Chapter 3: Transport Layer

Chapter 3: Transport Layer

Chapter 3: Transport Layer
our goals: understand principles behind transport layer services: multiplexing, demultiplexing reliable data transfer flow control congestion control learn about Internet transport layer protocols: UDP: connectionless transport TCP: connection-oriented reliable transport TCP congestion control Transport Layer

Chapter 3 outline 3.1 transport-layer services
3.2 multiplexing and demultiplexing 3.3 connectionless transport: UDP 3.4 principles of reliable data transfer 3.5 connection-oriented transport: TCP segment structure reliable data transfer flow control connection management 3.6 principles of congestion control 3.7 TCP congestion control Transport Layer

Transport services and protocols
provide logical (instead of physical!) communication between app processes running on different hosts transport protocols run in end systems send side: breaks app messages into segments, passes to network layer rcv side: reassembles segments into messages, passes to app layer more than one transport protocol available to apps Internet: TCP and UDP application transport network data link physical logical end-end transport abstract from the concrete path in physical network application transport network data link physical Transport Layer

Transport vs. network layer
household analogy: network layer: logical communication between hosts transport layer: logical communication between processes relies on & enhances network layer services 12 kids in Ann’s house sending letters to 12 kids in Bill’s house: hosts = houses processes = kids app messages = letters in envelopes transport protocol = Ann and Bill collect from/distribute to in-house siblings each week => give to/get from postal service mail carrier network-layer protocol = postal service Ann and Bill give the mail to a postal-service mail carrier (who makes daily visits to the house) Transport Layer

Multiplexing/demultiplexing
use header info to deliver received segments to correct socket (Bill receives batch of mails from carrier and delivers to his brothers and sister) demultiplexing at receiver: handle data from multiple sockets, add transport header (later used for demultiplexing) multiplexing at sender: application application P1 P2 application socket P3 transport P4 process transport network transport where to put the segments? link network network physical link link physical physical Transport Layer

How demultiplexing works
Remark: we call packets of the network layer DATAGRAMS and packets of the transport layer are called SEGMENTS host receives IP datagrams each datagram has source IP address, destination IP address each datagram carries one transport-layer segment each segment has source, destination port number host uses IP addresses & port numbers to direct segment to appropriate socket 32 bits source port # dest port # other header fields application data (payload) port number between 0 and (16 bits) reserved by well-known application protocols: (called well-known port numbers) e.g. HTTP uses port 80, FTP uses 21 TCP/UDP segment format Transport Layer

Connectionless demultiplexing
recall: when creating datagram to send into UDP socket, one must specify destination IP address destination port # port number is typically assigned autom. in client app when host receives UDP segment: checks destination port # in segment directs UDP segment to socket with that port # IP datagrams with same dest. port # and same destination IP address, but different source IP addresses and/or source port numbers will be directed to same socket at destination assigned port number is between 1024 and 65535 Transport Layer

Connectionless demux: example
P1: port number 6428 P3: port number 9157 P4: port number 5775 application application application P1 P3 P4 transport transport transport network network network link link link physical physical physical UDP segm source port: 6428 dest port: 9157 UDP segm source port: 6428 dest port: 5775 UDP segm source port: 9157 dest port: 6428 UDP segm source port: 5775 dest port: 6428 Transport Layer

Connection-oriented demux
TCP socket identified by 4-tuple: source IP address source port number dest IP address dest port number demux: receiver uses all four values to direct segment to appropriate socket server host may support many simultaneous TCP sockets: each socket identified by its own 4-tuple web servers have different sockets for each connecting client non-persistent HTTP will have different socket for each request Transport Layer

Connection-oriented demux: example
application application application P4 P5 P6 P1 P2 P3 transport transport transport network network network link link link physical physical physical server: IP address B TCP segm source IP,port: B,80 dest IP,port: A,9157 host: IP address C host: IP address A TCP segm source IP,port: C,5775 dest IP,port: B,80 TCP segm source IP,port: A,9157 dest IP, port: B,80 TCP segm source IP,port: C,9157 dest IP,port: B,80 three segments, all destined to IP address: B, dest port: 80 are demultiplexed to different sockets (different source IP/source port number!); one process per connection socket Transport Layer

threaded server application application P4 application P1 P2 P3 transport transport transport network network network link link link physical physical physical server: IP address B TCP segm source IP,port: B,80 dest IP,port: A,9157 host: IP address C host: IP address A TCP segm source IP,port: C,5775 dest IP,port: B,80 TCP segm source IP,port: A,9157 dest IP, port: B,80 TCP segm source IP,port: C,9157 dest IP,port: B,80 three segments, all destined to IP address: B, dest port: 80 are demultiplexed to different sockets (different source IP/source port number!); one THREAD per connection socket Transport Layer

threaded server application application P4 application P1 P2 P3 transport transport transport network network network link link link physical physical physical server: IP address B TCP segm source IP,port: B,80 dest IP,port: A,9157 host: IP address C host: IP address A TCP segm source IP,port: C,5775 dest IP,port: B,80 TCP segm source IP,port: A,9157 dest IP, port: B,80 TCP segm source IP,port: C,9157 dest IP,port: B,80 Note: during persisent HTTP, same server socket it used during non-persistent HTTP, new TCP connection (=> new socket) for every request/response Transport Layer

UDP: User Datagram Protocol [RFC 768]
“no frills,” “bare bones” Internet transport protocol “best effort” service, UDP segments may be: lost delivered out-of-order to app connectionless: no handshaking between UDP sender, receiver each UDP segment handled independently of others UDP use: streaming multimedia (loss tolerant, rate sensitive) DNS Internet phone real-time video conferencing reliable transfer over UDP: add reliability at application layer application-specific error recovery! Transport Layer

UDP: segment header why is there a UDP?
length: in bytes of UDP segment, including header; max. 16 bits => length of data field <2^16=65536 bytes (minus 8 bytes (UDP header), 20 bytes (IP header – not yet prepended!) 4 x 16 bits = 4 x 2 bytes source port # dest port # length checksum why is there a UDP? no connection establishment which can add delay (e.g. important for DNS) simple: no connection state at sender, receiver (server can support more active clients) small header size (UDP: 8 bytes, TCP: 20 bytes) no congestion control: UDP can blast away as fast as desired application data (payload) = 65508 IP datagram has also a length field of 16 bits => same restriction on size UDP segment format Transport Layer

UDP checksum sender: receiver:
Goal: detect “errors” (e.g., flipped bits) in transmitted segment sender: add IP pseudo header (transport layer must inform network layer about destination => certain information known); later ‘real’ IP header is generated treat segment contents, including header fields, as sequence of 16-bit integers checksum: addition of segment contents and one’s complement sender puts checksum value into UDP checksum field receiver: compute checksum of received segment check if computed checksum equals checksum field value: NO - error detected => pass damaged segment to app with warning or discard it YES - no error detected. But maybe errors nonetheless? More later …. Transport Layer

Internet checksum: example
example: add two 16-bit integers wraparound sum checksum wrap the carry and add it to low order bit Note: when adding numbers, a carryout from the most significant bit needs to be added to the result RFC 768 Transport Layer

Internet checksum Why do we have to add the carry from the most significant bit? 4-Bit example: we want to add -5 and -3 in 1’s complement arithmetic +5 is => -5 is 1010 +3 is => -3 is 1100 wrap the carry and add it to low order bit 1010 +’ 1100 = (ignoring the carry) 0110 (which is -9) = (with carry) 0111 (which is -8) Transport Layer

Internet checksum Why is the checksum calculated like this?
(efficient implementation: see RFC 1071) Why is the checksum calculated like this? Use notation [a,b] for 16-bit integer a⋅28+b, where a and b are bytes checksum computation corresponds to [A,B] +' [C,D] +' ... +' [Y,Z] where +' is1's complement addition 1’s complement addition has nice properties: - associative property: ( [A,B] +' [C,D] +' ... +' [J,0] ) +' ( [0,K] +' ... +' [Y,Z] ) wrap the carry and add it to low order bit - commutative property: ( [0,K] +' ... +' [Y,Z] ) +’ ( [A,B] +' [C,D] +' ... +' [J,0] ) - byte order independence: [B,A] +' [D,C] +' ... +' [Z,Y] - allows parallel summation Transport Layer

Reliable data transfer protocols
imagine the following (abstract) data transfer protocol: rdt_send(): called from above, (e.g., by app.). Passes data to deliver to receiver at upper layer deliver_data(): called by rdt to deliver data to upper some abstract upper layer send side receive side udt_send(): called by rdt, to transfer packet over unreliable channel to receiver rdt – reliable data transfer -> that is the protocol that we want to define in the sequel udt - unreliable data transfer rdt_rcv(): called when packet arrives on rcv-side of channel (receive function of the abstract rdt protocol) Transport Layer

Reliable data transfer: getting started
we’ll: incrementally develop sender and receiver sides of reliable data transfer protocol (rdt) consider only unidirectional data transfer but control info will flow on both directions! use finite state machines (FSM) to specify behavior of sender and receiver event causing state transition actions taken on state transition state: when in this “state” next state uniquely determined by next event state 1 state 2 event actions Transport Layer

rdt1.0: reliable transfer over a reliable channel
underlying channel perfectly reliable no bit errors no loss of packets separate FSMs for sender and receiver: sender sends data into underlying channel receiver reads data from underlying channel Wait for call from above rdt_send(data) Wait for call from below rdt_rcv(packet) extract (packet,data) deliver_data(data) packet = make_pkt(data) udt_send(packet) sender receiver Transport Layer

rdt2.0: channel with bit errors
underlying channel may flip bits in packet checksum to detect bit errors the question: how to recover from errors: acknowledgements (ACKs): receiver explicitly tells sender that pkt received OK negative acknowledgements (NAKs): receiver explicitly tells sender that pkt had errors sender retransmits pkt on receipt of NAK new mechanisms in rdt2.0 (beyond rdt1.0): error detection receiver feedback: control msgs (ACK,NAK) rcvr->sender How do humans recover from “errors” during conversation? humans say „ok“ or „what did you say?“ „Can you repeat?“ Transport Layer

rdt2.0: channel with bit errors
underlying channel may flip bits in packet checksum to detect bit errors the question: how to recover from errors: acknowledgements (ACKs): receiver explicitly tells sender that pkt received OK negative acknowledgements (NAKs): receiver explicitly tells sender that pkt had errors sender retransmits pkt on receipt of NAK new mechanisms in rdt2.0 (beyond rdt1.0): error detection feedback: control msgs (ACK,NAK) from receiver to sender Transport Layer

rdt2.0: FSM specification
rdt_send(data) sndpkt = make_pkt(data, checksum) udt_send(sndpkt) receiver rdt_rcv(rcvpkt) && isNAK(rcvpkt) Wait for ACK or NAK Wait for call from above udt_send(NAK) rdt_rcv(rcvpkt) && corrupt(rcvpkt) udt_send(sndpkt) rdt_rcv(rcvpkt) && isACK(rcvpkt) Wait for call from below L sender rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) stop and wait sender sends one packet, then waits for receiver response extract(rcvpkt,data) deliver_data(data) udt_send(ACK) Transport Layer

rdt2.0 has a fatal flaw! handling duplicates:
what happens if ACK/NAK corrupted? sender doesn’t know what happened at receiver! can’t just retransmit: possible duplicate handling duplicates: sender retransmits current pkt if ACK/NAK corrupted sender adds sequence number to each pkt receiver discards (doesn’t deliver up) duplicate pkt Transport Layer

rdt2.1: sender handles corrupted ACK/NAKs
sequence numbers 0 and 1 rdt_send(data) sndpkt = make_pkt(0, data, checksum) udt_send(sndpkt) rdt_rcv(rcvpkt) && ( corrupt(rcvpkt) || isNAK(rcvpkt) ) Wait for ACK or NAK 0 Wait for call 0 from above udt_send(sndpkt) rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) && isACK(rcvpkt) rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) && isACK(rcvpkt) L L Wait for ACK or NAK 1 Wait for call 1 from above rdt_rcv(rcvpkt) && ( corrupt(rcvpkt) || isNAK(rcvpkt) ) rdt_send(data) sndpkt = make_pkt(1, data, checksum) udt_send(sndpkt) udt_send(sndpkt) Transport Layer

rdt2.1: receiver, handles garbled ACK/NAKs
rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) && has_seq0(rcvpkt) extract(rcvpkt,data) deliver_data(data) sndpkt = make_pkt(ACK, chksum) udt_send(sndpkt) rdt_rcv(rcvpkt) && (corrupt(rcvpkt) rdt_rcv(rcvpkt) && (corrupt(rcvpkt) sndpkt = make_pkt(NAK, chksum) udt_send(sndpkt) sndpkt = make_pkt(NAK, chksum) udt_send(sndpkt) Wait for 0 from below Wait for 1 from below rdt_rcv(rcvpkt) && not corrupt(rcvpkt) && has_seq1(rcvpkt) rdt_rcv(rcvpkt) && not corrupt(rcvpkt) && has_seq0(rcvpkt) sndpkt = make_pkt(ACK, chksum) udt_send(sndpkt) sndpkt = make_pkt(ACK, chksum) udt_send(sndpkt) rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) && has_seq1(rcvpkt) extract(rcvpkt,data) deliver_data(data) sndpkt = make_pkt(ACK, chksum) udt_send(sndpkt) Transport Layer

rdt2.1: discussion sender: seq # added to pkt receiver:
two seq. #’s (0,1) will suffice since receiver can distinguish detect retransmission must check if received ACK/NAK corrupted twice as many states state must “remember” whether “expected” ACK is for seq # of 0 or 1 receiver: must check if received packet is duplicate state indicates whether 0 or 1 is expected pkt seq # note: receiver can not know if its last ACK/NAK received OK at sender two numbers suffice because at receiver side: if paket has same number as previously transmitted paket then RETRANSMISSION otherwise: NEW PAKET Transport Layer

rdt2.2: a NAK-free protocol
same functionality as rdt2.1, using ACKs only instead of NAK, receiver sends ACK for last packet received OK (=> may ACK same paket several times) receiver must explicitly include seq # of packet being ACKed duplicate ACK at sender results in same action as NAK: retransmit current packet if receiver gets corrupted paket, he sends an ACK with DIFFERENT paket number (number of previous paket) Transport Layer

rdt2.2: sender, receiver fragments
Wait for call 0 from above sndpkt = make_pkt(0, data, checksum) udt_send(sndpkt) rdt_send(data) rdt_rcv(rcvpkt) && ( corrupt(rcvpkt) || isACK(rcvpkt,1) ) rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) && isACK(rcvpkt,0) Wait for ACK sender FSM fragment rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) && has_seq1(rcvpkt) extract(rcvpkt,data) deliver_data(data) sndpkt = make_pkt(ACK1, chksum) udt_send(sndpkt) Wait for 0 from below rdt_rcv(rcvpkt) && (corrupt(rcvpkt) || has_seq1(rcvpkt)) receiver FSM fragment L acknowledgement now includes paket number Transport Layer

rdt3.0: channels with errors and loss
new assumption: underlying channel can also lose packets (data, ACKs) checksum, seq. #, ACKs, retransmissions will be of help … but not enough approach: sender waits “reasonable” amount of time for ACK retransmits if no ACK received in this time if pkt (or ACK) just delayed (not lost): retransmission will be duplicate, but seq. #’s already handles this receiver must specify seq # of pkt being ACKed requires countdown timer Transport Layer

rdt3.0 sender receiver FSM is homework (do it at home and compare
rdt_send(data) rdt_rcv(rcvpkt) && ( corrupt(rcvpkt) || isACK(rcvpkt,1) ) sndpkt = make_pkt(0, data, checksum) udt_send(sndpkt) start_timer rdt_rcv(rcvpkt) L L Wait for call 0from above Wait for ACK0 timeout udt_send(sndpkt) start_timer rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) && isACK(rcvpkt,1) rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) && isACK(rcvpkt,0) stop_timer stop_timer Wait for ACK1 Wait for call 1 from above timeout udt_send(sndpkt) start_timer if msg corrupt then do not retransmit since this already is done when timer expires receiver FSM is homework!!! rdt_rcv(rcvpkt) L rdt_send(data) rdt_rcv(rcvpkt) && ( corrupt(rcvpkt) || isACK(rcvpkt,0) ) sndpkt = make_pkt(1, data, checksum) udt_send(sndpkt) start_timer L receiver FSM is homework (do it at home and compare your solution to our solution during the tutorial) Transport Layer

rdt3.0 in action sender receiver sender receiver send pkt0 send pkt0
rcv pkt0 rcv pkt0 send ack0 send ack0 ack0 ack0 rcv ack0 rcv ack0 send pkt1 pkt1 send pkt1 pkt1 X loss rcv pkt1 ack1 send ack1 rcv ack1 send pkt0 pkt0 timeout resend pkt1 rcv pkt0 ack0 send ack0 pkt1 rcv pkt1 ack1 send ack1 rcv ack1 send pkt0 pkt0 (a) no loss rcv pkt0 ack0 send ack0 (b) packet loss Transport Layer

rdt3.0 in action (slide is repaired)
sender receiver sender receiver send pkt0 pkt0 rcv pkt0 send pkt0 pkt0 send ack0 rcv pkt0 ack0 rcv ack0 send ack0 ack0 send pkt1 pkt1 rcv ack0 rcv pkt1 send pkt1 pkt1 send ack1 rcv pkt1 ack1 ack1 X loss send ack1 timeout resend pkt1 timeout resend pkt1 pkt1 rcv pkt1 pkt1 send ack1 send pkt1 rcv ack0 pkt1 ack1 send pkt0 rcv ack1 pkt0 rcv pkt0 send ack0 ack0 rcv pkt1 rcv pkt1 (detect duplicate) (detect duplicate) ack1 send ack1 rcv ack1 send pkt0 pkt0 rcv pkt0 ack0 send ack0 (c) ACK loss (d) premature timeout/ delayed ACK Transport Layer

Performance of rdt3.0 rdt3.0 is correct, but performance is slow
e.g.: 1 Gbps link (109 bits/sec), from US west to east coast 15 ms prop. delay (RRT=30 ms=30/1000 sec), 8000 bit packet: Dtrans = L R 8000 bits 109 bits/sec = .008 ms Calculate U sender: utilization – fraction of time sender busy sending Transport Layer

rdt3.0: stop-and-wait operation
sender receiver first packet bit transmitted, t = 0 last packet bit transmitted, t = L / R first packet bit arrives RTT last packet bit arrives, send ACK assume no transmission delay for ACK ACK arrives, send next packet, t = RTT + L / R percentage of time the server is busy (with transmitting) Transport Layer

Performance of rdt3.0 rdt3.0 is correct, but performance is slow
e.g.: 1 Gbps link (109 bits/sec), from US west to east coast 15 ms prop. delay (RRT=30 ms=30/1000 sec), 8000 bit packet: Dtrans = L R 8000 bits 109 bits/sec = .008 ms U sender: utilization – fraction of time sender busy sending throughput = utilization x net bitrate = bits/sec ≈ 34kB/sec throughput over 1 Gbps link receiver has very slow download rate from that server! Transport Layer

Pipelined protocols pipelining: sender allows multiple, yet-to-be-acknowledged packets range of sequence numbers must be increased buffering at sender and/or receiver two generic forms of pipelined protocols: go-Back-N, selective repeat Transport Layer

Pipelining: increased utilization
maximal 3 unacknowledged packets in pipeline sender receiver first packet bit transmitted, t = 0 last bit transmitted, t = L / R first packet bit arrives RTT last packet bit arrives, send ACK last bit of 2nd packet arrives, send ACK last bit of 3rd packet arrives, send ACK ACK arrives, send next packet, t = RTT + L / R check out GBN applet at With Safari browser WAIT (!!!) FOR RETRANSMISSION (do not click new send all the time!) 3-packet pipelining increases utilization by a factor of 3! Transport Layer

Pipelined protocols: overview
Go-back-N: sender can have up to N unack’ ed packets in pipeline receiver only sends ‘cumulative’ ack doesn’t ack packet if there’s a gap sender has timer for oldest unack’ed packet when timer expires, retransmit all unacked packets Selective Repeat: sender can have up to N unack’ed packets in pipeline rcvr sends individual ack for each packet sender maintains timer for each unacked packet when timer expires, retransmit only that unacked packet Transport Layer

Go-Back-N protocol: sender
k-bit seq # in packet header “window” of up to N, consecutive unack’ed packets allowed ACK(n): ACKs all pkts up to, including seq # n - “cumulative ACK” may receive duplicate ACKs (see receiver) monitor timer for oldest in-flight pkt timeout(n): retransmit packet n and all higher seq # pkts in window Transport Layer

GBN: sender extended FSM (slide is repaired)
rdt_send(data) if (nextseqnum < base+N) { sndpkt[nextseqnum] = make_pkt(nextseqnum,data,chksum) udt_send(sndpkt[nextseqnum]) start_timer(nextseqnum) nextseqnum++ } else refuse_data(data) L base=1 nextseqnum=1 Timeout(base) restart_timer(base) udt_send(sndpkt[base]) restart_timer(base+1) udt_send(sndpkt[base+1]) … restart_timer(nextseqnum-1) udt_send(sndpkt[nextseqnum-1]) Wait rdt_rcv(rcvpkt) && corrupt(rcvpkt) L rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) base = getacknum(rcvpkt)+1 (drop all timers with # < base) Transport Layer

GBN: receiver extended FSM
default udt_send(sndpkt) rdt_rcv(rcvpkt) && notcurrupt(rcvpkt) && hasseqnum(rcvpkt,expectedseqnum) L Wait expectedseqnum=1 sndpkt = make_pkt(0,ACK,chksum) extract(rcvpkt,data) deliver_data(data) sndpkt = make_pkt(expectedseqnum,ACK,chksum) udt_send(sndpkt) expectedseqnum++ Transport Layer

default udt_send(sndpkt) rdt_rcv(rcvpkt) && notcurrupt(rcvpkt) && hasseqnum(rcvpkt,expectedseqnum) L Wait expectedseqnum=1 sndpkt = make_pkt(0,ACK,chksum) extract(rcvpkt,data) deliver_data(data) sndpkt = make_pkt(expectedseqnum,ACK,chksum) udt_send(sndpkt) expectedseqnum++ initialize: counter for expected sequence number make initial packet ready for sending but use seqnum 0 in case that 1st packet is corrupt or has wrong number Transport Layer

default udt_send(sndpkt) rdt_rcv(rcvpkt) && notcorrupt(rcvpkt) && hasseqnum(rcvpkt,expectedseqnum) L Wait expectedseqnum=1 sndpkt = make_pkt(0,ACK,chksum) extract(rcvpkt,data) deliver_data(data) sndpkt = make_pkt(expectedseqnum,ACK,chksum) udt_send(sndpkt) expectedseqnum++ if packet arrives that is not corrupt and has the expected seq number: - extract and deliver data to layer above send ack increase seq number Transport Layer

default udt_send(sndpkt) rdt_rcv(rcvpkt) && notcurrupt(rcvpkt) && hasseqnum(rcvpkt,expectedseqnum) L Wait expectedseqnum=1 sndpkt = make_pkt(0,ACK,chksum) extract(rcvpkt,data) deliver_data(data) sndpkt = make_pkt(expectedseqnum,ACK,chksum) udt_send(sndpkt) expectedseqnum++ if packet arrives that is corrupt or has the wrong seq number: resend ack for last correctly received packet Transport Layer

default udt_send(sndpkt) rdt_rcv(rcvpkt) && notcurrupt(rcvpkt) && hasseqnum(rcvpkt,expectedseqnum) L Wait expectedseqnum=1 sndpkt = make_pkt(0,ACK,chksum) extract(rcvpkt,data) deliver_data(data) sndpkt = make_pkt(expectedseqnum,ACK,chksum) udt_send(sndpkt) expectedseqnum++ ACK-only: always send ACK for correctly-received pkt with highest in-order seq # may generate duplicate ACKs (e.g. if there is a gap) need only remember expectedseqnum out-of-order pkt: discard (don’t buffer): no receiver buffering! re-ACK packet with highest in-order seq # Transport Layer

remark: packet numbers start at 1
GBN in action remark: packet numbers start at 1 (different to applet) sender sender window (N=4) receiver send pkt1 send pkt2 send pkt3 send pkt4 (wait) receive pkt1, send ack1 receive pkt2, send ack2 receive pkt4, discard, (re)send ack2 X loss rcv ack1, send pkt5 rcv ack2, send pkt6 receive pkt5, discard, (re)send ack2 ignore duplicate ACK receive pkt6, discard, (re)send ack2 pkt 3 timeout send pkt3 send pkt4 send pkt5 send pkt6 rcv pkt3, deliver, send ack3 rcv pkt4, deliver, send ack4 rcv pkt5, deliver, send ack5 rcv pkt6, deliver, send ack6 Transport Layer

GBN in action (variant 1)
remark: packet numbers start at 1 (different to applet) GBN in action (variant 1) sender sender window (N=4) receiver send pkt1 send pkt2 send pkt3 send pkt4 (wait) receive pkt1, send ack1 receive pkt2, send ack2 receive pkt4, discard, (re)send ack2 X loss rcv ack1, send pkt5 rcv ack2, send pkt6 receive pkt5, discard, (re)send ack2 ignore duplicate ACK receive pkt6, discard, (re)send ack2 pkt3 timeout send pkt3 send pkt4 send pkt5 send pkt6 rcv pkt3, deliver, send ack3 rcv pkt4, deliver, send ack4 rcv pkt5, deliver, send ack5 rcv pkt6, deliver, send ack6 Transport Layer

GBN in action (variant 2)
remark: packet numbers start at 1 (different to applet) sender sender window (N=4) receiver send pkt1 send pkt2 send pkt3 send pkt4 (wait) receive pkt1, send ack1 receive pkt2, send ack2 receive pkt4, discard, (re)send ack2 X loss increase base and restart timer rcv ack1, send pkt5 rcv ack2, send pkt6 receive pkt5, discard, (re)send ack2 ignore duplicate ACK receive pkt6, discard, (re)send ack2 timeout send pkt3 send pkt4 send pkt5 send pkt6 rcv pkt3, deliver, send ack3 rcv pkt4, deliver, send ack4 rcv pkt5, deliver, send ack5 rcv pkt6, deliver, send ack6 Transport Layer

Selective repeat improves GBN by not retransmitting a large number of packets receiver individually acknowledges all correctly received packets receiver buffers packets, as needed, for eventual in-order delivery to upper layer sender only resends packets for which ACK not received sender has timer for each unACKed packet (since only a single packet will be retransmitted on timeout) sender window N consecutive seq #’s limits seq #s of sent, unACKed packets Transport Layer

Selective repeat: sender, receiver windows
windows are not synchronized!! 4 ACKs are outstanding from viewpoint of sender 2 yellow ACKs are „on the way“ plus one red ACK Transport Layer

Selective repeat receiver sender data from above:
if next available seq # in window, then send packet timeout(n): resend packet n, restart timer of packet n ACK(n) in [sendbase,sendbase+N-1]: mark packet n as received if n smallest unACKed packet, then advance window base to next unACKed seq # packet n in [rcvbase, rcvbase+N-1] send ACK(n) out-of-order: buffer in-order: deliver (also deliver buffered, in-order pkts), advance window to next not-yet-received pkt packet n in [rcvbase-N,rcvbase-1] ACK(n) since previous ack got lost Transport Layer

Selective repeat in action
sender sender window (N=4) receiver send pkt0 send pkt1 send pkt2 send pkt3 (wait) receive pkt0, send ack0 receive pkt1, send ack1 receive pkt3, buffer, send ack3 X loss rcv ack0, send pkt4 rcv ack1, send pkt5 receive pkt4, buffer, send ack4 record ack3 arrived receive pkt5, buffer, send ack5 Q: what happens when ack2 arrives? Receiver will advance window until base = 6 sender will also advance window until base = 6 => show in applet with N=5 pkt 2 timeout resend pkt2 record ack4 arrived rcv pkt2; deliver pkt2, pkt3, pkt4, pkt5; send ack2 record ack5 arrived Q: what happens when ack2 arrives? Transport Layer

Selective repeat: short range of seqnum
sender window (after receipt) receiver window (after receipt) pkt0 pkt1 pkt2 X will accept packet with seq number 0 pkt3 (a) no problem example: seq #’s: 0, 1, 2, 3 window size=3 receiver does not know whether pkt0 is retransmitted or not duplicate data accepted as new in (b)? Q: what relationship between seq # range and window size to avoid problem in (b)? receiver can’t see sender side. receiver behavior identical in both cases! something’s (very) wrong! pkt0 pkt1 pkt2 timeout retransmit pkt0 X will accept packet with seq number 0 (b) oops! Homework: relationship between seq # range and window size Transport Layer

Packet Reordering assume now that packets my be reordered during transmission (since we do not have a single channel but a network of links) receiver may get packets with seqnum n where n is neither in sender’s nor in receiver’s window (e.g. delayed ACK => retransmission => windows move further) duplicate or new data????? sender must be ‘sure’ that when reusing seqnum n, a packet with this number is no longer on the way to receiver Solution: assume that packet cannot ‘live’ in network longer than maximum amount of time (e.g. 3 min for TCP) Transport Layer

Summary of Reliable Data Transfer
Mechanism Use, Comments Checksum Used to detect bit errors in transmitted packet. Timer Used to retransmit if packet or ACK was lost (or delayed). Sequence number Sequential numbering of packet flow. Use gaps to detect lost packets, use duplicate numbers to detect duplicate packets. Acknow-ledgment Receiver confirms correctly received packets; contains sequence number; may be cumulative or individual. (Negative ack-nowledgment) Receiver tells sender that packet has not been received. Window, pipelining Sending of packets only in restricted range => increases utilization of sender compared to stop-and-wait. Transport Layer

TCP: Overview RFCs: 793,1122,1323, 2018, 2581 full duplex data:
bi-directional data flow in same connection MSS: maximum segment size (max amount of data placed in segment); usually about 1460 B since Ethernet frames typically have 1500 Bytes (TCP/IP header: 40 B) connection-oriented: handshaking (exchange of control msgs) inits sender, receiver state before data exchange point-to-point: one sender, one receiver reliable mechanisms for message loss/ corrupted msgs pipelined: TCP congestion and flow control set window size flow controlled: sender will not overwhelm receiver 1452 Byte for most ISPs because DSL uses PPProtocol => 8 Bytes extra Transport Layer

TCP segment structure used for (de-)multiplexing data to/from
32 bits source port # dest port # used for (de-)multiplexing data to/from upper-layer applications sequence number acknowledgement number head len not used U A P R S F receive window checksum Urg data pointer options (variable length) application data (variable length) Transport Layer

TCP segment structure error detection source port # dest port #
32 bits source port # dest port # sequence number acknowledgement number head len not used U A P R S F receive window error detection checksum Urg data pointer options (variable length) application data (variable length) Transport Layer

TCP segment structure source port # dest port # sequence number
32 bits counting by bytes of data (not segments!); used for implementing a reliable data transfer service; more details later source port # dest port # sequence number acknowledgement number head len not used U A P R S F receive window checksum Urg data pointer options (variable length) application data (variable length) Transport Layer

TCP segment structure used for flow control; indicates number of bytes
32 bits source port # dest port # sequence number acknowledgement number head len not used U A P R S F receive window used for flow control; indicates number of bytes that receiver is willing to accept (more details later) checksum Urg data pointer options (variable length) application data (variable length) Transport Layer

TCP segment structure TCP header length (4 bits); because of
32 bits = 4 bytes source port # dest port # sequence number 5 rows acknowledgement number head len not used U A P R S F receive window TCP header length (4 bits); because of options field length can be greater than 20 bytes checksum Urg data pointer options (variable length) application data (variable length) Transport Layer

TCP segment structure flag fields (6 bits): A = ACK field
source port # dest port # flag fields (6 bits): A = ACK field acknowledgement number is valid (this msg is an ACK; used for reliable data transfer!) sequence number acknowledgement number head len not used U A P R S F receive window checksum Urg data pointer options (variable length) application data (variable length) Transport Layer

TCP segment structure flag fields (6 bits): R = RST S = SYN F = FIN
source port # dest port # flag fields (6 bits): R = RST S = SYN F = FIN used for connection setup and teardown (more details later) sequence number acknowledgement number head len not used U A P R S F receive window checksum Urg data pointer options (variable length) application data (variable length) Transport Layer

TCP segment structure flag fields (6 bits): P = PSH field
source port # dest port # sequence number flag fields (6 bits): P = PSH field TCP push function for cases where data needs to be sent immediately (no buffering! sender TCP does not wait for more data) acknowledgement number head len not used U A P R S F receive window checksum Urg data pointer Examples: - Telnet: want each keystroke to be sent immed. HTTP GET request: client has no further data to add and request should be sent to web daemon immediately (do not wait until segment “filled”) packet containing last bytes of requested file (for now sender has no further data to transmit) options (variable length) application data (variable length) Transport Layer

TCP segment structure flag fields (6 bits):
source port # dest port # flag fields (6 bits): U = URG field segment contains data that sending site marked as urgent => urgent data pointer (16 bit) points to last byte of urgent part in segment (isn’t employed much; usually in comb. with PSH) receiver TCP forwards the urgent data to the process with an indication that the data is marked as urgent by the sender sequence number acknowledgement number head len not used U A P R S F receive window checksum Urg data pointer options (variable length) example: you are sending a large file and suddenly realize that it is the wrong file => want abort message to arrive immediately (and not placed in queue behind file data) segment contains 400 bytes of urgent data followed by 200 bytes of regular data => URG bit would be set and the Urgent Pointer field would have a value of 400. application data (variable length) Transport Layer

TCP segment structure source port # dest port # sequence number
32 bits URG: urgent data (generally not used) counting by bytes of data (not segments!) source port # dest port # sequence number ACK: ACK # valid acknowledgement number head len not used PSH: push data now (generally not used) U A P R S F receive window # bytes rcvr willing to accept checksum Urg data pointer RST, SYN, FIN: connection estab (setup, teardown commands) options (variable length) application data (variable length) Internet checksum (as in UDP) Transport Layer

TCP seq. numbers, ACKs sequence numbers:
source port # dest port # sequence number acknowledgement number checksum rwnd urg pointer outgoing segment from sender sequence numbers: byte stream “number” of first byte in segment’s data acknowledgements: seq # of next byte expected from other side cumulative ACK (=> acks bytes up to first missing byte in stream) TCP spec doesn’t say how to handle out-of-order segments, - up to implementor (in practise: receiver buffers and waits for missing bytes to fill gaps) window size N sender sequence number space source port # dest port # sequence number acknowledgement number checksum rwnd urg pointer incoming segment to sender sent ACKed sent, not-yet ACKed (“in-flight”) usable but not yet sent not usable 32- bits number!! A Transport Layer

TCP seq. numbers, ACKs Example 2: Example 1:
Host A has received bytes numbered from Host B Host A waits for puts 536 in ACK number field when it sends next segment to B Example 2: Host A has received bytes numbered from Host B AND bytes (has not yet received 536 – 899) Host A waits for puts 536 in ACK number field when it sends next segment to B Transport Layer

seqnum at B starts at 79 (waiting for byte 42)
Telnet example each character typed by user (at client host A) is sent to sever (host B) and back to be displayed at Telnet’s user screen (echo back) => user sees what has already been processed on remote site seqnum at B starts at 79 (waiting for byte 42) Host A Host B seqnum at A starts at 42 (waiting for byte 79) User types ‘C’ Seq=42, ACK=79, data = ‘C’ host ACKs receipt of ‘C’, echoes back ‘C’ Seq=79, ACK=43, data = ‘C’ host ACKs receipt of echoed ‘C’ Seq=43, ACK=80 Transport Layer

Telnet example each character typed by user (at client host A) is sent to sever (host B) and back to be displayed at Telnet’s user screen (echo back) => user sees what has already been processed on remote site seqnum at B starts at 79 (waiting for byte 42) Host A Host B seqnum at A starts at 42 (waiting for byte 79) User types ‘C’ Seq=42, ACK=79, data = ‘C’ host ACKs receipt of ‘C’, echoes back ‘C’ because data is 1 byte (header is not counted) Seq=79, ACK=43, data = ‘C’ host ACKs receipt of echoed ‘C’ Seq=43, ACK=80 Transport Layer

Telnet example each character typed by user (at client host A) is sent to sever (host B) and back to be displayed at Telnet’s user screen (echo back) => user sees what has already been processed on remote site seqnum at B starts at 79 (waiting for byte 42) Host A Host B seqnum at A starts at 42 (waiting for byte 79) User types ‘C’ Seq=42, ACK=79, data = ‘C’ host ACKs receipt of ‘C’, echoes back ‘C’ Seq=79, ACK=43, data = ‘C’ host ACKs receipt of echoed ‘C’ send data and ack together (ack is piggybacked) Seq=43, ACK=80 Transport Layer

Telnet example each character typed by user (at client host A) is sent to sever (host B) and back to be displayed at Telnet’s user screen (echo back) => user sees what has already been processed on remote site seqnum at B starts at 79 (waiting for byte 42) Host A Host B seqnum at A starts at 42 (waiting for byte 79) User types ‘C’ Seq=42, ACK=79, data = ‘C’ host ACKs receipt of ‘C’, echoes back ‘C’ Seq=79, ACK=43, data = ‘C’ host ACKs receipt of echoed ‘C’ Seq=43, ACK=80 Transport Layer

HTTP example SYN bit = 1 to establish connection (more details later) => increase sequence number by one even though no bytes are sent. Note: we use relative seq num here. When connection is established a random initial number is chosen. Transport Layer

TCP round trip time, timeout
TCP uses a timeout/retransmission mechanism (as rdt protocol considered earlier) Q: how to set TCP timeout value? longer than connection’s round-trip time (RTT) but RTT varies too short: premature timeout, unnecessary retransmissions too long: slow reaction to segment loss Q: how to estimate RTT? SampleRTT: measured time from segment transmission until ACK receipt ignore retransmissions (it is ambiguous whether the reply was for the first instance of the packet or a later instance) SampleRTT will vary => use average average several recent measurements, not just current SampleRTT Transport Layer

EstimatedRTT = (1- )*EstimatedRTT + *SampleRTT exponential weighted moving average influence of past sample decreases exponentially fast (why?) typical value:  = 0.125 RTT: gaia.cs.umass.edu to fantasia.eurecom.fr RTT (milliseconds) sampleRTT EstimatedRTT time (seconds) Transport Layer

EstimatedRTT = (1- )*EstimatedRTT + *SampleRTT exponential weighted moving average influence of past sample decreases exponentially fast (why?) typical value:  = 0.125 RTT: gaia.cs.umass.edu to fantasia.eurecom.fr RTT (milliseconds) α close to 1 => weighted average immune to changes that last a short time (e.g., a single segment that encounters long delay) α close to 0 => weighted average respond to changes in delay very quickly sampleRTT EstimatedRTT time (seconds) Transport Layer

timeout interval: EstimatedRTT plus “safety margin” large variation in EstimatedRTT -> larger safety margin estimate SampleRTT deviation from EstimatedRTT: DevRTT = (1-)*DevRTT + *|SampleRTT-EstimatedRTT| (typically,  = 0.25) TimeoutInterval = EstimatedRTT + 4*DevRTT estimated RTT “safety margin” RFC 2988 Transport Layer

TCP reliable data transfer
TCP creates rdt service on top of IP’s unreliable service pipelined segments cumulative acks single retransmission timer (not one for each segm) retransmissions triggered by: timeout events duplicate acks let’s initially consider simplified TCP sender: data transfer in one direction only ignore duplicate acks ignore flow control, congestion control assume that data from above is less than MSS Transport Layer

TCP sender events: data rcvd from app: create segment with seq #
seq # is byte-stream number of first data byte in segment start timer if not already running think of timer as for oldest unacked segment expiration interval: TimeOutInterval timeout: retransmit segment that caused timeout restart timer ack rcvd: if ack acknowledges previously unacked segments update what is known to be ACKed (slide window to the right) (re)start timer if there are still unacked segments Transport Layer

TCP sender (simplified)
create segment, seq. #: NextSeqNum pass segment to IP (i.e., “send”) NextSeqNum = NextSeqNum + length(data) if (timer currently not running) start timer data received from application above L wait for event NextSeqNum = InitialSeqNum SendBase = InitialSeqNum retransmit not-yet-acked segment with smallest seq. # start timer timeout if (y > SendBase) { SendBase = y /* SendBase–1: last cumulatively ACKed byte */ if (there are currently not-yet-acked segments) start timer } ACK received, with ACK field value y hope that gap is just a single segment RFC 2988, page 3 Transport Layer

create segment, seq. #: NextSeqNum pass segment to IP (i.e., “send”) NextSeqNum = NextSeqNum + length(data) if (timer currently not running) start timer data received from application above L wait for event NextSeqNum = InitialSeqNum SendBase = InitialSeqNum retransmit not-yet-acked segment with smallest seq. # start timer timeout if (y > SendBase) { SendBase = y /* SendBase–1: last cumulatively ACKed byte */ if (there are currently not-yet-acked segments) start timer } ACK received, with ACK field value y RFC 2988, page 4 Transport Layer

TCP: retransmission scenarios
Host A Host B Host A Host B SendBase=92 Seq=92, 8 bytes of data Seq=92, 8 bytes of data Seq=100, 20 bytes of data timeout timeout since nextseqnum at A is 100 ACK=100 X ACK=100 ACK=120 Seq=92, 8 bytes of data Seq=92, 8 bytes of data SendBase=100 SendBase=120 ACK=100 ACK=120 SendBase=120 lost ACK scenario premature timeout Transport Layer

TCP: retransmission scenarios
Host A Host B timeout Seq=92, 8 bytes of data Seq=100, 20 bytes of data ACK=100 X ACK=120 Seq=120, 15 bytes of data cumulative ACK Transport Layer

TCP ACK generation [RFC 1122, RFC 2581]
maybe cumulative ACK possible => increases performance TCP ACK generation [RFC 1122, RFC 2581] event at receiver arrival of in-order segment with expected seq #. All data up to expected seq # already ACKed expected seq #. One other segment has ACK pending arrival of out-of-order segment higher-than-expect seq. # . Gap detected arrival of segment that partially or completely fills gap TCP receiver action receiver is allowed to delay ACK. Wait up to 500ms for next segment. If no next segment, send ACK. immediately send single cumulative ACK, ACKing both in-order segments immediately send duplicate ACK, indicating seq. # of next expected byte immediate send ACK, provided that segment starts at lower end of gap duplicate ACK: an ACK that has been sent before Transport Layer

TCP fast retransmit time-out period often relatively long:
long delay before resending lost packet if segment is lost, there will likely be many duplicate ACKs. TCP fast retransmit if sender receives 3 ACKs for same data (“triple duplicate ACKs”), immediately resend unacked segment with smallest seq # likely that unacked segment lost, so don’t wait for timer to expire Transport Layer

TCP fast retransmit X fast retransmit after sender
Host A Host B timeout Seq=92, 8 bytes of data Seq=100, 20 bytes of data X ACK=100 ACK=100 ACK=100 ACK=100 Seq=100, 20 bytes of data fast retransmit after sender receipt of triple duplicate ACK Transport Layer

create segment, seq. #: NextSeqNum pass segment to IP (i.e., “send”) NextSeqNum = NextSeqNum + length(data) if (timer currently not running) start timer data received from application above L wait for event NextSeqNum = InitialSeqNum SendBase = InitialSeqNum retransmit not-yet-acked segment with smallest seq. # start timer timeout if (y > SendBase) { SendBase = y if (there are currently not-yet-acked segments) start timer } else { /* duplicate ACK received */ increment number of dupl. ACKs received for y if (number of dupl. ACKs received for y==3) resend segment with seq. number y ACK received, with ACK field value y Transport Layer

TCP flow control receiver controls sender such that sender won’t overflow receiver’s buffer by transmitting too much, too fast sender maintains variable called receive window (rwnd) gives sender an idea of how much buffer space is available at receiver since TCP is full-duplex, both sides have receive window variable LastByteRead: number of last byte read by app (on rcv site) LastByteRcvd: number of last byte in received data stream rwnd = RcvBuffer – [LastByteRcvd - LastByteRead] currently being buffered on receiver site Transport Layer

TCP flow control receiver “advertises” free buffer space by including rwnd value in TCP header of receiver-to-sender segments RcvBuffer size set via socket options (typical default is 4096 bytes) sender limits amount of unacked (“in-flight”) data to receiver’s rwnd value sender guarantees that receive buffer will not overflow: to application process buffered data free buffer space RcvBuffer rwnd in the special case that rwnd=0 and the receiver has no data to send, the sender can‘t see any updated value of rwnd => sender sends packtes of 1 byte to receiver to get (with the ACK) the updated rwnd value see applet (in applet: use file size LARGER than buffer => several chunks are sent) TCP segment payloads receiver-side buffering LastByteSent – LastByteAcked ≤ rwnd applet: Transport Layer

Connection Management
before exchanging data, sender/receiver do “three-way handshake”: => agree to establish connection (each knowing the other willing to establish connection) and synchronize seq. numbers Step 1: client send special TCP segment with SYN flag =1 and initial seq. num. (client_isn) in seq. num. field Step 2: SYN segm. arrives at server => sever allocates TCP buffers and variables and sends SYNACK segment with SYN and ACK flag =1 ack number field is equal to client_isn+1 puts own initial seq. num. (server_isn) in seq. num. field Step 3: Client receives SYNACK segm., allocates buffers & vars, and sends ACK with ack num (server_isn+1) (which may already contain app data) why not two-way? Homework Transport Layer

TCP 3-way handshake client state server state LISTEN SYNSENT
SYNbit=1, Seq=x choose init seq num, x send TCP SYN msg SYN RCVD ESTAB SYNbit=1, Seq=y ACKbit=1; ACKnum=x+1 choose init seq num, y send TCP SYNACK msg, acking SYN ACKbit=1, ACKnum=y+1 received SYNACK(x) indicates server is live; send ACK for SYNACK; this segment may contain client-to-server data received ACK(y) indicates client is live ESTAB Transport Layer

TCP: closing a connection
client or server can start closing of connection host A sends TCP segment with FIN bit = 1 host B responds to received FIN with ACK host B also sends FIN (on receiving FIN, ACK can be combined with own FIN) host A receives FIN and sends final ACK host A waits certain time in case that last ACK got lost and FIN is resent (=> A will ack again) Transport Layer

TCP: closing a connection
client state server state ESTAB ESTAB FIN_WAIT_1 FINbit=1, seq=x can no longer send but can receive data clientSocket.close() CLOSE_WAIT FIN_WAIT_2 ACKbit=1; ACKnum=x+1 wait for server close can still send data can no longer send data LAST_ACK TIMED_WAIT FINbit=1, seq=y CLOSED timed wait (usually 30 sec, 1 min or 2 min) CLOSED ACKbit=1; ACKnum=y+1 Transport Layer

Possible TCP states of client
CLOSED wait X seconds send SYN SYN_SENT TIME _WAIT receive SYNACK send ACK receive FIN send ACK ESTAB FIN_WAIT_2 receive ACK (send nothing) send FIN FIN_WAIT_1 Transport Layer

Possible TCP states of server
receive ACK (send nothing) CLOSED create listen socket serverPort = 12000 serverSocket = socket(AF_INET,SOCK_STREAM) serverSocket.bind((‘’,serverPort)) serverSocket.listen(1) LISTEN LAST_ACK receive SYN send SYNACK send FIN SYN_RCVD CLOSE_WAIT receive FIN send ACK receive ACK (send nothing) ESTAB Transport Layer

Principles of congestion control
informally: “too many sources sending too much data too fast for network to handle” different from flow control! problems related to congestion: lost packets (buffer overflow at routers) long delays (queueing in router buffers) important problem in data networks! Transport Layer

Causes/costs of congestion: scenario 1
original data: lin throughput: lout two senders, two receivers one router, infinite buffers output link capacity: R no retransmission Host A unlimited shared output link buffer Host B R/2 packet delay lin R/2 lout lin maximum per-connection throughput: R/2 large delays as arrival rate, lin, approaches capacity Transport Layer

one router, finite buffer sender retransmission of timed-out packet application-layer sends at rate lin transport-layer includes retransmissions : lin’ ≥ lin lin : original data lout l'in: original data, plus retransmitted data Host A finite shared output link buffers Host B Transport Layer

lout lin=l'in idealization: perfect knowledge sender sends only when router buffers available (=> no retransmissions) lin : original data lout copy l'in: original data, plus retransmitted data A free buffer space! finite shared output link buffers Host B Transport Layer

Idealization: known loss packets can be lost, dropped at router due to full buffers sender only resends if packet known to be lost lin : original data lout copy l'in: original data, plus retransmitted data A no buffer space! Host B Transport Layer

Idealization: known loss packets can be lost, dropped at router due to full buffers sender only resends if packet known to be lost R/2 lin lout when sending at R/2, some packets are retransmissions but asymptotic goodput is still R/2 (only original data leaves router) lin : original data lout l'in: original data, plus retransmitted data because the output from the router is always original data (no duplicates!!) A free buffer space! Host B Transport Layer

Realistic: duplicates packets can be lost, dropped at router due to full buffers sender times out prematurely, sending two copies, both of which are delivered R/2 when sending at R/2, some packets are retransmissions including duplicates that are delivered! lout lin R/2 timeout lin lout copy l'in A free buffer space! Host B Transport Layer

Realistic: duplicates packets can be lost, dropped at router due to full buffers sender times out prematurely, sending two copies, both of which are delivered R/2 when sending at R/2, some packets are retransmissions including duplicates that are delivered! lout lin R/2 “costs” of congestion: more work (retrans) for given “goodput” unneeded retransmissions: link carries multiple copies of pkt decreasing goodput Transport Layer

four senders multihop paths timeout/retransmit R: capa. of router links Q: what happens as lin and lin’ increase in A-C connection ? A: as red lin’ increases, all arriving blue pkts at upper queue are dropped, blue throughput g 0 blue in-rate at upper router is at most R<< lin‘; router is ‘flooded’ with red packets => blue packets get lost Host A lout lin : original data Host B l'in: original data, plus retransmitted data finite shared output link buffers Host D Host C Transport Layer

lout lin’ R/2 buffer overflows rare for small lin‘ throughput increases as long as lin‘ is not too large (buffer overflows still rare, more original data arrives) Transport Layer

lout lin’ R/2 lin large for all connections: consider D-B conn. at upper router => most blue packets lost another “cost” of congestion: when packets are dropped, any upstream transmission capacity (left router!) used for that packet was wasted! Transport Layer

Approaches towards congestion control
two broad approaches towards congestion control: end-end congestion control: no explicit feedback from network congestion inferred from end-system observed loss, delay approach taken by TCP network-assisted congestion control: routers provide feedback to end systems single bit indicating congestion explicit rate for sender to send at Transport Layer

Case study: ATM ABR congestion control
Some remarks about ATM networks: Asynchronous Transfer Mode (ATM) (also called cell relay) originally designed to carry both voice and data traffic over WANs only used in some backbone networks of providers (mostly because of its high costs compared to e.g. Ethernet) Price of 100Mbps Enet NIC: < $10 Price of 155Mbps ATM NIC: > $500 fixed-sized packets called cells (Internet has variable sized packets) In order to interconnect with the TCP/IP world, an ATM gateway is used that converts TCP/IP and Ethernet frames into ATM cells and then converts them back once they have reached their destination network. e.g. server cluster needs to be accessed by different departments of a company that are located at different geographical locations Transport Layer

Some remarks about ATM networks: establish virtual-circuit (VC) between two endpoints before data exchange => all cells (of a fixed connection) take same path via certain routers (here called switches) data is delivered in correct order switches can track state of VC => know average transmission rate of sender switches can signal sender to reduce rate in case of congestion => network-assisted congestion control Transport Layer

ABR: available bit rate: transmisson rate is adjusted by sender based on returned RM cells if sender’s path “underloaded”: sender should use available bandwidth if sender’s path congested: sender throttled to minimum guaranteed rate RM (resource management) cells: sent by sender, interspersed with data cells (default: every 32 data cells) travel along the data path to the destination and sent back bits in RM cell set by switches (“network-assisted”) NI bit: no increase in rate (mild congestion) CI bit: congestion indication RM cells returned to sender by receiver (probably modified by receiver and switches) Transport Layer

RM cell data cell two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senders’ send rate thus max supportable rate on path Explicit Forward Congestion Indication (EFCI) bit in data cells: set to 1 in congested switch if data cell preceding RM cell has EFCI set, receiver sets CI bit in returned RM cell Transport Layer

older switch -> does not support RM cells source of illustration: Transport Layer

detailed ABR flow control mechanism is complex and ATM switches allow for many different configurations => ABR is not very prevalent today nice overview article: “ABR Service on ATM Networks: What is ?“ by R. Jain (1995) Transport Layer

TCP congestion control algorithm:
approach: use congestion window variable cwnd choose sending rate accordingly goal: LastByteSent – LastByteAcked ≤ min{cwnd, rwnd} what sender knows receive window (free buffer space at receiver) let’s assume for simplicity that rwnd is large and sender has always data to send => sender sequence number space vorher hatten wir “rwnd” receive window für flow control cwnd last byte ACKed last byte sent sent, not-yet ACKed (“in-flight”) Transport Layer

TCP Congestion Control: details
sender sequence number space TCP sending rate: roughly: send cwnd bytes, wait RTT for ACKs, then send more bytes cwnd last byte ACKed last byte sent sent, not-yet ACKed (“in-flight”) cwnd RTT rate ~ bytes/sec cwnd is dynamic, function of perceived network congestion cwnd indirectly limits sending rate idea: sender tries to find maximal rate at which no losses occur = bandwidth probing Transport Layer

TCP congestion control:
approach: sender increases window size (=> transmission rate) probing for usable bandwidth, until loss occurs additive increase: increase cwnd by 1 MSS for every received ACK until loss detected corresponds to doubling of cwnd every RTT since each segment is of size 1 MSS (or less) and # of received ACKs is old-cwnd-value loss detection: timeout event or three duplicate ACKS Transport Layer

TCP Slow Start (not really slow)
Host A Host B when connection begins, increase rate until first loss event: initially cwnd = 1 MSS double cwnd every RTT done by incrementing cwnd for every ACK received summary: initial rate is slow but ramps up exponentially fast slow start ends when congestion avoidance mode starts (later) of when loss occurs one segment RTT two segments four segments time Transport Layer

TCP: detecting, reacting to loss
loss indicated by timeout: before resetting cwnd store slow start threshold: ssthresh:= cwnd/2 cwnd set to 1 MSS; window then grows exponentially (as in slow start) until threshold ssthres, then enter congestion avoidance mode, where cwnd increased by 1 MSS every RRT (linear growth) loss indicated by 3 duplicate ACKs: dup ACKs indicate network capable of delivering some segments cwnd is cut in half window then grows linearly Earlier TCP versions: always set cwnd to 1 (timeout or 3 duplicate acks) Transport Layer

TCP: switching from slow start to congestion avoidance
Q: when should the exponential increase (slow start) switch to linear? A: when cwnd gets to 1/2 of its value before timeout. Implementation: variable ssthresh on loss event, ssthresh is set to 1/2 of cwnd just before loss event Transport Layer

TCP: additive increase, multiplicative decrease (AIMD) algorithm
(Ignoring slow start and timeouts.) - we have additive increase in congestion avoidance mode: - we have multiplicative decrease in case of 3 duplicate ACKs (after RTT time): after RTT time: cwnd = cwnd + 1 MSS (linear growth) congestion control applet! cwnd = 0.5 * cwnd Transport Layer

Congestion Window: Reno vs Tahoe
Tahoe and Reno curve is identical actually, it should start at 6+3=9! TCP Tahoe: early version of TCP; cut cong. window to 1 MSS in both cases (timeout AND triple dupl. ACK) TCP Reno: newer version; uses ‘fast recovery’ (increase cwnd by 1 MSS for every duplicate ACK) Transport Layer

Summary: TCP Congestion Control
New ACK! congestion avoidance cwnd = cwnd + MSS (MSS/cwnd) dupACKcount = 0 transmit new segment(s), as allowed new ACK . dupACKcount++ duplicate ACK slow start timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment cwnd = cwnd+MSS transmit new segment(s), as allowed new ACK dupACKcount++ duplicate ACK L ssthresh = 64 KiB L cwnd > ssthresh timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment ssthresh= cwnd/2 cwnd = ssthresh + 3 retransmit missing segment dupACKcount == 3 timeout ssthresh = cwnd/2 cwnd = 1 dupACKcount = 0 cwnd = ssthresh dupACKcount = 0 New ACK ssthresh= cwnd/2 cwnd = ssthresh + 3 retransmit missing segment dupACKcount == 3 64*1024/1460 = 44,… fast recovery cwnd = cwnd + MSS transmit new segment(s), as allowed duplicate ACK Transport Layer

slow start timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment cwnd = cwnd+MSS transmit new segment(s), as allowed new ACK dupACKcount++ duplicate ACK L ssthresh = 64 KB increase cwnd by 1 MSS for every ACK if cwnd/MSS segements are sent in time interval [0,RTT] then cwnd is twice as large after receiving cwnd/MSS ACKs Example: 1 MSS = 1460 Bytes cwnd = 14,600 Bytes, then 10 segments are being sent within at RTT Transport Layer

avoidance cwnd = cwnd + MSS (MSS/cwnd) dupACKcount = 0 transmit new segment(s), as allowed new ACK . dupACKcount++ duplicate ACK slow start timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment cwnd = cwnd+MSS transmit new segment(s), as allowed new ACK dupACKcount++ duplicate ACK L ssthresh = 64 KB L cwnd > ssthresh timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment increase cwnd by 1 MSS for every RTT receiving cwnd/MSS ACKs per RTT means we add cwnd/MSS times the factor MSS*(MSS/cwnd) add 1 MSS per RTT example: cwnd/MSS = 10 segments Transport Layer

avoidance cwnd = cwnd + MSS (MSS/cwnd) dupACKcount = 0 transmit new segment(s), as allowed new ACK . dupACKcount++ duplicate ACK slow start timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment cwnd = cwnd+MSS transmit new segment(s), as allowed new ACK dupACKcount++ duplicate ACK L ssthresh = 64 KB L cwnd > ssthresh timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment ssthresh= cwnd/2 cwnd = ssthresh + 3 retransmit missing segment dupACKcount == 3 set cwnd to cwnd/2 since it was too high try to find ‘optimal’ value for cwnd Transport Layer

avoidance cwnd = cwnd + MSS (MSS/cwnd) dupACKcount = 0 transmit new segment(s), as allowed new ACK . dupACKcount++ duplicate ACK slow start timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment cwnd = cwnd+MSS transmit new segment(s), as allowed new ACK dupACKcount++ duplicate ACK L ssthresh = 64 KB L cwnd > ssthresh timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment cwnd = ssthresh dupACKcount = 0 New ACK ssthresh= cwnd/2 cwnd = ssthresh + 3 retransmit missing segment dupACKcount == 3 - increase cwnd on every dupl. ACK (note that dupl. ACK means that certain packets get through) - won’t have many dupl. ACK since we retransmit missing packet - additionally we can transmit further segments (cwnd increases quickly) - see RFC 2581 fast recovery cwnd = cwnd + MSS transmit new segment(s), as allowed duplicate ACK Transport Layer

New ACK! applet congestion avoidance cwnd = cwnd + MSS (MSS/cwnd) dupACKcount = 0 transmit new segment(s), as allowed new ACK . dupACKcount++ duplicate ACK slow start timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment cwnd = cwnd+MSS transmit new segment(s), as allowed new ACK dupACKcount++ duplicate ACK L ssthresh = 64 KB L cwnd > ssthresh timeout ssthresh = cwnd/2 cwnd = 1 MSS dupACKcount = 0 retransmit missing segment ssthresh= cwnd/2 cwnd = ssthresh + 3 retransmit missing segment dupACKcount == 3 timeout ssthresh = cwnd/2 cwnd = 1 dupACKcount = 0 cwnd = ssthresh dupACKcount = 0 New ACK ssthresh= cwnd/2 cwnd = ssthresh + 3 retransmit missing segment dupACKcount == 3 fast recovery cwnd = cwnd + MSS transmit new segment(s), as allowed duplicate ACK Transport Layer

TCP throughput ‘saw tooth’ behavior (if slow start is ignored)
compute avg. TCP throughput as function of window size and RTT using some simplifying assumptions ignore slow start, assume always data to send fixed window size W (measured in bytes) when loss occurs transmission rate ranges from 0.5W/RTT to W/RTT avg. window size (# in-flight bytes) is ¾ W avg TCP throughput = 3 4 W RTT Bytes/sec W W/2 Transport Layer

High-Speed TCP Connections
avg TCP throughput = 3 4 W RTT Bytes/sec example: 1500 Byte segments, 100ms RTT, want 10 Gbps throughput (≈ 1010 / (1500*8) = 1/12 * 107 segments/sec) => W = thr * RTT * 4/3 = 1/12 * 107 * 1/10 * 4/3 = 1/9 * 106 ≈ 105 in-flight segments (!!!!) This is a lot!!! Transport Layer

High-Speed TCP Connections
What fraction of packets could be lost so that we still have a throughput of 10Gbps? throughput in terms of segment loss probability, L: ➜ to achieve 10 Gbps throughput, need a loss rate of L ≈ 2· – very small loss rate (one loss every 5 billion segm) new versions of TCP for high-speed (RFC 3649) TCP throughput = 1.22 . MSS RTT L homework 1 lost packet in every 1000 packets is more realistic arguments against large frames: - Large packets can occupy a slow link for some time, causing greater delays to following packets and increasing lag and minimum latency. - frame size must be compatible with all link types (e.g. dial up connection!) otherwise: fragmentation Remark: one cannot simply increase the MSS since this will lead to a lot of IP fragmentation (=> less efficient and in case of fragment loss whole segment is lost) default is 536 B => plus 40 B header gives MTU for IP networks local networks use larger MSS (e.g Bytes for Ethernet) modern LANs can use Jumbo frames (MTU up to 9000 B) Transport Layer

TCP Fairness fairness goal: if K TCP sessions share same bottleneck link of bandwidth R, each should have average rate of R/K TCP connection 1 bottleneck router capacity R TCP connection 2 Transport Layer

Why TCP is fair (Intuition)
two competing sessions (same MSS, RTT, no other traffic): below full utilization, increase cwnd by 1 MSS per RTT in congestion avoidance mode (ignore slow start phase) both detect losses during increase => halve their windows equal bandwidth share full bandwidth utilization R loss: decrease window by factor of 2 congestion avoidance: additive increase Connection 2 throughput can be generalized to more dimensions in reality: connections with small RTT are able to grab available bandwidth more quickly (faster increase of window) => higher throughput loss: decrease window by factor of 2 congestion avoidance: additive increase Connection 1 throughput R => realized bandwidth converges to equal bandwidth share! Transport Layer

Fairness (more) Fairness and UDP Fairness, parallel TCP connections
multimedia apps often do not use TCP do not want rate throttled by congestion control instead use UDP: send audio/video at constant rate, tolerate packet loss => crowd out TCP traffic need for congestion control for UDP Fairness, parallel TCP connections application can open multiple parallel connections between two hosts => larger fraction of bandwidth at congested link (e.g. multiple tabs in browser) unfair! e.g., link of rate R with 9 existing connections: new app asks for 1 TCP, gets rate R/10 new app asks for 11 TCPs, gets 11* R/20 Transport Layer

Chapter 3: summary principles behind transport layer services:
multiplexing, demultiplexing reliable data transfer flow control congestion control their instantiation, implementation in the Internet UDP TCP Remark: TCP is actually more complex than what we have described (variety of patches, fixes improvements) next: leaving the network “edge” (application, transport layers) into the network “core” Transport Layer

Chapter 3: Transport Layer

Similar presentations

Presentation on theme: "Chapter 3: Transport Layer"— Presentation transcript:

Similar presentations

About project

Feedback

Log in

Auth with social network:

Chapter 3: Transport Layer

Similar presentations

Presentation on theme: "Chapter 3: Transport Layer"— Presentation transcript:

Similar presentations

About project

Feedback