the acknowledged byte have received correctly and in order.
∗ this is a duplicate ACK (DUPACK):
some data have been received out of order, retransmit the segment.

Lecture: Transport layer protocols for ad hoc networks 9

Ad hoc networks D.Moltchanov, TUT, Fall 2009

3.2. Flow control

rcvBuf

data from IP reading

Available space TCP data in the buffer

rcvWin

Figure 4: The receive window (rcvWin) at the receive buffer (rcvBuf).

Flow control service is a rate matching provided by TCP using the dynamic rcvWin:
• rcvWin is sent with every ACK to the sender;
• initially, rcvWin = rcvBuf;
• for next ACK the receiver sends: rcvWin = rcvBuf − ( lastByteRcv − lastByteRead );
• the sender must ensure that ( lastByteSent − lastByteAcked ) ≤ rcvWin to not overflow.

Lecture: Transport layer protocols for ad hoc networks 10

Ad hoc networks D.Moltchanov, TUT, Fall 2009

3.3. TCP Tahoe congestion control

Every transmission starts with connection setup and followed by slow start phase:
• the sender starts the session with a congestion window set to maximum segment size (MSS):
– it sends MSS bytes of data;
– starts retransmission timeout (RTO) and waits for acknowledgement packet (ACK).
• if ACK is received in RTO the congestion window is doubled and two MSSs of data are sent;
• the congestions window is doubled with every ACK until it reaches slow-start threshold;
Congestion window size, MSSs

16

1
time

Figure 5: Slow-start phase in TCP.

Lecture: Transport layer protocols for ad hoc networks 11

Ad hoc networks D.Moltchanov, TUT, Fall 2009
The slow-start phase is followed by congestion avoidance phase:
• when the slow-start threshold is reached, the congestion window grows linearly (AI);
• if the ACKs are received before timers (RTOs) expire:
– the congestion window grows until the receiver window advertised in connection setup;
– (lastByteSent − lastByteAcked) ≤ rcvWin to NOT allow overflow!
Congestion window size, MSSs

16

1
time

Figure 6: Congestion avoidance phase in TCP.

Lecture: Transport layer protocols for ad hoc networks 12

Ad hoc networks D.Moltchanov, TUT, Fall 2009
If the ACK is not received in RTO, TCP assumes the packets is lost (congestion):
• TCP enters the congestion control phase performing the following:
– reduces the slow-start threshold to 1/2 of current CW (MD);
– resets the congestion window to one MSS;
– activates the slow-start algorithm and resets the timeout.
Congestion window size, MSSs

18
16

1
time

Figure 7: Congestion control phase in TCP.

Lecture: Transport layer protocols for ad hoc networks 13

Ad hoc networks D.Moltchanov, TUT, Fall 2009

3.4. TCP Tahoe with fast retransmit

TCP Tahoe uses the fast retransmit procedure to respond to losses:
• if the sender receives three out-of-order segment with higher than expected seq. number;
• resend the out-of-order segment before timeout expires and enter slow start phase.
Sender Receiver

out of order
Segment 1

DUPACK
timeout Segment 2

DUPACK
Segment 3

DUPACK
Segment 0

Figure 8: Fast retransmit procedure of TCP Tahoe.

Lecture: Transport layer protocols for ad hoc networks 14

Ad hoc networks D.Moltchanov, TUT, Fall 2009

3.5. TCP Reno

DUPACKs are not indication of severe congestion. On arrival of three DUPACKs:
• TCP Reno enters the fast recovery phase and performs the following actions;
– retransmits the lost segment and does not enter slow start phase;
– reduces the slow-start threshold to 1/2 of the current CW ;
– reduces the CW to a 1/2 of the current CW + 3.
• increases the CW linearly with reception of subsequent DUPACKs;
– one MSS per a DUPACK (meaning that one more packet left the network).
• on reception of ACK (this ACK is due to retransmission) the sender:
– resets the CW with the slow-start threshold;
– enters the congestion avoidance similar to TCP Tahoe.
More improvements are available:
• TCP NewReno;
• TCP SACK.

Lecture: Transport layer protocols for ad hoc networks 15

Ad hoc networks D.Moltchanov, TUT, Fall 2009

Congestion window size, MSSs

18
16

TCP Reno

4 TCP Tahoe

1
time

Figure 9: Fast recovery procedure of TCP Reno.

Lecture: Transport layer protocols for ad hoc networks 16

Ad hoc networks D.Moltchanov, TUT, Fall 2009

4. Why does not TCP work well in ad-hoc networks?

The following reasons are behind the poor performance of TCP in ad-hoc networks:
• Misinterpretation of packet loss:
– wired networks: packets losses are mainly due to congestions;
– ad-hoc wireless networks: high packet loss due to:
∗ high BER (wired link < 10E − 9, wireless link: 10E − 5 ∼ 10E − 4 and even higher);
∗ collisions due to hidden terminal problem;
∗ interference: wired links are well isolated, wireless links interfere;
∗ frequent path breaks: wired: due to failures, wireless: mobility;
∗ large-scale and small-scale propagation phenomenons.
• Frequent path breaks:
– Frequent path path breaks → topology changes → route reconfigurations;
– RouteReconf. = f (Number of node, Tx, topology, bandwidth, traffic, nature of routing);
– If RouteRecong. > RTO: TCP enters a slow start phase;
– Slow start: inefficient use of the resources.

Lecture: Transport layer protocols for ad hoc networks 17

 Ad hoc networks D.Moltchanov, TUT, Fall 2009
• Effects of contention (dependency on the path length):
With the increase in the number of hops in the path throughput decreases exponentially.

Throughput

2 4 6 8 Number of hops

Figure 10: Throughput as a function of the number of hops.

What are the reasons? The probability of the path break is given as:

pb = 1 − (1 − pl )k , (1)

where pl is the probability of the link break.

Lecture: Transport layer protocols for ad hoc networks 18

Ad hoc networks D.Moltchanov, TUT, Fall 2009

Pr of path break
1

0.5

2 4 6 8 10 12 14 Number of hops

Figure 11: Probability of path break as a function of the number of hops.

• Misinterpretation of congestion window:

CW is the rate that is acceptable for the network and the receiver.
– old path: CW reflects its rates;
– reconfiguration occurs: what are new CW and RTO?
– new path: old CW may not reflect acceptable rates.

Lecture: Transport layer protocols for ad hoc networks 19

Ad hoc networks D.Moltchanov, TUT, Fall 2009
• Asymmetric link behavior:
Sometimes wireless links are directional in ad-hoc networks leading to:
– delivery of a packet to a node and failure in the delivery of ACK back;
– congestion control algorithm could be invoked due to this;
– some routing protocols require the forward and backward paths to be the same.
• Resource contention:
– both DATA and ACK require RTS-CTS-DATA-ACK at the data-link layer;
– contention for resources in the same link at forward and backward paths;
– this contention may not be the same.
• Multipath routing:
Some routing protocols use multiple paths between the source and destination leading to:
– high number of out-of-order packets leading to DUPACKs;
– different RTO values leading to unnecessary retransmissions.

Lecture: Transport layer protocols for ad hoc networks 20

Ad hoc networks D.Moltchanov, TUT, Fall 2009
• Network partitioning and merging:

Figure 12: Partitioning and merging in ad-hoc networks.

Lecture: Transport layer protocols for ad hoc networks 21

Ad hoc networks D.Moltchanov, TUT, Fall 2009

5. TCP enhancements
5.1. Feedback-based TCP (TCP-F)
TCP-F requires the following to enhance performance:
• support of reliable data-link layer protocols;
• routing support to inform the TCP sender about path breaks;
• routing protocol is expected to repair the broken path within a reasonable time.
The aim of TCP-F: minimize the throughput degradation resulting from path breaks.

Sender
Receiver Receiver

Figure 13: Link break in ad-hoc network.

Lecture: Transport layer protocols for ad hoc networks 22

Ad hoc networks D.Moltchanov, TUT, Fall 2009
In TCP-F an intermediate node upon detection of the link break:
• obtains information from TCP-F sender’s packets routed via this node;
• generates a route failure notification (RFN) packet;
• routes this packet to the TCP-F sender;
• does not forward any packet from this connection;
• updates its routing table;
• stores information about generation of a RFN packet.

Any intermediate node that forwards the RFN packet:

• if this node has an alternative route to destination:
– discards the RFN packet and uses this path to forward other packets:
∗ this allows to reduce an overhead involved in route re-establishment!
• if this node does not alternate route to destination:
– updates its routing table and forwards the RFN packet to the source.

Lecture: Transport layer protocols for ad hoc networks 23

Ad hoc networks D.Moltchanov, TUT, Fall 2009
When TCP-F sender receives the RFN packet it enters the so-called snooze state:
• stops sending packet to the destination;
• cancels all the timers;
• freezes the congestion window;
• sets up a route failure timer = f (routing protocol, network size, network dynamic):
• when failure timer expires TCP-F enters the connected state.

If the broken links rejoins or intermediate node obtains a new path to destination:
• route reestablishment notification (RRN) is sent to TCP-F sender;

When the sender receives RRN packet:

• reactivates all timers and congestion window assuming that the network is back;
• starts transmitting data available in the buffer;
• takes care of packets lost due to path break.

Lecture: Transport layer protocols for ad hoc networks 24

Ad hoc networks D.Moltchanov, TUT, Fall 2009

Sender (connected)

A B C D

Sender (from connected to snooze)

A B C D
RFN RFN

Sender (from snooze to connected)

A B C D
RRN RRN

Figure 14: Operation of TCP-F.

Lecture: Transport layer protocols for ad hoc networks 25

Ad hoc networks D.Moltchanov, TUT, Fall 2009
The following advantages and shortcomings are inherent for TCP-F:
• +: provides simple feedback to minimize problems due to link breaks;
• +: still allows congestion control occurring due to buffer overflows;
• −: requires merging of transport and network layer features (at least, cross-layering);
• −: requires ability of nodes to detect path breaks;
• −: requires ability of routing protocols to repair a link within a reasonable time;
• −: requires ability of node to determine the TCP-F sender;
• −: reactivated congestion window may not reflect allowed network rate.

TCP-F End-to-end transport

IP layer (routing) Routing updates

Data-link layer (MAC part) Detection of link failures

Figure 15: Communications in the protocol stack as required by TCP-F.

Lecture: Transport layer protocols for ad hoc networks 26

Ad hoc networks D.Moltchanov, TUT, Fall 2009

5.2. TCP with explicit link failure notification (TCP-ELFN)

According to TCP-ELFN an explicit link failure notification is used.
When an intermediate node detects a link failure:
• sends an explicit link failure notification (ELFN) to TCP-ELFN sender:
– either sending an ICMP destination unreachable message (DUR);
– or inserting info regarding link break in RouteError message of the routing protocol.

Once the TCP-ELFN sender receives the ELFN packet:

• it disables its retransmission timer and CW ;
• enters a standby state.

Being in standby state the TCP-ELFN sender:

• periodically originates probe packets to see if a new route is established;
• when ACK for a probe packet is received TCP-ELFN continues to perform as usual.

Lecture: Transport layer protocols for ad hoc networks 27

Ad hoc networks D.Moltchanov, TUT, Fall 2009

Sender (connected)

A B C D

Sender (standby)

A B C D
ICMP(DUP), RouteError
Sender (standby)

A B C D
probe probe
Sender (connected)

A B C D
routeOK routeOK

Figure 16: Example of TCP-ELFN.

• +: provides path break information to the sender;

• +: does not heavily depend on routing protocol capabilities;
• −: periodic packets consume bandwidth, sometimes it may not help (partitioning);
• −: reactivated congestion window may not reflect the new allowed network rate.

Lecture: Transport layer protocols for ad hoc networks 28

Ad hoc networks D.Moltchanov, TUT, Fall 2009

5.3. TCP with buffering and sequence information (TCP-BuS)

TCP-BuS is characterized by the following:
• protocol tries to notify the source about the path breaks using the feedback info;
• this protocol is more dependent on routing protocol compared to TCP-F and TCP-ELFN.
TCP-BuS was proposed for usage with associativity-based routing (ABR) and uses:
• localized query (LQ) message of ABR;
• REPLY message of ABR.
Both these messages are modified to carry TCP connection and segment information.

Sender intermediate upstream Receiver

A B C D E
intermediate downstream

Figure 17: Basic definitions for TCP-BuS protocol.

Lecture: Transport layer protocols for ad hoc networks 29

Ad hoc networks D.Moltchanov, TUT, Fall 2009
When a link break is detected, intermediate downstream node:
• generates a route notification (RN) message to TCP-BuS receiver:
– RN includes the sequence number of packet belonging to that flow in the head of its queue;
• all packets belonging to this flow are discarded at all intermediate nodes that forward RN.

When a link break is detected, intermediate upstream node:

• generate explicit route disconnection notification (ERDN);
• when ERDN is received by the sender, it stops sending and freezes timers CW ;
• all packets in transit nodes are buffered, till new partial path is found by source of ERDN;
• tries to find a new (partial) route to the TCP-BuS receiver;
• if it finds, explicit route successful notification packet (ERSN) to the sender is sent.

When ERDN is received the following is done at the sender:

• comparing sequence numbers segments are selectively retransmitted.

Lecture: Transport layer protocols for ad hoc networks 30

Ad hoc networks D.Moltchanov, TUT, Fall 2009

TCP-BuS receiver
12 RN TCP flow

1 LQ
5
6 REPLY
2
8
10
ERDN
3
9

11 ERDN
4 7
TCP-BuS sender

Figure 18: Operation of TCP-BuS connection.

Lecture: Transport layer protocols for ad hoc networks 31

Ad hoc networks D.Moltchanov, TUT, Fall 2009
The following advantages and shortcomings are inherent for TCP-BuS:
• +: avoidance of retransmissions due to buffering, usage of sequence numbers;
• +: selective retransmission improves the performance of TCP;
• −: dependency on the routing protocol (ABR);
• −: buffering at intermediate nodes (they could be overflowed, thus, may fail).

5.4. Ad hoc TCP

This is a wise realization of TCP with feedback due to:
• usage of explicit congestion notification field (currently is under consideration in IETF);
• usage of intermediate slim layer between traditional TCP and IP layer;
• only a few functions;
• does not require a certain routing protocol;
• adaptation of congestion window to a new path.
Aim is the same: to treat path breaks independently from congestion situations.

Lecture: Transport layer protocols for ad hoc networks 32

Ad hoc networks D.Moltchanov, TUT, Fall 2009

Data

TCP tcp_input() tcp_output()

ATCP atcp_input() atcp_output()

IP ip_input() ip_output()

Data

Figure 19: TCP/IP layered model featuring ATCP.

The following are advantages of such a layered structure:

• ATCP logic is separated from classic TCP;
• no changes to TCP are required;
• in principle, ATCP can be realized and installed separately from TCP;
• simple primitives can be defined between ATCP and TCP layers.

Lecture: Transport layer protocols for ad hoc networks 33

Ad hoc networks D.Moltchanov, TUT, Fall 2009

retransmits a DISCONN DUR: destination unreachable

segment from
a buffer Tx: packet transmission
DUPACK/ACK DUR
DUR DUR DUPACK: duplicate ACK
NORMAL

3 DUPACKs Tx

ACK ECN
ECN CONGESTED
LOSS

Figure 20: ATCP state transition diagram.

• when a TCP connection is established, ATCP enters the NORMAL state;

• in the NORMAL state ATCP does not interfere with classic TCP and remains invisible.

Lecture: Transport layer protocols for ad hoc networks 34

Ad hoc networks D.Moltchanov, TUT, Fall 2009
Consider the case when packets are lost or arrive out-of-order at destination:
• the receiver generate DUPACKs:
• Traditional TCP (e.g. Tahoe):
– the sender retransmits the segment;
– decrease its congestion window accordingly.
• ATCP:
– counts, waits if the number of DUPACKs reaches three;
– if so, it puts ATCP in the LOSS state and does not invoke the congestion control;
– in the LOSS state ATCP retransmits unacked segments from the sender’s buffer.

When ATCP is in the LOSS state:

• if new ACK is received from the receiver ATCP enters the NORMAL state;
• if ECN is received it enters the CONGESTED state;
• if ICMP DUR is received it enters the DISCONNECTED state.

Lecture: Transport layer protocols for ad hoc networks 35

 Ad hoc networks D.Moltchanov, TUT, Fall 2009
When ATCP is in the NORMAL state:
• three DUPACKs are received ATCP enters goes to LOSS state;
• if ECN is received it enters the CONGESTED state and remains invisible for TCP;
• if ICMP DUP is received it enters the DISCONNECTED state.

When ATCP is in the DISCONNECTED state:

• decrease the congestion window to one;
• remains in this state;
• if either ACK or DUPACK is received, it goes into NORMAL state.
It is expected that the new route is found and the source is informed.

When ATCP is in the CONGESTED state:

• if ICMP DUR is received it goes into DISCONNECTED state.

Lecture: Transport layer protocols for ad hoc networks 36

Ad hoc networks D.Moltchanov, TUT, Fall 2009
In general, the following is made in ATCP:
• packet loss due to high BER:
– retransmits lost packet
• route reconfiguration delay and transient partitions:
– stops transmission until the new route has been found
• out-of-order packet delivery:
– makes classic TCP unaware of it and retransmits the packet from TCP buffer
• change in route:
– recomputes the congestion window
The following advantages and shortcomings are inherent:
• +: compatible with traditional TCP;
• +: maintains the end-to-end semantics of TCP;
• −: requires support from routing protocol (route changes, partition detection);
• −: requires changes to interface functions.

Lecture: Transport layer protocols for ad hoc networks 37

Ad hoc networks D.Moltchanov, TUT, Fall 2009

5.5. Split TCP

The following are two major problems with TCP:
• degradation of throughput with increase in the path length:
Short connections obtain more throughput than long connections.
Throughput

2 4 6 8 Number of hops

Figure 21: Throughput as a function of the number of hops.

Lecture: Transport layer protocols for ad hoc networks 38

Ad hoc networks D.Moltchanov, TUT, Fall 2009
• unfairness among TCP flows:
– MAC layer contention (IEEE 802.11 MAC: channel capture effect):
– lengthy TCP flows – more points to contend.

Figure 22: Unfairness in TCP flows in ad-hoc networks.

Lecture: Transport layer protocols for ad hoc networks 39

Ad hoc networks D.Moltchanov, TUT, Fall 2009
Split-TCP provides the solution by splitting the TCP functionalities into two aims:
• congestion control;
• end-to-end reliability.
Why it is possible?
• congestion control: local phenomenon due to high contention for resources;
• end-to-end reliability: end-to-end phenomenon.

End-to-end delivery Congestions

Figure 23: Congestion control and end-to-end reliability.

Lecture: Transport layer protocols for ad hoc networks 40

Ad hoc networks D.Moltchanov, TUT, Fall 2009
Split-TCP splits the connection into a set of concatenated TCP connections.

segment 1 (zone 1) S
segment 2 (zone 2)
proxy node

Figure 24: Splitting of the TCP connection into segments (zones).

Proxy node is responsible for:

• terminating the connection from the sender/precessor proxy node;
• setting up a connection with receiver/successor node.

Lecture: Transport layer protocols for ad hoc networks 41

Ad hoc networks D.Moltchanov, TUT, Fall 2009
Proxy nodes are chosen using the distributed algorithm:
• simplest way: packet traversed n hops - behave as a proxy.

Transmission control at the TCP sender window is split into:

• end-to-end CW :
– it is updated according to arrival of end-to-end ACKs.
• CW : (CW ≤ end-to-end CW )
– it is updated according to arrival of local ACKs (LACKs) from the next node.

The proxy node behaves as follows:

• it maintains CW that governs transmission in a segment;
• when packet arrives from predecessor the LACK is sent back;
• arrived packet is buffered;
• the buffered packet is forwarded to the next node.

Lecture: Transport layer protocols for ad hoc networks 42

Ad hoc networks D.Moltchanov, TUT, Fall 2009

segment 1 (zone 1) S
segment 2 (zone 2)
proxy node
TCP data flow; end-to-end ACK;
LACK; network link.

Figure 25: Flows in split-TCP.

Lecture: Transport layer protocols for ad hoc networks 43

Ad hoc networks D.Moltchanov, TUT, Fall 2009
The following advantages and shortcomings are inherent for split-TCP:
• +: improved throughput: reduction in the path length;
• +: improved throughput fairness: each segment works at the most suitable rate;
• −: requires modifications to TCP;
• −: the end-to-end connection handling is violated;
• −: the failure of proxy nodes may lead to throughput degradation;
• −: security encryption schemes may not work (intermediate nodes have to process).
• −: high resource consumption (buffer space).

Lecture: Transport layer protocols for ad hoc networks 44

Ad hoc networks D.Moltchanov, TUT, Fall 2009

Table 1: Comparison of TCP solutions for ad-hoc networks.

Issue TCP-F TCP-ELFN TCP-BuS ATCP Split-TCP

loss due to BER TCP TCP TCP no cong. cntrl TCP

path break TCP snoozes TCP standby TCP snoozes TCP TCP

out-of-order pckts TCP TCP resend reordering TCP

congestion TCP TCP TCP ECN notif. TCP

path break notif. yes yes yes yes no

path reest. notif. yes no yes no no

depend. on routing. yes yes yes yes no

end-to-end semant. yes yes yes yes no

buffer at inter. nodes no no yes no yes

Lecture: Transport layer protocols for ad hoc networks 45

Ad hoc networks D.Moltchanov, TUT, Fall 2009

6. Other transport layer protocols

TCP was not developed for wireless links. What are advantageous features of TCP:
• end-to-end reliability;
• congestion control due to high traffic contention;
• end-to-end flow control.

What are reasons to deploy TCP in ad-hoc networks:

• HUGE INSTALLATION BASE;
• TCP is not a good candidate for wireless mobile ad-hoc environment.

New protocol:
• could be more suitable than TCP for ad-hoc networks and wireless links;
• could be useful in military/medical or other specific applications.

Lecture: Transport layer protocols for ad hoc networks 46

Ad hoc networks D.Moltchanov, TUT, Fall 2009

6.1. Application controlled transport protocol (ACTP)

ACTP is a light-weight transport layer protocol:
• TCP:
High reliability → a lot of retransmissions → low throughput.
• UDP:
Is not reliable at all → a lot of data just lost → low reliability.
• ACTP:
Is between TCP and UDP → some lost data may be recovered → higher performance.

ACTP is characterized by the following:

• it is responsible for feedback to end application;
• end-to-end reliability is left to applications;
• priorities of packet are implemented;
• implementation of priorities is left to lower layers.

Lecture: Transport layer protocols for ad hoc networks 47

Ad hoc networks D.Moltchanov, TUT, Fall 2009

Application layer sendTo(delay,message,priority)

ACTP layer IPoutput() isAcked()

IP

Figure 26: Interface function used in ACTP is shown here.

Each data to be transmitted contains the following information:

• delay (a maximum tolerable delay);
• number of the packet;
• priority of the packet.
The delivery status is maintained at ACTP and available to application via isAcked():
• successful delivery of the packet (ACK was returned);
• possible loss of the packet (ACK is not returned within a deadline);
• remaining time for the packet (ACK is not returned but the deadline has not expired);
• no information available.

Lecture: Transport layer protocols for ad hoc networks 48

Ad hoc networks D.Moltchanov, TUT, Fall 2009
The following peculiarities are used in ACTP:
• packet lifetime 4 × RT O;
• packet can be retransmitted on a request from application layer;
• delay 0 – high priority packet;
• packet without a message number is handled as classic UDP packet.

The following advantages and shortcomings are inherent for ACTP:

• +: scalable for large networks (light-weight);
• +: allows to set priorities to data;
• +: no congestion window;
• −: congestions are possible;
• −: is not compatible with TCP;
• −: suitable for particular applications only.

Lecture: Transport layer protocols for ad hoc networks 49

Ad hoc networks D.Moltchanov, TUT, Fall 2009

6.2. Ad-hoc transport protocol (ATP)

The following are the major difference between ATP and TCP:
• ATP uses rate based transmission;
• ATP separates congestion control and reliability:
– network congestion information is obtained from intermediate nodes;
– flow control and reliability information is obtained from ATP receiver.
• ATP uses assisted congestion control;
• ATP uses selective ACKs (SACKs).

ATP uses the information available from underlying layers for:

• estimation of the initial transmission rate;
• detection, avoidance and control of congestion;
• detection of path breaks.

Lecture: Transport layer protocols for ad hoc networks 50

Ad hoc networks D.Moltchanov, TUT, Fall 2009
The congestion information obtained from intermediate node is expressed in terms:
• weighted average queuing delay DQ :

DQ = xDQnew + (1 − x)DQold , (2)

• contention delay DC .

During the connection setup phase or when ATP recovers from path break:
• ATP sender determines the transmission rate sending probe packets (quick start phase);
– intermediate node attaches the rate info in form of DC and DQ ;
– receiver responds with ACK.

At the uplink path the following always occurs:

• DQ and DC at each intermediate node is included in rate feedback field (ABR);
• ATP receiver collects DQ and DC and includes them in periodic SACKs.

Lecture: Transport layer protocols for ad hoc networks 51

Ad hoc networks D.Moltchanov, TUT, Fall 2009
When congestion occurs TCP uses:
• decrease of the CW .

Instead, ATP defines three phases:

• increase:
– If new rate (R) is higher than the current rate and beyond the threshold γ then:

R − Sold
Snew = Sold + , (3)
k
is used (increased) where k is the fraction used to avoid rapid fluctuations.
• decrease:
– If new rate is lower than current the rate is decreased to a new rate.
• maintain:
– if new rate is higher that current but less than the threshold.

Lecture: Transport layer protocols for ad hoc networks 52

Ad hoc networks D.Moltchanov, TUT, Fall 2009
Consider special cases:
• ATP sender has not received ACK for two consecutive feedback periods:
– it undergoes multiplicative decrease of the transmission rate.
• ATP sender has not received ACK for three consecutive feedback periods:
– it assumes that the connection is lost and enters the connection establishment phase.
• When path break occurs:
– the network layer detects it and originated ELFN packet toward ATP sender;
– the ATP sender freezes the sender state and enters the connection establishment phase;
– if the connection is established the sender begins transmission again.

The following advantages and shortcomings are inherent for ATP:

• +: improved performance in ad hoc network:
– decoupling of congestion control and reliability;
– avoiding fluctuations of the congestion window.
• −: lack of interoperability with TCP.

Lecture: Transport layer protocols for ad hoc networks 53