TCP: Transmission Control Protocol UDP: User Datagram Protocol TCP - 1

Transcription

1 TCP/IP Family of Protocols (cont.) TCP: Transmission Control Protocol UDP: User Datagram Protocol TCP - 1

2 What is TCP? TCP is the reliable transport protocol of TCP/IP TCP is an Automatic Repeat request (ARQ), Continuous RQ (packets are continually sent), protocol TCP utilizes acknowledgements (ACKs) TCP maintains a sliding window TCP maintains a retransmission timeout Douglas E. Comer: Computer Networks TCP - 2

3 Layer 4 Addressing: Port Numbers To talk to another port, a sender needs to know both the IP address and the port number at the destination Each message must carry destination IP and port number, as well as the source IP and port number TCP - 3

4 Structure of a TCP-segment Bit Octet 4 8 Source Port Sequence Number Acknowledgement Number Destination Port buffer size of sending node used as an indication of what the node is willing to accept (advertised window size, max. window size) 12 Data Offset Reserved U R G A C K P S H R S T S Y N F I N Window 16 Checksum Urgent Pointer 20 Options Padding Data TCP - 4

5 Code Bit (Flag) Settings URG: The data contained within this packet is urgent (i.e. keystroke data). The urgent pointer field is active and valid. ACK: The acknowledgment field is valid. PSH: Forces an immediate pushing of all data through lower layers to transmit or the pushing of all data from the transport layers to the upper layer protocols. This feature circumvents TCP s typical collection routines that may hold data before releasing to higher or lower layers if other processes are underway. RST: Reset the connection. This occurs when some unexpected event interferes with communication. SYN: Synchronize the sequence numbers. This bit initiates a communication relationship. The sender will send a packet with this code active. The receiver will acknowledge with a properly sequenced acknowledgment and its own sequence number. The sender will then respond with its own properly sequenced acknowledgment to begin the communication. FIN: Data transmission is finished. This connection is a candidate for termination. TCP - 5

6 Sliding Window The window size represents the maximum amount of unacknowledged packets on flight at any given time The Sliding Window Algorithm is: Transmit all new segments according to window Wait for acknowledgement Slide the window TCP - 6

7 Sliding Window (cont.) [ ] > ACK0... (slide window) < [ ] >... ACK4... (slide window) < [ ] > TCP - 7

8 Reaching Equilibrium P r P b Bandw width Sender P b Receiver A s A r Time TCP - 8

9 Reaching Equilibrium (cont.) A connection is said to have reached equilibrium when: Transmitted data acquires an inter-packet spacing equal to the transmission time of the packets on the slowest link in the path TCP - 9

10 TCP Round Trip Time Control is based on the Round Trip Time RTT (Average) Round-Trip-Time (RTT) estimations New_Round_Trip_Sample: measured as the time between sending a segment and receiving the acknowledgement RTT:=α*old_RTT + (1 - α)*new_round_trip_sample 0 α < 1, e.g. α:=0.9 α close to 1: reacts slowly to changes α close to 0: reacts quickly to changes e.g. see ping results TCP - 10

11 TCP Timeout Value & Karn s Algorithm TCP timeout value timeout:= β * RTT is based on Round-Trip-Time (RTT) estimations In the event of retransmission, TCP is unable to distinguish between ACKs for the same sequence number. Karn suggested to cease RTT estimations during TCP timeouts. TCP - 11

12 Karn s Algorithm (cont.) Ignore RTTs from retransmitted segments TCP - 12

13 Slow-Start Aims to identify link capacity Algorithm Description: 1. Start sending one packet 2. For each ACK received, send two packets Slow-Start forces an exponential increase of cwnd (congestion window size) cwnd doubles after approx. RTT cwnd reaches maximum size (MaxWin) in RTT*log 2 (MaxWin) TCP - 13

14 Congestion Avoidance (state) Aims to preserve equilibrium: if cwnd <= ssthresh slow start else congestion avoidance endif Congestion Avoidance is always implemented with Slow-Start: When an ACK is received: if (cwnd <= ssthresh) cwnd += 1; else cwnd += 1/cwnd; If a timeout occurs: ssthresh=cwnd/2; cwnd=1 TCP - 14

15 Fast Retransmit Aims to avoid long timeout intervals TCP may only acknowledge in order received packets TCP does not utilize NACKs In the event of single packet loss all subsequent packets are received out-of-order TCP responds with duplicate ACKs (dupacks) If the dupacks exceed a threshold, TCP responds with retransmission without prior timeout Fast Retransmit is succeeded by Slow-Start TCP - 15

16 Fast Recovery Aims at a faster window recovery Fast Recovery dictates that Fast Retransmit should be followed by Congestion Avoidance: When a dupack is received: if (dupack = 3) { retransmit(last); ssthresh = min(cwnd,advwnd)/2; cwnd = ssthresh + 3; else cwnd ++; When an ACK is received: cwnd = ssthresh; TCP - 16

17 TCP: finite state machine/ state transition diagram begin passive open syn/syn + ack anything/reset CLOSED close active open/syn (1) LISTEN SYN RECVD reset ack syn/syn + ack send/syn syn + ack/ack (2) SYN SENT close/ time out/ reset A connection starts with the state CLOSED. Input/Output denotes the input causing a state transition and the output generated. FIN WAIT-1 ack/ FIN WAIT-2 close/fin close/fin fin/ack fin/ack ESTAB- LISHED CLOSING ack/ TIME WAIT fin/ack (3) close/fin CLOSE WAIT LAST ACK timeout after 2 segment lifetimes ack/ TCP - 17

18 Opening atcp-connection TCP - 18

19 TCP A Closing a TCP-connection TCP B Sends segment with activated FIN Flag Connection from A to B closed Receives FIN segment and informs application layer Sends confirmation Sends segment with activated FIN Flag Receives FIN segment and confirms receipt Connection from B to A closed TCP - 19

20 TCP Animation Slow-start = For every Ack two data packets are transmitted Slower links require more transmission time (bottleneck) Eventual cause of packet loss (Dropped) 256Kb, 2ms 256Kb, 2ms Sender 1 Data, 1500bytes 2 Dropped Ack, 40bytes Receiver 6 10Mb, 2ms 1Mb, 5ms 10Mb, 2ms TCP - 20

21 4.3BSD Tahoe Slow-start with Congestion Avoidance Congestion Window Slow-start Threshold Packets Time (seconds) TCP - 21

22 4.3BSD Tahoe Packet Trace 2500 TCP Packet Dropped Packets 2000 TCP Packets (Sequenc ce Number) Time (seconds) TCP - 22

23 4.3BSD Reno 4.3BSD Tahoe with Fast Retransmit & Recovery Congestion Window Slow-start Threshold Packets Time (seconds) TCP - 23

24 4.3BSD Reno Packet Trace 2500 TCP Packet Dropped Packets 2000 TCP Packets (Sequenc ce Number) Time (seconds) TCP - 24

25 TCP - 25

26 Problems with TCP TCP provokes packet losses to get an idea about the state of the network. The protocol has no explicit feedback only implicit feedback when something goes wrong. The use of the available bandwidth oscillates. This means that the load of the network varies; small load: space left, unused available capacity overload: network overflow, packet losses, delays, retransmissions, i.e. slower than it has to be. There has been a shift in the Internet traffic. Due to short transmissions, TCP is almost always in slow start, measurements show 85% in slow start. There is no difference between different trace classes, such as ftp files, video streams or . Originally less than 1% packet losses, now 5-7%, which means many retransmissions (13%), most of which are due to timeouts. TCP always assumes that packet losses are due to congestion. Problem in Mobile Networks!!! TCP - 26

27 Animation Ns2-network simulator Look for animation on the net: e.g.: Linux Demonstrations: route ping etc. Exercise: When is ssthresh increased? TCP - 27

28 UDP (User Datagram Protocol) Appeared first in IETF RFC 768 a procedure for application programs to send messages to other programs with a minimum of protocol mechanism TCP - 28

29 UDP (cont.) Resides at the same level as TCP Connectionless service Uses underlying IP to carry messages A thin protocol limited functionalities NO ACK, NO segmentation, NO sequencing Applications: e.g., NFS, SNMP, DNS TCP - 29

30 Consists of: UDP Message Format 1) UDP header 2) UDP data area TCP - 30

31 Data Integrity UDP IP does not compute a checksum on the data portion of an IP datagram UDP has to make sure the data portion (IP address, actual data.) is correct UDP requests IP information from lower level TCP - 31

32 Data Integrity UDP (cont.) Pseudo-header is prepended to a UDP datagram TCP - 32

33 Data Integrity (cont.) 16 bit 1 s complement on the entire object (pseudo-header, header, data) To ensure data integrity: Sender-side: Create pseudo-header, then checksum is performed Checksum is saved at the checksum field Receiver-side: Re-create the pseudo-header and recompute the checksum TCP - 33

34 UDP Datagram Transmission Multiplexes data from different applications/processes UDP datagram is encapsulated with different headers before actual 0 s and 1 s physical transmission Application Transport UDP header User Data User Data Internet IP header UDP Datagram Network Frame header IP Datagram TCP - 34

35 UDP Datagram Transmission (cont.) When a packet arrives, headers are removed layer by layer UDP is responsible for datagram demultiplexing based on the destination port number Application #1 Application #2 Port #1 Port #2 UDP IP Layer.Previous Layers TCP - 35

36 Port Assignment How do we assign port numbers to different applications? Both sender and receiver must agree on port numbers before they can interoperate Two approaches: 1) Universal Assignment Everyone agrees on some port assignments 2) Dynamic Binding Port numbers are assigned locally, sender must ask the receiver for the current port assignments TCP - 36

37 Port Assignment (cont.) A hybrid approach has been chosen Ports are well-known ports Rest are dynamic ports Port number Services 21 FTP 22 SSH 25 SMTP 53 DNS 69 TFTP 80 WWW 110 POP3 161 SNMP see: TCP - 37