Computer Networking Concept

Transcription

1 Copyright, 2011 MMLab, Dept. of ECE, UOS Computer Networking Concept Seong Jong Choi, PhD. Professor University of Seoul Dept. of Electrical and Computer Eng. Web:

2 주요개념 프로토콜 프로토콜계층 인터페이스 서비스 FTP Telnet Web TCP UDP IP Ethernet Twisted Pair IPX Token Ring Fiber Same Interfaces July, 2011 Computer Network Concepts 2

3 전송매체 매체사용 : shared, switched, collision, carrier sense, collision detect, multiple access, slots, tokens 장비 repeaters, hubs, bridges, switches, routers, application gateways July, 2011 Computer Network Concepts 3

4 연결방법 Circuit-Switching 자원확보 최대지연시간보장 사용하지않으면 : Expensive 멀티미디어전송 Packet-Switching 헤더 저장후전송 (Store-and-Forward) 효율적 : 저렴함 Connectionless vs Connection-oriented July, 2011 Computer Network Concepts 4

5 데이터통신 흐름제어 (Flow Control) Sliding window 오류탐지 / 오류정정 crc Congestion Control July, 2011 Computer Network Concepts 5

6 네트워크계층 도착지로메시지전송하는방법 네트워크주소 라우팅 (Routing) July, 2011 Computer Network Concepts 6

7 TCP/IP Reference Model TCP = Transmission Control Protocol IP = Internet Protocol (Routing) TCP/IP Ref Model Application Transport Internetwork Host to Network FTP TCP IP Ether net TCP/IP Protocols Telnet Packet Radio UDP HTTP Point-to- Point OSI Ref Model Application Presentation Session Transport Network Datalink Physical July, 2011 Computer Network Concepts 7

8 Encapsulation TCP Header FTP Header TCP Data FTP Data IP Header IP Data Ethernet Header Ethernet Data Ethernet Trailer July, 2011 Computer Network Concepts 8

9 The Internet Protocol Tree SMTP RPC rlogin & rsh Telnet DNS SNMP TFTP RPC XDR BOOTP & DHCP FTP NFS TCP UDP IP + ICMP + IGMP ARP RARP H/W Device Drivers and Media Access Protocols July, 2011 Computer Network Concepts 9

10 연결장비 Repeater: PHY device that restores data and collision signals: a digital amplifier Hub: Multi-port repeater + fault detection Bridge: Data link layer device connecting two or more collision domains. MAC multicasts are propagated throughout extended LAN. Router: Network layer device. IP, IPX, AppleTalk. Does not propagate MAC multicasts. Switch: Multi-port bridge with parallel paths July, 2011 Computer Network Concepts 10

11 연결장비 Application Transport Network Datalink Physical Gateway Router Bridge/Switch Repeater/Hub Application Transport Network Datalink Physical July, 2011 Computer Network Concepts 11

12 TCP: Key features Connection-oriented Point-to-point: 2 end-points (no broadcast or multicast) Reliable transfer: Data is delivered in-order Full duplex communication Byte-stream I/f: sequence of octets Reliable startup: Data on old connection does not confuse new connections Graceful shutdown: Data sent before closing a connection is not lost. Reset or immediate shutdown also possible.

13 Reliability Reliability provided by: Reliable connection startup: Data on old connection does not confuse new connections Graceful connection shutdown: Data sent before closing a connection is not lost. Data segmented for transmission and acknowledged by destination. Timeout + Retransmission provided if data unacknowledged Checksum provided to catch errors. Resequencing of out-of-order data; discarding of duplicate data. Window flow control => sender cannot overrun receiver buffers

14 End-to-end Service

15 TCP Header Format Source Port Dest Port Seq No Ack No Header length Resvd Control Window Checksum Urgent Options Pad Data x y Size in bits

16 IP Features Connectionless service Addressing Data forwarding Fragmentation and reassembly Supports variable size datagrams Best-effort delivery: Delay, out-of-order, corruption, and loss possible. Higher layers should handle these. Provides only Send and Delivery services Error and control messages generated by Internet Control Message Protocol (ICMP)

17 What IP does NOT provide End-to-end data reliability & flow control (done by TCP or application layer protocols) Sequencing of packets (like TCP) Error detection in payload (TCP, UDP or other transport layers) Error reporting (ICMP) Setting up route tables (Routing protocols:rip, OSPF) Connection setup (it is connectionless) Address/Name resolution (ARP, RARP, DNS) Configuration (BOOTP, DHCP) Multicast (IGMP, MBONE)

18 IP Datagram Format Vers H. Len TOS Total Length Identification Flags Fragment Offset Time to live Protocol Header Checksum Source IP Address Destination IP Address IP Options (if any) Padding Data

19 IP Datagram Format 1st Word (purpose: info, variable size header, packet) Version (4 bits) Internet header length (4 bits): units of 32-bit words. Min header is 5 words or 20 bytes. Type of service (TOS: 8 bits): Reliability, precedence, delay, and throughput. Not widely supported Total length (16 bits): header + data in bytes Total must be less than 64 kb. 2nd Word (purpose: fragmentation): Identifier (16 bits): Helps uniquely identify the datagram between any source, destination address

20 IP Header (Cont) Flags (3 bits):more Flag (MF) - for fragmentation Don t Fragment (DF) Reserved Fragment offset (13 bits): In units of 8 bytes 3rd word: (purpose: demuxing, error, timeout) Time to live (8 bits): Specified in router hops Protocol (8 bits): Next level protocol to receive the data Header checksum (16 bits): 1 s complement sum of all 16- bit words in the header. Change header => modify checksum using 1 s complement arithmetic.

21 IP Header (Cont) Source Address (32 bits): Original source. Does not change along the path. Destination Address (32 bits): Final destination. Does not change along the path. Options (variable): Security, source route, record route, stream id (used for voice) for reserved resources, timestamp recording Padding (variable): Makes header length a multiple of 4 Data (variable): Data + header < 65,535 bytes

22 Fragmentation Datagrams larger than MTU are fragmented Original header is copied to each fragment and then modified (fragment flag, fragment offset, length,...) Some option fields are copied (see RFC 791) IP Header Original Datagram IP Hdr 1 Data 1 IP Hdr 2 Data 2 IP Hdr 3 Data 3 Fig 17.4

23 Maximum Transmission Unit Each subnet has a maximum frame size Ethernet: 1518 bytes FDDI: 4500 bytes Token Ring: 2 to 4 kb Transmission Unit = IP datagram (data + header) Each subnet has a maximum IP datagram length: MTU S Net 1 MTU=1500 R Net 2 MTU=1000 R

24 Fragmentation Example MTU = 1500B MTU = 280B IHL = 5, ID = 111, More = 0 Offset = 0W, Len = 472B IHL=5, ID = 111, More = 1 Offset = 0W, Len = 276B IHL=5, ID = 111, More = 0 Offset = 32W, Len = 216B Payload size 452 bytes needs to be transmitted across a Ethernet (MTU=1500B) and a SLIP line (MTU=280B) Length = 472B, Header = 20B => Payload = 452B Fragments need to be multiple of 8-bytes. Nearest multiple to 260 (280-20B) is 256B First fragment length = 256B + 20B = 276B. Second fragment length = (452B- 256B) + 20B = 216B

25 Reassembly Reassambly only at the final destination Partial datagrams are discarded after a timeout Fragments can be further fragmented along the path. Subfragments have a format similar to fragments. It is not possible to tell how many times fragmented. Minimum MTU along a path Path MTU S D Net 1 MTU=1500 R1 Net 2 MTU=1000 R2 Net 3 MTU=1500

26 Further notes on Fragmentation Performance: single fragment lost => entire packet useless. Waste of resources all along the way. Ref: Kent & Mogul, 1987 Don t Fragment (DF) bit set => datagram discarded if need to fragment. ICMP message generated: may specify MTU (default = 0) Used to determine Path MTU (both in TCP, UDP) The transport and application layer headers do not appear in all fragments. Problem if you peep into those headers.

27 IP Protocol Numbers Decimal Key word Protocol 0 Reserved 1 ICMP Internet Control Message Protocol 2 IGMP Internet Group Management Protocol 4 ST Stream Protocol 5 TCP Transmission Control Protocol 8 EGP Exterior Gateway Protocol 9 IGP Interior Gateway Protocol 17 UDP User Datagram Protocol

28 IP Options Coding Type Length Value 1B 1B nb Flag Copy Class Number 1b 2b 5b Flag Copy: 0 = Copy the option only into the first fragment of a fragmented datagram 1 = Copy into all fragments Class: 0 =User or control, 1=Reserved, 2=Diagnostics, 3=reserved

29 IP Options Class Number Length Description End of Options No Op Security 0 3 Var Loose Source Routing 0 7 Var Record Route Stream ID (obsolete) 0 9 Var Strict Source Routing 2 4 Var Internet Time-Stamp

30 IP Source Routing Code Length Pointer Router Data P P Loose Source Routing (LSSR): Specify partial route list Strict Source Routing: Specify full route.

31 Route Recording Code Length Pointer Route Data P Empty Empty P Empty Need to allow enough space to record IP addresses on route. Datagram size does not change as it goes through internet.

32 Timestamp Option Code Length Pointer Data Oflw Flags IP Address 1 Timestamp 1 IP Address n Timestamp n Record timestamps along route Overflow (Oflw) counter incremented if out of space Flags: allows some further options for flexibility

33 Discussion on IP Header Design If fragmentation is going to be avoided all the time, why not have the 4-bytes of fragmentation info as an IP option? Is 32-bit addresses going to be enough? Why mess with variable length headers? Can the variability in header length be controlled to allow better encoding? Are the IP options really that useful? Why variable length option headers? Many of these issues addressed in IPv6.

34 TCP Header Source Port (16 bits): Identifies source user process 20 = FTP, 23 = Telnet, 53 = DNS, 80 = HTTP,... Destination Port (16 bits) Sequence Number (32 bits): Sequence number of the first byte in the segment. If SYN is present, this is the initial sequence number (ISN) and the first data byte is ISN+1. Ack number (32 bits): Next byte expected

35 TCP Header (Cont) Checksum (16 bits): covers the segment + pseudo header. Protection from mis-delivery. Urgent pointer (16 bits): Points to the byte following urgent data. Lets receiver know how much data it should deliver right away. Options (variable): Max segment size (does not include TCP header, default 536 bytes), Window scale factor, Selective Ack permitted, Timestamp, No-Op, End-of-options

36 TCP Checksum Checksum is the 16-bit one's complement of the one's complement sum of a pseudo header, the TCP header, and the data, (padded with zero octets at the end if necessary to make a multiple of two octets.) Checksum field filled with zeros initially Pseudo header (similar to UDP) used in calculations, but not transmitted. RFC Source Adr Dest. Adr Zeros Protocol TCP Length TCP Header TCP data

37 Connection Establishment Fig 18.3 Client sends SYN, with an initial sequence number (ISN) and a Max Segment Size (MSS). Called active open. Server acks the SYN (for the forward connection), and also sets the SYN bit, with its own ISN (for the reverse connection). Called passive open. Client acks the reverse direction SYN. 3 segments transmitted.

38 Connection Termination Fig 18.3 again, also fig 18.5 Client sends FIN. Server acks this and notifies its application. However it can keep its halfconnection open. Each connection closed separately. Server app issues a close and server sends FIN to client. Client acks this. 4 segments transmitted.

39 Three-Way Handshake 3-way handshake: necessary and sufficient for unambiguous setup/teardown even under conditions of loss, duplication, and delay

40 More Connection Establishment Socket: BSD term to denote an IP address + a port number. A connection is fully specified by a socket pair i.e. the source IP address, source port, destination IP address, destination port. Initial Sequence Number (ISN): counter maintained in OS. BSD increments it by every 500ms or new connection setup => time to wrap around < 9.5 hours.

41 SYN pkt lost => retransmitted. Exponential timeout backoff (6, 12, 24 s etc) Connection timeout is 75 s. Timer granularity is 500 ms => first timeout between 5.5 and 6s. See Fig. 18.7

42 MSS Largest chunk sent between TCPs. Default = 536 bytes. Announced in connection establishment. Not negotiated. Different MSS possible for forward/reverse paths. Does not include TCP header Many BSD systems restrict MSS to be multiples of 512 bytes: inefficient. Path MTU restricts size of MSS further.

43 Half close, Half open, Reset Possible for one end to close while the other end sends data. Used in rsh command. Fig 18.10, Half-open: one side crashed and lost memory of connection while other side thinks connection is open. Usually connection is reset upon communication. Reset => used to abort connection. Queued data (if any) is dumped. Orderly release => FIN sent after queued data transmitted.

44 TCP state transition diagram Figure 18.12: client (dark line), server (dashed line) transitions. 2MSL wait: wait for final segment to be transmitted before releasing connection (typically 2 min) Socket pair cannot be reused during 2MSL Delayed segments dropped Conn Establishment: SYN_SENT, SYN_RCVD, ESTABLISHED, LISTEN Close: FIN_WAIT_1, FIN_WAIT_2, CLOSING, TIME_WAIT, CLOSE_WAIT, LAST_ACK

45 Effect of 2MSL wait Can t kill server & restart immediately to use the same well known port (1-4 min!) Reason: TCP cannot reallocate the socket pair (i.e. the connection) till 2MSL. If you kill client and restart, it will get a different port 2MSL wait protects against delayed segments from the previous incarnation of the connection. If server crashes and reboots within 2 MSL wait, it is still safe because RFC 793 prevents having connections for 1 MSL after reboot.

46 Simultaneous open/close Figs and Simultaneous open is very rare. Requires same socket pair i.e. both the ports must be well known too. Two simultaneous telnets (A to B and B to A) will not create this because client ports are not well-known. Possible in long RTT cases Requires 4 messages

47 TCP Options Kind Length Meaning 0 1 End of Valid options in header 1 1 No-op 2 4 Maximum Segment Size 3 3 Window Scale Factor 8 10 Timestamp End of Options: Stop looking for further option No-op: Ignore this byte. Used to align the next option on a 4-byte word boundary

48 TCP Servers Most TCP servers are concurrent i.e. separate process to handle each client - for ease of connection management Server listens to well-known port. Socket pair distinguishes connections A separate endpoint in the ESTABLISHED state is associated with each connection One endpoint is used to listen (LISTEN state) for new connections

49 TCP Servers (Contd) Endpoints in the ESTABLISHED state cannot receive SYN packets Possible to wildcard or select specific interfaces (local IP addresses) to listen to. Multiple connection requests => backlog queue of connections established but new process not yet created by server to handle it. Queue full => send RESET to new connection requests

50 TCP Keepalive timer Optional timer. Not part of TCP spec, but found in most implementations. Not necessary, because connection defined by endpoints. Connection can be up as long as source/destination up. Typical use: to detect idle clients or half-open connections and de-allocate server resources tied up to them. Eg: telnet, ftp.

51 July, 2011 Computer Network Concepts 51