Introduction to TCP A first look at the sockets API for ‘connection-oriented’ client/server application programs

Benefits of TCP Communication is ‘transparently reliable’ Data is delivered in the proper sequence An application programmer does not need to worry about issues such as: –Lost or delayed packets –Timeouts and retransmissions –Duplicated packets –Packets arriving out-of-sequence –Flow and congestion control

network layer transport layer application layer Overview T C P port P port Q ………… I P unreliable IP datagrams process A process B reliable TCP byte-stream connection

Transport-Layer’s duties Create the illusion of a reliable two-way point-to-point connection linking a client application with a server application –Manage the ‘error-control’ mechanism –Manage the ‘flow-control’ mechanism –Manage the connection’s ‘persistence’ –Manage the connection’s ‘shutdown’

Interaction overview socket() bind() listen() accept() read() write() close() socket() bind() connect() write() read() close() The ‘server’ application The ‘client’ application 3-way handshake data flow to server data flow to client 4-way handshake

What is a ‘connection’? An application’s socket is ‘connected’ if it has a defined pair of socket-addresses: –An IP-address and port-number for the ‘host’ –An IP-address and port-numbet for the ‘peer’ ‘hopper.usfca.edu’ port port USF’s web-server ‘hrn23501.usfca.edu’ port port 80 classroom workstation 2-way data stream

Layout of TCP header destination port address sequence number acknowledgment number Header Length options and padding source port address 32-bits window size urgent pointer FINFIN SYNSYN RSTRST PSHPSH ACKACK URGURG reserved checksum

Sequence number data (256 bytes) The sequence number field defines the number being assigned to the first byte of data contained in this segment. During connection setup, each party to the connection uses a random number generator to get the value it will assign to the first byte of data it will transmit, called its initial sequence number (ISN). Thereafter, the sequence number in each succeeding segment will equal the sequence number used in the prior segment plus the number of data bytes in that prior segment. By this scheme the receiver can arrange all the incoming data bytes in the proper order, even if some segments happen to arrive out-of-order. data (256 bytes) data (256 bytes) data (232 bytes) ISNISN + 256ISN + 512ISN segmented stream of 1000 bytes

Acknowledgment number This field holds the number of the byte that the source of this segment is expecting to receive next from its connection partner This field’s value is meaningful only when this segment’s ACK control flag bit is set sender receiver 9, 8, 7 6, 5  4, 3, 2, 1 ACK 5

TCP Header Length Like IP headers, the TCP Header’s length is expressed in multiples of 32-bits: it’s at least 5 (i.e., 20-bytes) if there aren’t any ‘TCP Options’ included in the TCP header The amount of DATA in a TCP packet can be calculated from the IP Header’s ‘Total Length’ field, minus the number of bytes that comprise these two headers (IP header + TCP header) IP headerTCP headerDATA Total Length (in bytes)

Control flags URGURG ACKACK PSHPSH RSTRST SYNSYN FINFIN Legend: FIN = Terminate the connection SYN = Synchronize sequence numbers RST = Reset the connection PSH = Push the data ACK = The value in the acknowledgement field is valid URG = The value in the urgent pointer field is valid

Establishing the connection server application (passive) client application (active) timeline SYN J SYN K, ACK J+1 ACK K+1 The 3-way Handshake

Exchanging data server application (passive) client application (active) timeline PSH+ACK ACK PSH+ACK A typical Request and Reply transaction

Connection shutdown server application (passive) client application (active) timeline ACK+FIN ACK The 4-way Handshake ACK+FIN

3-way handshake SYN ACK + SYN ACK

‘request-and-reply’ ACK+PSH ACK ACK+PSH ACK+FIN

4-way handshake ACK+FIN ACK ACK+FIN ACK

TCP Timers To achieve transparent reliability, the TCP subsystem maintains some internal timers One of these is the ‘Retransmission Timer’ If a packet is sent, but its ACK does not arrive before this timer expires, then the packet will be ‘retransmitted’ Of course, this could result in the receiver getting duplicate packets (if it’s a bit slow)

‘lost’ versus ‘late’ client application server application timeline ACK PSH+ACK Busy server might be ‘slow’ to acknowledge PSH+ACK retransmit timeout ACK arrives late same PSH arrives twice

‘piggyback’ To reduce traffic-flow when possible, TCP delays sending an immediate ACK for an arriving data-packet, in case the receiver might soon have some data of its own to send back – in which case the ACK can ‘piggyback’ on the outgoing data PSH This mechanism, of course, requires TCP to maintain a ‘Delayed ACK’ timer

Delayed ACK senario transport layer application layer T C P port P …… process A buffer for outgoing data buffer for incoming data writeread Retransmit timer Delayed ACK timer Keep Alive timer Window Probe timer to/from the IP layer

Window Size During the ‘connection setup’ handshake, each host communicates to its partner a ‘window size’ parameter, to let be known some information about its capacity for buffering packets It also conveys its ‘MSS’ parameter (as a ‘TCP header option’) to inform its partner of its buffers’ Maximum Segment Size

MSS versus MTU A diagram shows the distinction between the protocol’s MSS (Maximum Segment Size) and the interface’s MTU (Maximum Transmission Unit); no TCP packets will be sent with a segment-size that’s larger datalink header IP header TCP header packet DATA FCS MSS MTU

Main TCP option types The TCP Header contains an options list, occupying from 0 to 11 longword values Each option is identified by an 8-bit ‘type’: Type 0: End of the options list: data follows this Type 1: No option: used for alignment padding Type 2: Maximum Segment Size (MSS) Type 3: Window scaling option (WSOPT) Type 4: Selective Acknowledgments supported Type 5: Selective Acknowledgment (SACK) Type 8: Timestamp value and echo reply (TSOPT)

Option formats Option types 0 and 1 are single-bytes All other option types are at least 2-bytes, with the second byte containing the length type 2 length 4 MSS value type 3 length 3 WSOPT value 4 bytes2 bytes type 4 length 2 3 bytes type 8 length bytes timestamp value timestamp echo reply value

Looking at TCP options HLEN

‘Wrapped’ sequences The TCP header’s ‘Sequence Number’ is a 32-bit value, initially chosen at random It could happen that a large number gets selected as an Initial Sequence Number and that a large amount of data gets sent, thus causing the 32-bit field to ‘overflow’ So how does the receiver tell a ‘wrapped sequence’ from a ‘late-arriving’ segment?

Type 8: TSOPT The TCP timestamps have two purposes: –RTTM: Round-Trip Time Measurement –PAWS: Protect Against Wrapped Sequences TYPE (=8) Timestamp Value Timestamp Echo Reply LENGTH (=10) 16 bits

The SACK option It conveys extended ‘acknowledgment’ information from a receiver to a sender about ‘gaps’ in the received data-stream Type (=5)Length Left Edge of first Block Right Edge of first Block... Left Edge of n-th Block Right Edge of n-th Block 32-bits 2+n*8 bytes no gaps here

Demo programs We put ‘tcpserver.cpp’ and ‘tcpclient.cpp’ on our class website, so you can watch actual TCP packets being exchanged by using our ‘nicwatch’ application (or some other packet-sniffer, e.g., ‘wireshark’) We deliberately used loops which write to, or read from, sockets one-byte-at-a-time so that you can observe ‘TCP buffering’!