Chapter 3: Network and Communication

1. Chapter 3: Network and Communication • What is a network? • What types of network are there? • What networking standards are there? • How do you represent information? • What is communication protocol? • What are communication models? (message-passing, stream communication and RPC) Distributed Systems

2. Anatomy of a network • A set of interconnected resources • Hosts that run network applications software • Clients and servers • Set of peers • The network infrastructure that interconnects the hosts • The networking hardwareand software • Network node devices such as routers and switches • Links: cables, connectors, network interfaces Distributed Systems

3. Types of network • Main types: • LAN, WAN, MAN, and Internet • LAN (Local Area Network) is mainly private • Ether net, Token ring • Or interconnected • WAN (Wide Area Network) can be private or public • Interconnected • MAN (Metropolitan Area Network) is mainly public • Interconnected by Optical fibre • Global Network is public • The internet • The telephone network Distributed Systems

4. To the Internet To offsite LANs Interconnecting LANs and WANs • Host systems usually connect into a LAN switch • Number of hosts limited by the number of ports on the switch • Routers have two main uses • Interconnecting LANs • Connecting to a WAN or to the Internet • Routers interconnect LANs • To separate the users • To separate the traffic switch router Distributed Systems

5. The OSI and IETF Protocol Suites IETF: The Internet Engineering Task Force OSI: Open System Interconnection Reference Model OSI model IETF model Layer 7- Application application protocols layers 5/6/7:Application Layer 6 - Presentation Layer 5 - Session TCP, UDP Layer 4 - Transport IP Layer 3 - Network networking protocols Logical MAC PPP, 802.3,5,11, etc Layer 2 - Link: Layer 1 - Physical Physical Distributed Systems

6. Application Layer Application data Transport header andpayload (e.g. TCP segment) Transport Layer T hdr App data Network header andpayload (e.g. IP packet) Network Layer N hdr Transport Link header and payload (e.g. Ethernet frame) Link Layer L hdr Network CRC Physical Layer 101011100101 Protocol Data Encapsulation Distributed Systems

7. A Typical Message on the Network Distributed Systems

8. Addresses Addresses Webclient DNSclient URL Web server Source port Destination port TCP UDP TCP UDP Transport protocol Transport protocol Encapsulation Encapsulation IP IP Source IP Destination IP Physical network Physical network Source NIC Destination NIC Protocol Data Flow Distributed Systems

9. Communication Models • Message Passing • lowest level of communication, e.g. sockets • unstructured peer-peer IPC • varieties of communication patterns • Data Stream • continuous media • satisfy real time data services • Request / Reply semantics • basis of Client-Server • RPC (Remote Procedure Call) • RMI (Remote Method Invocation) Distributed Systems

10. Message Passing Definitions(1) • Procedures: send, receive, accept, create, connect, locate , reply, acknowledge • Multiplicity: point-to-point,broadcast, multicast • Message Content: data or instruction, by value or by reference (address) • Channels: • - link, port, mailbox • - direction can be uni-diection or bi-direction • - capacity can be unbounded (i.e. asynchronous, no blocking) • or null (implies synchronous) or fixed (implies buffering) • Message Receipt: • explicit receive – receiver can select message, call get, receive • implicit receive – receiver must receive from sender, reference the variable causes message reception (do not have to call get) Distributed Systems

11. Message Passing Definitions(2) • Synchronous/Asynchronous • Synchronous – receiver waits ready for sender message and responds in real time (e.g. phone call). Both sender and receiver return when transfer is complete. No buffer is required. (both must block until comm. ends) • Asynchronous – sender sends message into buffer, message picked up later at receivers convenience (e.g. mailbox). Sender process returns whether or not a message is received. Receiver blocks until the message is available • Blocking/Non-Blocking • Blocking – sender cannot proceed after sending message until receiver picks up message • Non Blocking – sender can continue as soon as message send is done (e.g. added to buffer) • Sender/Receiver Naming • Static – sender and receiver names (location) fixed • Dynamic – names may change (e.g. ask a static name server Distributed Systems

12. Message Passing Definitions(3) • Connection Link • Connection Oriented – link is established and held for duration of service. Guaranteed link but bandwidth may be wasted. • Connectionless – connection not established until message send occurs e.g. different packets sent by different routes • Transient • message is only stored by system while sender and receiver are executing (e.g. MSN messenger) (both must be running) • Persistent • message is stored and delivered by system, even if receiver is not executing (e.g. email) Distributed Systems

13. Persistence and Synchronicity in Communication (1) General organization of a communication system in which hosts are connected through a network Distributed Systems

14. Persistence and Synchronicity in Communication (2) (a) Persistent asynchronous communication (b) Persistent synchronous communication Distributed Systems

15. Persistence and Synchronicity in Communication (3) (c) Transient asynchronous communication (d) Receipt-based transient synchronous communication Distributed Systems

16. Persistence and Synchronicity in Communication (4) (e) Delivery-based transient synchronous communication at message delivery (f) Response-based transient synchronous communication Distributed Systems

17. Socket Programming • A socket is a communication endpoint between processes • A socket forms the API that allows processes to communicate point-to-point over the internet, within a LAN or within a single computer • Each internet host implements the TCP/IP family of protocols • A socket is identified by a socket address consisting of an IP (version 4) address and port number e.g. 129.86.5.20:80 • IP addresses are stored as unsigned 32 bit integer, and frequently represented in dotted decimal notation. • /* Internet address structure */ • structin_addr {unsigned ints_addr;}; • Port numbers are unsigned 16 bit integers (range 0-65535). Port numbers 0-1024 are well known and reserved, e.g. 21 ftp, 23 telnet, 25 email, 80 http... Distributed Systems

18. Socket Families and Types • AF_UNIX – for communicating between processes on the same (UNIX) computer. • AF_INET – for communicating between processes on different machines connected by the internet or a LAN. • SOCK_STREAM is for reliable TCP (Transmission Control Protocol) connection oriented communication that can be for AF_UNIX or AF_INET sockets. These streaming sockets allow for continuous communication. • SOCK_DGRAM is for unreliable UDP (User Datagram Protocol) connectionless communication in which process are not required to connect to the socket continuously. These datagram sockets allow data to be sent in finite packets (or datagrams). The datagram protocol applies only to internet AF_INET sockets. Distributed Systems

19. Socket primitives for TCP/IP Distributed Systems

20. socket() create socket bind() bind local IP address of socket to port place socket in passive mode ready to accept requests listen() take next request from queue (or wait) then forks and create new socket for client connection accept() socket() Blocks until connection from client Issue connection request to server connect() recv() send() Transfer message strings with send/recv or read/write Process request send() recv() Close socket close() close() Server Client TCP/IP Socket Calls for Connection Distributed Systems

21. socket() create socket bind() bind local IP address of socket to port Receive senders address and senders datagram recvfrom() blocks until datagram received from a client socket() request specify senders address and send datagram sendto() Process request reply sendto() recvfrom() close() Close socket Server Client UDP/IP Socket Calls for Connection Distributed Systems

22. UDP/IP Socket program example(1) #include <stdio.h> #include <sys/types.h> #include <sys/socket.h> #include <netinet/in.h> #define PORT_NUM 2222 char message[20]; main(){ /* process：send1.c */ intsend1,fromlen; structsockaddr_insend1_addr, send2_addr; send1 = socket(AF_INET, SOCK_DGRAM, 0); send1_addr.sin_family = AF_INET; send1_addr.sin_addr.s_addr = INADDR_ANY; send1_addr.sin_port = 0; bind(send1, (structsockaddr*)&send1_addr, sizeof(send1_addr)); send2_addr.sin_family = AF_INET; send2_addr.sin_addr.s_addr = inet_addr(“xxx.xxx.xxx.xxx”); send2_addr.sin_port = PORT_NUM; strcpy(message, “send2, hi!”); sendto(send1, message, sizeof(message), 0, (structsockaddr) &send2_addr, sizeof(send2_addr)); fromlen = sizeof(send2_addr); recvfrom(send1, message, sizeof(message), 0, (structsockaddr) &send2_addr, &fromlen); printf(“send2 says: %s\n”, message); close(send1); } Distributed Systems

23. UDP/IP Socket program example(2) #include <stdio.h> #include <sys/types.h> #include <sys/socket.h> #include <netinet/in.h> #define PORT_NUM 2222 char message[20]; main(){ /* process: send2.c */ intsend2,fromlen; structsockaddr_insend1_addr, send2_addr; send2 = socket(AF_INET, SOCK_DGRAM, 0); send2_addr.sin_family = AF_INET; send2_addr.sin_addr.s_addr = INADDR_ANY; send2_addr.sin_port = PORT_NUM; bind(send2, (structsockaddr*)&send2_addr, sizeof(send2_addr)); fromlen = sizeof(send2_addr); recvfrom(send2, message, sizeof(message), 0, (structsockaddr) &send1_addr, &fromlen); printf(“send1 says: %s\n”, message); strcpy(message, “hi send1, how are you!”); sendto(send2, message, sizeof(message), 0, (structsockaddr) &send1_addr, sizeof(send1_addr)); close(send2); } Distributed Systems

24. Stream Oriented Communication sender receiver • Continuous Media: the temporal (time) relationships between different data item are fundamental to correctly interpreting what the data actually means (movies, audio streams). • Discrete Media: the temporal relationships between data items are not important (text, still images). Distributed Systems

25. Data Stream transmission modes • Asynchronous mode: no timing constraints on data stream. • Synchronous mode: there is a max end-to-end delay, but how about too fast? • Isochronous mode: data items are transferred on time, both max delay and min delay. Distributed Systems

26. Request / Reply Model Principle of RPC between a client and server program. Distributed Systems

27. Local Procedure Call main(){ char cip[] = “Buubdl!bu!ebxo”; /* cipher*/ int key = 1;/* secret key */ intlen = decrypt(cip, key); /* LPC */ /* other processing */ } int decrypt(char * s, int key){ /* decryption */ inti = 0; while( *s) { *s -= key; i++; s++;} return i; } call LPC procedure return stack stack stack Return address i -> 0 s -> main.cip key -> 1 before after cip ->Buubdl!bu!ebxo len -> ? key -> 1 cip ->Buubdl!bu!ebxo len -> ? key -> 1 cip ->Attack at dawn len -> 14 key -> 1 Distributed Systems

28. Remote Procedure Call Binding server binder recvreq register or search return (0) program stub (1) (8) stub procedure (5) (6) (2) LPC Bind req Recv bind marshal Send req Recv result unmarsh return recv req unmarsh LPC marshal send result execute return (3) (5) (1) (4) (8) (7) (6) client server Distributed Systems

29. Goal: make RPC transparent • (i.e., behave just like normal (local) procedure call) • Potential Problems • address space (server location) • parameters passing methods • and failures of server or clients • Basic RPC • - normal procedure call : count = read(fd,buf,nbytes) • Parameters passing methods • - call-by-value • - call-by-reference • call-by-copy/restorRPC • implementation uses client stubs in the place actual procedure Distributed Systems

30. Remote Procedure Call: steps (0) Remote procedures registration; (1) Client procedure calls client stub in normal way; (2) Client stub sends a binding request asking for information; (3) Binding server searches for binding and reply to client stub; (4) Client stub packs a message (marshalling) and send to server stub; (5) Server stub unpacks parameters (unmarshalling), invokes LPC; (6) Server procedure executes and returns results to server stub; (7) Server stub packs results (marshalling) and sends to client stub; (8) Client stub unpacks results and returns to client procedure. Call-by-value: parameter is a straight value (int, float, …) Call-by-reference: parameter is a pointer to anything (int, record, array, pointer, …) Distributed Systems

31. Parameter Passing - parameter marshaling -- packing parameters into a msg - n = sum(4,7) must be able to: - handle different data representations EBCDIC, ASCII, Unicode, floating point numbers, little endian (bytes from right to left, 486) and big endian (bytes from left to right, SPARC) -Need to represent data in standard format e.g XDR - how to handle pointers : address at 1000 send a copy of data and restore value on return at a ddress Binding client to sever: Static binding: server address known at complete time Dynamic binding (server location resolved at run time) - how to locate a server: use name server - how to register a server : need registry service at registry directory - Server is identified by (name, version, handle, unique-id) Distributed Systems

32. RPC Semantics in the Presence of Failures (1) - goal: RPC behave like local procedure call 1) cannot locate the server - server may be down - Version number mismatch - In this case : need to raise an exception 2) request msg is lost use timer and resend request msg Distributed Systems

33. Failure semantics continue (2) 3) reply msg is lost - use timer and resent another request - server need to tell whether the request is duplicate unless the request is idempotent Idempotent means: the operation can be executed more than one without harm e.g., read (fd, buf, n) read (fd, buf, pos, n) deposit (money) -- not possible to make it Idempotent ii) assign request nos and keep track Distributed Systems

34. Failure semantics continue (3) • 4) server crashes after receiving a request • unique no not enough since cannot know whether or not RPC done or not • - Possible Semantics • i) at least once (retransmit after time out) • ii) at most once ( do not retransmit request) • iii) exactly once (not possible to implement) • 5) client crashes after sending a request • - make request and then crash • Server computation becomes orphan - potential problems -- waste cycles at server , etc. • how to deal with orphans: extermination for long computations, problem if server is holding a lock Distributed Systems

35. Example: SUN RPC (1) /* eXtended Data Representation (XDR) definition，file name：caesar.x */ const MAX = 100; typedefstruct { /* return type */ intlen; char code[MAX]; } Data; typedefstruct { /* parameter type */ int key; char cipher[MAX]; } Args; program CAESAR { /* CAESAR program */ version VERSION { Data DECRYPT(Args) = 1; /* decryption procedure */ Data ENCRYPT(Args) = 2; /* encryption procedure*/ } = 5; } = 8888; Distributed Systems

36. Example: SUN RPC(2) Invoke XDR compiler rpcgento generate the following files: • Client stub • Server main program and server stub • XDR parameter marshalling/unmarshalling functions • Program header file, caesar.h, which includes constants, user defined types, remote procedure prototypes. Now, we are ready to design other programs. Distributed Systems

37. Example: SUN RPC(3) /* client program file：client.c */ #include <rpc/rpc.h> #include “caesar.h” main(){ CLIENT *cp; char *serverName = “Caesar_server”; Argsarg; Data * plaintext; /* create client pointer */ cp = clnt_create(serverName, CRESAR, VERSION, “udp”); if (cp == NULL) exit(1); arg.key = 1; /* set RPC parameters */ arg.cipher = “Buubdl!bu!ebxo”; plaintext = decrypt_2(&arg, cp); /* issue RPC */ /* other processing */ … clnt_destroy(cp); /* delete client pointer */ } /* server program file：server.c */ #include <rpc/rpc.h> #include “ceasar.h” Data* decrypt_2(Args *a){ /* decryption */ static Data output; /* must be static */ char s = a->cipher; int i = 0; while( *s) { output.code[i] = *s - key; i++; s++;} output.len = i; return &output; /* return result */ } Data* encrypt_2(args *a){ /* encryption */ /* … */ } Distributed Systems

38. Example: SUN RPC(4) Server program Server code Server stub RPC library XDR Definition Header file client stub Client program Client code The steps in writing a client and a server in SUN RPC Distributed Systems

39. The steps in writing a client and a server in DCE RPC IDL: Interface Definition Language; uuidgen: IDL file generator DCE: Distributed Computing Environment (Open Software foundation) Distributed Systems

40. Re-sending RPC Replica filtering Re-sending results RPCsemantics no no no maybe yes no no at-least-once yes yes no maybe-once yes yes yes at-most-once RPC Semantics LPC has exact-once semantics, how about RPC? Server dead? RPC request lost? Reply lost? • Re-sending RPC (time out) • Replica filtering • Re-sending results Distributed Systems

41. Remote Method Invocation • Remote Method Invocation (RMI) is a Java mechanism similar to RPCs. (java implementation of RPC) • RMI allows a Java program on one machine to invoke a method on a remote object.

42. Marshalling Parameters

43. Remote Method Invocation • RMI and RPC differs in two ways: • RPCs support procedural programming whereby only remote procedures or functions may be called. RMI is object based: It supports invocation of methods on remote objects. • The parameters to remote procedures are ordinary data structures in RPC; with RMI it is possible to pass objects as parameters to remote methods. • If the marshaled parameters are local (non remote) objects, they are passed by copy using a technique known as object serialization. • Object serialization allowed the state of an object to be written to a byte stream.

44. Creating a Distributed System with RMI • Four major steps • Define remote interface • Describes client/server communication • Client calls remote methods, server implements them • Define server application to implement remote interface • Same name as remote interface, ends with Impl • Define client application that uses remote interface reference • Interacts with server implementation • Compile and execute server and client