User:Jsoon eu/thoughts on protocol

A communications protocol is a formal description of digital message formats and the rules for exchanging those messages in or between computing systems and in telecommunications. Protocols may include signaling, authentication and error detection and correction capabilities. A protocol describes the syntax, semantics, and synchronization of communication and may be implemented in hardware or software, or both.

Protocols are used to structure communications in and between systems. Together with programming languages and operating systems, protocols form the foundation on which modern computers are build.

Introduction
In a diplomatic context the word protocol refers to a diplomatic document or a rule,guideline etc which guides diplomatic behaviour. Synonyms are procedure and policy. While there is no generally accepted formal definition of "protocol" in computer science, an informal definition, based on the previous, could be "a description of a set of procedures to be followed when communicating". In computer science the word algorithm is a synonym for the word procedure, so a ''protocol is to communications what an algorithm is to computations. 

Communicating systems use well-defined formats for exchanging messages. Each message has an exact meaning intended to provoke a defined response of the receiver. A protocol therefore describes the syntax, semantics, and synchronization of communication. A programming language describes the same for computations, so there is a close analogy between protocols and programming languages: protocols are to communications what programming languages are to computations.

Diplomatic documents build on each other, thus creating document-trees. The way the sub-documents making up a document-tree are written has an impact on the complexity of the tree. By imposing a development model on the documents, overall readability can be improved and complexity can be reduced. An effective model to this end is the layering scheme or model. In a layering scheme the documents making up the tree are thought to belong to classes, called layers. The distance of a sub-document to its root-document is called its level. The level of a sub-document determines the class it belongs to. The sub-documents belonging to a class all provide similar functionality and, when form follows function, have similar form.

The communications protocols in use on the Internet are designed to function in very complex and diverse settings, so they tend to be very complex. Unreliable transmission links add to this by making even basic requirements of protocols harder to achieve. To ease design, communications protocols are also structured using a layering scheme as a basis. Instead of using a single universal protocol to handle all transmission tasks, a set of cooperating protocols fitting the layering scheme is used. The layering scheme in use on the Internet is called the TCP/IP model. The actual protocols are collectively called the Internet protocol suite. The group responsible for this design is called the Internet Engineering Task Force (IETF).

Obviously the number of layers of a layering scheme and the way the layers are defined can have a drastic impact on the protocols involved. This is where the analogies come into play for the TCP/IP model, because the designers of TCP/IP employed the same techniques used to conquer the complexity of programming language compilers (design by analogy) in the implementation of its protocols and its layering scheme.

Like diplomatic protocols, communications protocols have to be agreed upon by the parties involved. To reach agreement a protocol is developed into a technical standard. International standards are developed by the International Organization for Standardization (ISO).

Communicating systems
The information exchanged between devices on a network or other communications medium is governed by rules (conventions) that can be set out in a technical specification called a communication protocol standard. The nature of the communication, the actual data exchanged and any state-dependent behaviors are defined by the specification. This approach is often taken for protocols in use by telecommunications.

In digital computing systems, the rules can be expressed by algorithms and datastructures, raising the opportunity of hardware independency. Expressing the algorithms in a portable programming language, makes the protocol software operating system independent. The protocols in use by an operating system itself, lend themselves to be described this way and are usually, just like the rest of the operating system, distributed in binary or source form.

Operating systems are usually conceived of as consisting of a set of cooperating processes that manipulate a shared store (on the system itself) to communicate with each other. This communication is governed by well understood protocols and is only a small part of what a process is supposed to accomplish (managing system resources like cpu's, memory, timers, I/O devices etc, and providing controlled access to the resources), so these protocols can be embedded in the process code itself as small additional code fragments.

In contrast, communicating systems have to communicate with each other using shared transmission media, because there is no common memory. Unlike a memory store operation, a transmission doesnot need to be reliable and can involve different hardware and operating systems on different systems. This complicates matters up to a point that some kind of structuring is necessary to conquer the complexity of networking protocols, especially, when used on the Internet. The communicating systems can make use of different operating systems, as long as they agree to use the same kind of structuring and the same protocols for their communications.

To implement a networking protocol, the protocol software modules are to be interfaced with a framework assumed to be implemented on the machine's operating system. This framework implements the networking functionality of the operating system. Obviously, the framework needs to be as simple as it can be, to allow for an easier incorporation into the operating systems. The best known frameworks are the TCP/IP model and the OSI model.

At the time the Internet was formed, layering had proven to be a successful design approach for both compiler and operating system design and given the similarities between programming languages and communication protocols, it was intuitively felt that layering should be applied to the protocols as well. This gave rise to the concept of layered protocols which nowadays forms the basis of protocol design.

Systems do not use a single protocol to handle a transmission. Instead they use a set of cooperating protocols, sometimes called a protocol family or protocol suite. Some of the best known protocol suites include: IPX/SPX, X.25, AX.25, AppleTalk and TCP/IP. To cooperate the protocols have to communicate with each other, so there is an unnamed 'protocol' to do this. A technique used by this 'protocol' is called encapsulation, which makes it possible to pass messages from layer to layer in the framework.

The protocols can be arranged on functionality in groups, for instance there is a group of transport protocols. The functionalities are mapped on the layers, each layer solving a distinct class of problems relating to, for instance: application-, transport-, internet- and network interface-functions. To transmit a message, a protocol has to be selected from each layer, so some sort of multiplexing/demultiplexing takes place. The selection of the next protocol, also part of the aforementioned 'protocol' is accomplished by extending the message with a protocolselector for each layer.

There's a myriad of protocols, but they all only differ in the details. For this reason the TCP/IP protocol suite can be studied to get the overall protocol picture. The Internet Protocol (IP) and the Transmission Control Protocol (TCP) are the most important of these, and the term Internet Protocol Suite, or TCP/IP, refers to a collection of its most used protocols. Most of the communication protocols in use on the Internet are described in the Request for Comments (RFC) documents of the Internet Engineering Task Force (IETF). RFC1122, in particular, documents the suite itself.

Basic requirements of protocols
The data representing the messages is to be sent and received on communicating systems to establish communications. Protocols should therefore specify rules governing the transmission. In general, much of the following should be addressed:
 * Data formats for data exchange. In digital message bitstrings are exchanged. The bitstrings are divided in fields and each field carries information relevant to the protocol. Conceptually the bitstring is divided into two parts called the header area and the data area. The actual message is stored in the data area, so the header area contains the fields with more relevance to the protocol. The transmissions are limited in size, because the number of transmission errors is proportional to the size of the bitstrings being send. Bitstrings longer than the maximum transfer unit (MTU) are divided in pieces of appropriate size. Each piece has almost the same header area contents, because only some fields are dependent on the contents of the data area (notably CRC fields, containing checksums that are calculated from the data area contents).


 * Address formats for data exchange. The addresses are used to identify both the sender and the intended receiver(s). The addresses are stored in the header area of the bitstrings, allowing the receivers to determine whether the bitstrings are intended for themselves and should be processed or (when not to be processed) should be discarded. A connection between a sender and a receiver can be identified using an address pair (sender address, receiver address). Usually some address values have special meanings. An all-1s address could be taken to mean all stations on the network, so sending to this address would result in a broadcast on the local network. Likewise, an all-'0's address could be taken to mean the sending station itself (as a synonym of the actual address). Stations have addresses unique to the local net, so usually the address is conceptually divided in two parts: a network address and the station address. The network address uniquely identifies the network on the internetwork (a network of networks). The rules describing the meanings of the address value are collectively called an addressing scheme.


 * Address mapping. Sometimes protocols need to map addresses of one scheme on addresses of another scheme. For instance to translate a logical IP address specified by the application to a hardware address. This is referred to as address mapping. The mapping is implied in hierarchical address schemes where only a part of the address is used for the map address. In other cases the mapping needs to be described using tables.


 * Routing. When systems are not directly connected, intermediary systems along the route to the intended receiver(s) need to forward messages (instead of discarding them) on behalf of the sender. Determining the route the message should take is called routing. On the Internet, the networks are connected using routers (gateways). This way of connecting networks is called internetworking. To determine the next router on the path to the destination, all systems consult locally stored tables consisting of (destination network address, delivery address) - entries and a special entry consisting of (a 'catch-all' address, default router address). The delivery address is either the address of a router assumed to be closer to the destination and the hardware interface to be used to reach it, or the address of a hardware interface on the system directly connecting a network. The default router is used when no other entry matches the intended destination network.


 * Detection of transmission errors is necessary, because no network is error-free. Bits of the bitstring become corrupted or lost. Usually, CRCs of the data area are added to the end of packets, making it possible for the receiver to notice many (nearly all) differences caused by errors, whilst recalculating the CRCs of the received packet and comparing them with the CRCs given by the sender. The receiver rejects the packets on CRC differences and arranges somehow for retransmission.


 * Acknowledgements of correct reception of packets by the receiver are usually used to prevent the sender from retransmitting the packets. Some protocols, notably datagram protocols like the Internet Protocol (IP), do not acknowledge.


 * Loss of information - timeouts and retries. Sometimes packets are lost on the network or suffer from long delays. To cope with this, a sender expects an acknowledgement of correct reception from the receiver within a certain amount of time. On timeouts, the packet is retransmitted. In case of a broken link the retransmission has no effect, so the number of retransmissions is limited. Exceeding the retry limit is considered an error.


 * Direction of information flow needs to be addressed if transmissions can only occur in one direction at a time (half-duplex links). To gain control of the link a sender must wait until the line becomes idle and then send a message indicating its wish to do so. The receiver responds by acknowledging and waits for the transmissions to come. The sender only begins transmitting after the acknowledgement. Arrangements have to be made to accommodate the case when two parties want to gain control at the same time.


 * Sequence control. We have seen that long bitstrings are divided in pieces, that are send on the network individually. The pieces may get 'lost' on the network or arrive out of sequence, because the pieces can take different routes to their destination. Sometimes pieces are needlessly retransmitted, due to network congestion, resulting in duplicate pieces. By sequencing the pieces at the sender, the receiver can determine what was lost or duplicated and ask for retransmissions. Also the order in which the pieces are to be processed can be determined.


 * Flow control is needed when the sender transmits faster than the receiver can process the transmissions or when the network becomes congested. Sometimes, arrangements can be made to slow down the sender, but in many cases this is outside the control of the protocol.

Getting the data across is only part of the problem. The data received has to be evaluated in the context of the progress of the conversation, so a protocol has to specify rules describing the context and explaining whether the (form of the) data fits this context or not. These kind of rules are said to express the syntax of the communications. Other rules determine whether the data is meaningful for the context in which the exchange takes place. These kind of rules are said to express the semantics of the communications.

Both intuitive descriptions as well as more formal specifications in the form of finite state machine models are used to describe the expected interactions of the protocol. Formal ways for describing the syntax of the communications are Abstract Syntax Notation One (a ISO standard) or Augmented Backus-Naur form (a IETF standard).

An example protocol: the ethernet protocol
To get a feel for protocols and what needs to be described, the ethernet protocol is explained in some detail.

Ethernet was invented at Xerox PARC in the early 1970s. It has gained widespread use on LANs and became standardized as IEEE standard 802.3. To connect to a LAN, a computer has to be equipped with an ethernet network interface card. The combination of computer and interface card is called a station. Using wiring, interface cards are all connected to a hub.

Conceptually, all stations share a single communication channel called a shared bus. Transmissions on this channel are received by all stations at (nearly) the same time. The hardware provides no indication to the sender about whether the transmission was delivered and is therefore called a best-effort delivery mechanism. A carrier wave is used to transmit the data in packets, referred to as ethernet frames. Each station is constantly tuned in on the channel, so each transmission is noticed. In order to determine whether the channel is free, the carrier wave can be sensed by the hardware; if not present the channel is free for transmission.

Stations wanting to transmit, wait for the channel to become free and then start transmitting one single frame and then stop for a small amount of time before transmitting a next frame to allow others to transmit. The transmissions take time to reach all the stations due to the limited travelling speed of the signal (approximately 70% of the speed of light), so sometimes two stations start transmitting very shortly after each other, creating what is called collisions, because the second station notices too late that the channel is not free. The result is a scrambled signal, so the frames need to be retransmitted. During the transmissions, the hardware monitors for interference and if detected stops transmitting. To avoid further collisions, each station should wait for a different amount of time before starting the retransmission. The amount of time to wait is picked randomly from a range using an exponential backoff policy, which ensures that further collisions become increasingly improbable by increasing the random range exponentially (2n) on each retry.

Ethernet cards are assigned unique 48 bit numbers, known as ethernet addresses, which are used to address the stations. To broadcast a message, a special broadcast address (all 1s) is used. Ethernet establishes link level connections, which can be defined using both the destination and sources addresses. The frame format is as follows: a preamble (8 octets), destination address (6 octets), source address (6 octets), frame type (2 octets), frame data (46-1500 octets), a CRC (4 octets). The preamble is used to synchronize the receiving interface. It consists of 64 bits of alternating 0s and 1s. This bit pattern is relatively easy to detect and synchronize to, and therefore suitable as a start marker of the transmission. The frame type is a 16 bit integer identifying the type of data in the frame data. The cyclic redundancy check (CRC) is derived from the frame data and used by the receiver to verify the integrity of the transmitted data.

On reception of a transmission, the receiver uses the destination address to determine whether the transmission is relevant to the station or should be ignored. Both broadcast address and a destination address equal to the address of the receiving interface card are to be handled. The frame type is used by the operating system on the receiving station to select the appropriate protocol module (i.e. the ethernet protocol module). Ethernet frames are said to be self-identifying, because of the frame type. Self-identifying frames make it possible to intermix multiple protocols on the same physical network and allow a single computer to use multiple protocols together.

The details given are from a networking point of view. Data transmission details like the electrical properties of the cable and the electrical circuits used for signal modulation are outside the scope of this article, but should obviously be described by a communication protocol standard as well.

Such a protocol is called a low-level protocol. The Internet Protocol specifies higher level abstractions like IP addressing, datagram format, and the concept of unreliable, connectionless delivery, and is therefore called a high-level protocol.
 * Programming like this would require knowledge about the hardware, like the details of the Ethernet frame format, network access policy, and frame error handling.

Protocols and programming languages
The following two analogies between communications and computations are used throughout this text:


 * Protocols are to communications what algorithms are to computations.


 * Protocols are to communications what programming languages are to computations.

The analogies have important consequences for both the design and the development of protocols.

The word protocol has different meanings in the two analogies: 1. a variant of (or something like) an algorithm. 2. a variant of (or something like) a programming language. Arguably, it would make more sense to speak of a protocolling language (compare algorithmic language) when using the word protocol in its second meaning, but this is not a standard practice in a networking context. To further clarify on this, one has to consider the fact that algorithms, programs and protocols are just different ways of describing expected behaviour of interacting objects. Assuming the use of pseudo-code in the case of both algorithms and protocols, all of the representations can be referred to as applications of programming languages. In other words protocols are applications of protocolling languages.

A familiar example of a protocolling language is the HTML language used to describe webpages which are the actual webprotocols.

In programming languages the association of identifiers to a value is termed a definition. Program text is structured using block constructs and definitions can be local to a block. The localized association of an identifier to a value established by a definition is termed a binding and the region of program text in which a binding is effective is known as its scope. The computational state is kept using two components: the environment, used as a record of identifier bindings, and the store, which is used as a record of the effects of assignments.

In communications, message values are transferred using transmission media. By analogy, the equivalent of a store would be a collection of transmission media, instead of a collection of memory locations. A valid assignment in a protocol (as an analog of programming language) could be  Ethernet:='message' , meaning a message is to be broadcast on the local ethernet.

On a transmission medium there can be many receivers. For instance a mac-address identifies an ether network card on the transmission medium (the 'ether'). In our imaginary protocol, the assignment ethernet[mac-address]:=message value could therefore make sense.

By extending the assignment statement of an existing programming language with the semantics described, a protocolling language could easily be imagined.

Operating systems provide reliable communication and synchronization facilities for communicating objects confined to the same system by means of system libraries. A programmer using a general purpose programming language (like C or  ADA) can use the routines in the libraries to implement a protocol, instead of using a dedicated protocolling language.

Language layering
Early compilers translated high-level language sources to machine code. Instead of translating directly into machine code, modern compilers translate to a machine independent intermediate code in order to enhance portability of the compiler and minimize design efforts. The intermediate language defines a virtual machine that can execute all programs written in the intermediate language (a machine is defined by its language and vice versa). The intermediate code instructions are translated into equivalent machine code sequences by a code generator to create executable code. It is also possible to skip the generation of machine code by actually implementing the virtual machine in machine code. This virtual machine implementation is called an interpreter, because it reads in the intermediate code instructions one by one and after each read executes the equivalent machine code sequences (the interpretation) of the read intermediate instruction directly.

Translation of a programming language into intermediate code and intermediate code into machine code to generate an executable is an example of layering of languages. Because languages define machines, this can be viewed as a layering of virtual machines as well. Using machines (or mechanisms) to build more complex machines, which in their turn are used for even more complex machines etcetera, results in what is referred to as a multi-level machine.

A modern computer system is usually a six level machine with the following levels (also called layers) present: the digital logic level, the microprogramming level, the conventional machine level, the operating system level, the assembly language level and the problem-oriented programming language level.

The Internet can be viewed as a layered machine as well: a (best-effort) hardware delivery mechanism is used to build a connectionless packet delivery system on top of which a reliable transport system is build, which is used to build an application. The delivery system is defined by the IP protocol and the transport system by the TCP protocol.

Universal protocols
Despite their numbers, networking protocols show little variety, because all networking protocols use the same underlying principles and concepts, in the same way. So, the use of a general purpose programming language would yield a large number of applications only differing in the details. A suitably defined (dedicated) protocolling language would therefore have little syntax, perhaps just enough to specify some parameters or optional modes of operation, because its virtual machine would have incorporated all possible principles and concepts making the virtual machine itself a universal protocol. The protocolling language would have some syntax and a lot of semantics describing this universal protocol and would therefore in effect be a protocol, hardly differing from this universal networking protocol. In this (networking) context a protocol is a language.

The notion of a universal networking protocol provides a rationale for standardization of networking protocols; assuming the existence of a universal networking protocol, development of protocol standards using a consensus model (the agreement of a group of experts) might be a viable way to coordinate protocol design efforts.

Networking protocols operate in very heterogeneous environments consisting of very different network technologies and a (possibly) very rich set of applications, so a single universal protocol would be very hard to design and implement correctly. Instead, the IETF decided to reduce complexity by assuming a relatively simple network architecture allowing decomposition of the single universal networking protocol into two generic protocols, TCP and IP, and two classes of specific protocols, one dealing with the low-level network details and one dealing with the high-level details of common network applications (remote login, file transfer, email and web browsing). ISO choose a similar but more general path, allowing other network architectures, to standardize protocols.

Protocol design
Communicating systems operate in parallel. The programming tools and techniques for dealing with parallel processes are collectively called concurrent programming. Concurrent programming only deals with the synchronization of communication. The syntax and semantics of the communication governed by a low-level protocol usually have modest complexity, so they can be coded with relative ease. High-level protocols with relatively large complexity could however merit the implementation of language interpreters. An example of the latter case is the HTML language.

Concurrent programming has traditionally been a topic in operating systems theorie texts. Formal verification seems indispensable, because concurrent programs are notorious for the hidden and sophisticated bugs they contain. A mathematical approach to the study of concurrency and communication is referred to as Communicating Sequential Processes (CSP). Concurrency can also be modelled using finite state machines like Mealy- and Moore machines. Mealy- and Moore machines are in use as design tools in digital electronics systems, which we encounter in the form of hardware used in telecommunications or electronic devices in general. This kind of design can be a bit of a challenge to say the least, so it is important to keep things simple. For the Internet protocols, in particular and in retrospect, this meant a basis for protocol design was needed to allow decomposition of protocols into much simpler, cooperating protocols.

Concurrent programming
A concurrent program is an abstraction of cooperating processes suitable for formal treatment and study. The goal of the abstraction is to prove correctness of the program assuming the existence of some basic synchronization or data exchange mechanisms provided by the operating system (or other software) or hardware. The mechanisms are complex, so more convenient higher level primitives are implemented with these mechanisms. The primitives are used to construct the concurrent program. The basic primitive for synchronization is the semaphore. All other primitives (locks, reentrant mutexes, semaphores, monitors, message passing, tuple space) can be defined using semaphores. The semaphore is sufficiently elementary to be successfully studied by formal methods.

In order to synchronize or exchange data the processes must communicate by means of either a shared memory, used to store data or access-restricted procedures, or the sending/receiving of signals (message passing) using a shared transmission medium. Most third generation operating systems implement seperate processes that use special instructions to ensure only one process can execute the restricted procedures. On distributed systems there is no common central memory so the communications are always by means of message passing. In this case the processes simply have to wait for each other (synchronization by rendezvous) before exchanging data.

Conceptually, the concurrent program consists of several sequential processes whose execution sequences are interleaved. The execution sequences are divided into sections. A section manipulating shared resources is called a critical section. The interleaving scheme makes no timing assumptions other than that no process halts in its critical section and that ready processes are eventually scheduled for execution. For correct operation of the program, the critical sections of the processes need to be properly sequenced and synchronized. This is achieved using small code fragments (protocols) at the start and the end of the critical sections. The code fragments determine whether the critical sections of two communicating processes should execute in parallel (rendezvous of processes) or should be executed sequentially (mutual exclusion of processes).

A concurrent program is correct if it does not violate some safety property such as mutual exclusion or rendezvous of critical sections and does not suffer of liveness properties such as deadlock or lockout. Correctness of the concurrent program can only be shown using a mathematical argument. Specifications of concurrent programs can be formulated using formal logics (like CSP) which make it possible to prove properties of the programs. Incorrectness can be shown using execution scenarios.

Mutual exclusion is extensively studied in the mutual exclusion problem. The rendezvous is studied in the producer-consumer problem in which a producer process only produces data if and only if the consumer process is ready to consume the data. Although both problems only involve two processes, their solutions require rather complex algorithms (Dekker's algorithm, Lamport's bakery algorithm). The readers-writers problem is a generalization of the mutual exclusion problem. The dining philosophers problem is a classical problem sufficiently difficult to expose many of the potential pitfalls of newly proposed primitives.

A basis for protocol design
Systems do not use a single protocol to handle a transmission. Instead they use a set of cooperating protocols, sometimes called a protocol family or protocol suite. To cooperate the protocols have to communicate with each other, so some kind of conceptual framework is needed to make this communication possible. Also note that software is needed to implement both the 'xfer-mechanism' and a protocol (no protocol, no communication).

In literature there are numerous references to the analogies between computer communication and programming. By analogy we could say that the aforementioned 'xfer-mechanism' is comparable to a cpu; a 'xfer-mechanism' performs communications and a cpu performs computations and the 'framework' introduces something that allows the protocols to be designed independent of one and another by providing separate execution environments for the protocols. Furthermore, it is repeatedly stated that protocols are to computer communication what programming languages are to computation.

Protocol layering
Protocol layering now forms the basis of protocol design. It allows the decomposition of single, complex protocols into simpler, cooperating protocols, but it is also a functional decomposition, because each protocol belongs to a functional class, called a protocol layer. The protocol layers each solve a distinct class of communications problems. The Internet protocol suite consists of the following layers: application-, transport-, internet- and network interface-functions. Together, the layers make up a layering scheme or model.

In computations, we have algorithms and data, and in communications, we have protocols and messages, so the analog of a data flow diagram would be some kind of message flow diagram. To visualize protocol layering and protocol suites, a diagram of the message flows in and between two systems, A and B, is shown in figure 3.

The systems both make use of the same protocol suite. The vertical flows (and protocols) are in system and the horizontal message flows (and protocols) are between systems. The message flows are governed by rules, and dataformats specified by protocols. The blue lines therefore mark the boundaries of the (horizontal) protocol layers.

The vertical protocols are not layered because they don't obey the protocol layering principle which states that a layered protocol is designed so that layer n at the destination receives exactly the same object sent by layer n at the source. The horizontal protocols are layered protocols and all belong to the protocol suite. Layered protocols allow the protocol designer to concentrate on one layer at a time, without worrying about how other layers perform.

The vertical protocols neednot be the same protocols on both systems, but they have to satisfy some minimal assumptions to ensure the protocol layering principle holds for the layered protocols. This can be achieved using a technique called Encapsulation.

Usually, a message or a stream of data is divided into small pieces, called messages or streams, packets, IP datagrams or network frames depending on the layer in which the pieces are to be transmitted. The pieces contain a header area and a data area. The data in the header area identifies the source and the destination on the network of the packet, the protocol, and other data meaningful to the protocol like CRC's of the data to be send, data length, and a timestamp.

The rule enforced by the vertical protocols is that the pieces for transmission are to be encapsulated in the data area of all lower protocols on the sending side and the reverse is to happen on the receiving side. The result is that at the lowest level the piece looks like this: 'Header1,Header2,Header3,data' and in the layer directly above it: 'Header2,Header3,data' and in the top layer: 'Header3,data', both on the sending and receiving side. This rule therefore ensures that the protocol layering principle holds and effectively virtualizes all but the lowest transmission lines, so for this reason some message flows are coloured red in figure 3.

To ensure both sides use the same protocol, the pieces also carry data identifying the protocol in their header.

The design of the protocol layering and the network (or Internet) architecture are interrelated, so one cannot be designed without the other. Some of the more important features in this respect of the Internet architecture and the network services it provides are described next.


 * The Internet offers universal interconnection, which means that any pair of computers connected to the Internet is allowed to communicate. Each computer is identified by an address on the Internet. All the interconnected physical networks appear to the user as a single large network. This interconnection scheme is called an internetwork or internet.
 * Conceptually, an Internet addresses consists of a netid and a hostid. The netid identifies a network and the hostid identifies a host. The term host is misleading in that an individual computer can have multiple network interfaces each having its own Internet address. An Internet Address identifies a connection to the network, not an individual computer. The netid is used by routers to decide where to send a packet.
 * Network technology independence is achieved using the low-level address resolution protocol (ARP) which is used to map Internet addresses to physical addresses. The mapping is called address resolution. This way physical addresses are only used by the protocols of the network interface layer. The TCP/IP protocols can make use of almost any underlying communication technology.
 * Message flows and Routing.svgPhysical networks are interconnected by routers. Routers forward packets between interconnected networks making it possible for hosts to reach hosts on other physical networks. The message flows between two communicating system A and B in the presence of a router R are illustrated in figure 4. Datagrams are passed from router to router until a router is reached that can deliver the datagram on a physically attached network (called direct delivery). To decide whether a datagram is to be delivered directly or is to be send to a router closer to the destination, a table called the IP routing table is consulted. The table consists of pairs of networkids and the paths to be taken to reach known networks. The path can be an indication that the datagram should be delivered directly or it can be the address of a router known to be closer to the destination.  A special entry can specify that a default router is chosen when there are no known paths.
 * All networks are treated equal. A LAN, a WAN or a point-to-point link between two computers are all considered as one network.
 * A Connectionless packet delivery (or packet-switched) system (or service) is offered by the Internet, because it adapts well to different hardware, including best-effort delivery mechanisms like the ethernet. Connectionless delivery means that the messages or streams are divided in pieces that are multiplexed separately on the high speed intermachine connections allowing the connections to be used concurrently. Each piece carries information identifying the destination. The delivery of packets is said to be unreliable, because packets may be lost, duplicated, delayed or delivered out of order without notice to the sender or receiver. Unreliability arises only when resources are exhausted or underlying networks fail. The unreliable connectionless delivery system is defined by the Internet Protocol (IP). The protocol also specifies the routing function, which chooses a path over which data will be send. It is also possible to use TCP/IP protocols on connection oriented systems. Connection oriented systems build up virtual circuits (paths for exclusive use) between senders and receivers. Once build up the IP datagrams are send as if they were data through the virtual circuits and forwarded (as data) to the IP protocol modules. This technique, called tunneling, can be used on X.25 networks and ATM networks.
 * A reliable stream transport service using the unreliable connectionless packet delivery service is defined by the transmission control protocol (TCP). The services are layered as well and the application programs residing in the layer above it, called the application services, can make use of TCP. Programs wishing to interact with the packet delivery system itself can do so using the user datagram protocol (UDP).

Software layering
Having established the protocol layering and the protocols, the protocol designer can now resume with the software design. The software has a layered organization and its relationship with protocol layering is visualized in figure 5. The software modules implementing the protocols are represented by cubes. The information flow between the modules is represented by arrows. The (top two horizontal) red arrows are virtual. The blue lines mark the layer boundaries.

To send a message on system A, the top module interacts with the module directly below it and hands over the message to be encapsulated. This module reacts by encapsulating the message in its own data area and filling in its header data in accordance with the protocol it implements and interacts with the module below it by handing over this newly formed message whenever appropriate. The bottom module directly interacts with the bottom module of system B, so the message is send across. On the receiving system B the reverse happens, so ultimately (and assuming there were no transmission errors or protocol violations etc.) the message gets delivered in its original form to the topmodule of system B.

On protocol errors, a receiving module discards the piece it has received and reports back the error condition to the original source of the piece on the same layer by handing the error message down or in case of the bottom module sending it across.

The division of the message or stream of data into pieces and the subsequent reassembly are handled in the layer that introduced the division/reassembly. The reassembly is done at the destination (i.e. not on any intermediate routers).

TCP/IP software is organized in four layers.
 * Application layer. At the highest layer, the services available across a TCP/IP internet are accessed by application programs. The application chooses the style of transport to be used which can be a sequence of individual messages or a continuous stream of bytes. The application program passes data to the transport layer for delivery.
 * Transport layer. The transport layer provides communication from one application to another. The transport layer may regulate flow of information and provide reliable transport, ensuring that data arrives without error and in sequence. To do so, the receiving side sends back acknowledgments and the sending side retransmits lost pieces called packets. The stream of data is divided into packets by the module and each packet is passed along with a destination address to the next layer for transmission. The layer must accept data from many applications concurrently and therefore also includes codes in the packet header to identify the sending and receiving application program.
 * Internet layer. The Internet layer handles the communication between machines. Packets to be send are accepted from the transport layer along with an identification of the receiving machine. The packets are encapsulated in IP datagrams and the datagram headers are filled. A routing algorithm is used to determine if the datagram should be delivered directly or send to a router. The datagram is passed to the appropriate network interface for transmission. Incoming datagrams are checked for validity and the routing algorithm is used to decide whether the datagram should be processed locally or forwarded. If the datagram is addressed to the local machine, the datagram header is deleted and the appropriate transport protocol for the packet is chosen. ICMP error and control messages are handled as well in this layer.
 * Network interface layer. The network interface layer is responsible for accepting IP datagrams and transmitting them over a specific network. A network interface may consist of a device driver or a complex subsystem that uses its own data link protocol.

Program translation has been divided into four subproblems: compiler, assembler, link editor, and loader. As a result, the translation software is layered as well, allowing the software layers to be designed independently. Noting that the ways to conquer the complexity of program translation could readily be applied to protocols because of the analogy between programming languages and protocols. The designers of the TCP/IP protocol suite were keen on imposing the same layering on the software framework. This can be seen in the TCP/IP layering by considering the translation of a pascal program (message) that is compiled (function of the application layer)  into an assembler program that is assembled (function of the transport layer) to object code (pieces) that is linked (function of the Internet layer) together with library object code (routing table) by the link editor, producing relocatable machine code (datagram)  that is passed to the loader which fills in the memory locations (ethernet addresses) to produce executeable code (network frame) to be loaded (function of the network interface layer) into physical memory (transmission medium). To show just how closely the analogy fits, the terms between parentheses in the previous sentence denote the relevant analogs and the terms written cursively denote data representations. Program translation forms a linear sequence, because each layer's output is passed as input to the next layer. Furthermore, the translation process involves multiple data representations. We see the same thing happening in protocol software where multiple protocols define the datarepresentations of the data passed between the software modules.

The network interface layer uses physical addresses and all the other layers only use IP addresses. The boundary between network interface layer and Internet layer is called the high-level protocol address boundary. The modules below the application layer are generally considered part of the operating system. Passing data between these modules is much less expensive than passing data between an application program and the transport layer. The boundary between application layer and transport layer is called the operating system boundary.

Strict layering
Strictly adhering to a layered model, a practice known as strict layering, is not always the best approach to networking. Strict layering, can have a serious impact on the performance of the implementation, so there is at least a trade-off between simplicity and performance. Another, perhaps more important point can be shown by considering the fact that some of the protocols in the Internet Protocol Suite cannot be expressed using the TCP/IP model, in other words some of the protocols behave in ways not described by the model. To improve on the model, an offending protocol could, perhaps be split up into two protocols, at the cost of one or two extra layers, but there is a hidden caveat, because the model is also used to provide a conceptual view on the suite for the intended users. There is a trade-off to be made here between preciseness for the designer and clarity for the intended user.

Protocol development
For communication to take place, protocols have to be agreed upon. Recall that in digital computing systems, the rules can be expressed by algorithms and datastructures, raising the opportunity of hardware independency. Expressing the algorithms in a portable programming language, makes the protocolsoftware operating system independent. The sourcecode could be considered a protocol specification. This form of specification, however is not suitable for the parties involved.

For one thing, this would enforce a source on all parties and for another, proprietary software producers would not accept this. By describing the software interfaces of the modules on paper and agreeing on the interfaces, implementers are free to do it their way. This is referred to as source independency. By specifying the algorithms on paper and detailing hardware dependencies in an unambiguous way, a paper draft is created, that when adhered to and published, ensures interoperability between software and hardware.

Such a paper draft can be developed into a protocol standard by getting the approval of a standards organization. To get the approval the paper draft needs to enter and successfully complete the standardization process. This activity is referred to as protocol development. The members of the standards organization agree to adhere to the standard on a voluntary basis. Often the members are in control of large market-shares relevant to the protocol and in many cases, standards are enforced by law or the government, because they are thought to serve an important public interest, so getting approval can be very important for the protocol. It should be noted though that in some cases protocol standards are not sufficient to gain widespread acceptance i.e. sometimes the sourcecode needs to be disclosed enforced by law or the government in the interest of the public.

The need for protocol standards
The need for protocol standards can be shown by looking at what happened to the bi-sync protocol (BSC) invented by IBM. BSC is an early link-level protocol used to connect two separate nodes. It was originally not intended to be used in a multinode network, but doing so revealed several deficiencies of the protocol. In the absence of standardization, manufacturers and organizations felt free to 'enhance' the protocol, creating incompatible versions on their networks. In some cases, this was deliberately done to discourage users from using equipment from other manufacturers. There are more than 50 variants of the original bi-sync protocol. One can assume, that a standard would have prevented at least some of this from happening.

In some cases, protocols gain market dominance without going through a standardization process. Such protocols are referred to as de facto standards. De facto standards are common on emerging markets, niche markets, or markets that are monopolized (or oligopolized). They can hold a market in a very negative grip, especially when used to scare away competition. From a historical perspective, standardization should be seen as a measure to counteract the ill-effects of de facto standards. Positive exceptions exist; a 'de facto standard' operating system like GNU/Linux does not have this negative grip on its market, because the sources are published and maintained in an open way, thus inviting competition. Standardization is therefore not the only solution for open systems interconnection.

Standards organizations
Some of the standards organizations of relevance for communications protocols are the International Organization for Standardization (ISO), the International Telecommunications Union (ITU), the Institute of Electrical and Electronics Engineers (IEEE), and the Internet Engineering Task Force (IETF). The IETF maintains the protocols in use on the Internet. The IEEE controls many software and hardware protocols in the electronics industry for commercial and consumer devices. The ITU is an umbrella organization of telecommunications engineers designing the public switched telephone network (PSTN), as well as many radio communication systems. For marine electronics the NMEA standards are used. The World Wide Web Consortium (W3C) produces protocols and standards for Web technologies.

International standards organizations are supposed to be more impartial than local organizations with a national or commercial self-interest to consider. Standards organizations also do research and development for standards of the future. In practice, the standards organizations mentioned, cooperate closely with each other.

The standardization process
The standardization process starts off with ISO commissioning a sub-committee workgroup. The workgroup issues working drafts and discussion documents to interested parties (including other standards bodies) in order to provoke discussion and comments. This will generate a lot of questions, much discussion and usually some disagreement on what the standard should provide and if it can satisfy all needs (usually not). All conflicting views should be taken into account, often by way of compromise, to progress to a draft proposal of the working group.

The draft proposal is discussed by the member countries' standard bodies and other organizations within each country. Comments and suggestions are collated and national views will be formulated, before the members of ISO vote on the proposal. If rejected, the draft proposal has to consider the objections and counter-proposals to create a new draft proposal for another vote. After a lot of feedback, modification, and compromise the proposal reaches the status of a draft international standard, and ultimately an international standard.

The process normally takes several years to complete. The original paper draft created by the designer will differ substantially from the standard, and will contain some of the following 'features': International standards are reissued periodically to handle the deficiencies and reflect changing views on the subject.
 * Various optional modes of operation, for example to allow for setup of different packet sizes at startup time, because the parties could not reach consensus on the optimum packet size.
 * Parameters that are left undefined or allowed to take on values of a defined set at the discretion of the implementor. This often reflects conflicting views of some of the members.
 * Parameters reserved for future use, reflecting that the members agreed the facility should be provided, but could not reach agreement on how this should be done in the available time.
 * Various inconsistencies and ambiguities will inevitably be found when implementing the standard.

Future of standardization (OSI)
A lesson learned from ARPANET (the predecessor of the Internet) is that standardization of protocols is not enough, because protocols also need a framework to operate. It is therefore important to develop a general purpose, future-proof framework suitable for structured protocols (such as layered protocols) and their standardization. This would prevent protocol standards with overlapping functionality and would allow clear definition of the responsibilities of a protocol at the different levels (layers). This gave rise to the ISO Open Systems Interconnection reference model (RM/OSI), which is used as a framework for the design of standard protocols and services conforming to the various layer specifications.

In the OSI model, communicating systems are assumed to be connected by an underlying physical medium providing a basic (and unspecified) transmission mechanism. The layers above it are numbered (from one to seven); the nth layer is referred to as (n)-layer. Each layer provides service to the layer above it (or at the top to the application process) using the services of the layer immediately below it. The layers communicate with each other by means of an interface, called a service access point. Corresponding layers at each system are called peer entities. To communicate, two peer entities at a given layer use a (n)-protocol, which is implemented by using services of the (n-1)-layer. When systems are not directly connected, intermediate peer entities (called relays) are used. An address uniquely identifies a service access point. The address naming domains need not be restricted to one layer, so it is possible to use just one naming domain for all layers. For each layer there are two types of standards: protocol standards defining how peer entities at a given layer communicate, and service standards defining how a given layer communcates with the layer above it. In the original version of RM/OSI, the layers and their functionality are (from highest to lowest layer): In contrast to the TCP/IP layering scheme, which assumes a connectionless network, RM/OSI assumed a connection oriented network. Connection oriented networks are more suitable for wide area networks and connectionless networks are more suitable for local area networks. Using connections to communicate implies some form of session and (virtual) circuits, hence the (in the TCP/IP model lacking) session layer. The constituent members of ISO were mostly concerned with wide area networks, so development of RM/OSI concentrated on connection oriented networks and connectionless networks were only mentioned in an addendum to RM/OSI. At the time, the IETF had to cope with this and the fact that the Internet needed protocols which simple were not there. As a result the IETF developed its own standardization process based on "rough consensus and running code". The standardization process is described by RFC2026.
 * The application layer may provide the following services to the application processes: identification of the intended communication partners, establishment of the necessary authority to communicate, determination of availability and authentication of the partners, agreement on privacy mechanisms for the communication, agreement on responsibility for error recovery and procedures for ensuring data integrity, synchronization between cooperating application processes, identification of any constraints on syntax (e.g. character sets and data structures), determination of cost and acceptable quality of service, selection of the dialogue discipline, including required logon and logoff procedures.
 * The presentation layer may provide the following services to the application layer: a request for the establishment of a session, data transfer, negotiation of the syntax to be used between the application layers, any necessary syntax transformations, formatting and special purpose transformations (e.g. data compression and data encryption).
 * The session layer may provide the following services to the presentation layer: establishment and release of session connections, normal and expedited data exchange, a quarantine service which allows the sending presentation entity to instruct the receiving session entity not to release data to its presentation entity without permission, interaction management so presentation entities can control whose turn it is to perform certain control functions, resynchronization of a session connection, reporting of unrecoverable exceptions to the presentation entity.
 * The transport layer provides reliable and transparent data transfer in a cost effective way as required by the selected quality of service. It may support the multiplexing of several transport connections on to one network connection or split one transport connection into several network connections.
 * The network layer does the setup, maintenance and release of network paths between transport peer entities. When relays are needed, routing and relay functions are provided by this layer. The quality of service is negotiated between network and transport entities at the time the connection is setup. This layer is also responsible for (network) congestion control.
 * The data link layer does the setup, maintenance and release of data link connections. Errors occurring in the physical layer are detected and may be corrected. Errors are reported to the network layer. The exchange of data link units (including flow control) is defined by this layer.
 * The physical layer describes details like the electrical characteristics of the physical connection, the transmission techniques used, and the setup, maintenance and clearing of physical connections.

Nowadays, the IETF has become a standards organization for the protocols in use on the Internet. RM/OSI has extended its model to include connectionless services and because of this, both TCP and IP could be developed into international standards.

Taxonomies
Classification schemes for protocols usually focus on domain of use and function. As an example of domain of use, connection-oriented protocols and connectionless protocols are used on connection-oriented networks and connectionless networks respectively. For an example of function consider a tunneling protocol, which is used to encapsulate packets in a high-level protocol, so the packets can be passed across a transport system using the high-level protocol.

A layering scheme combines both function and domain of use. The dominant layering schemes are the ones proposed by the IETF and by ISO. Both layering schemes are considered less than perfect, but they do represent current progress in protocol design and development. Despite the fact that the underlying assumptions of the layering schemes are different enough to warrant distinguishing the two, it is a common practice to compare the two by relating common protocols to the layers of the two schemes. For an example of this practice see: List of network protocols.

The layering scheme from the IETF is called Internet layering or TCP/IP layering. The functionality of the layers has been described in the section on software layering and an overview of protocols using this scheme is given in the article on Internet protocols.

The layering scheme from ISO is called the OSI model or ISO layering. The functionality of the layers has been described in the section on the future of standardization and an overview of protocols using this scheme is given in the article on OSI protocols.

Common types of protocols
The Internet Protocol is used in concert with other protocols within the Internet Protocol Suite. Prominent members of which include:
 * Transmission Control Protocol (TCP)
 * User Datagram Protocol (UDP)
 * Internet Control Message Protocol (ICMP)
 * Hypertext Transfer Protocol (HTTP)
 * Post Office Protocol (POP3)
 * File Transfer Protocol (FTP)
 * Internet Message Access Protocol (IMAP)

Other instances of high level interaction protocols are:
 * IIOP
 * RMI
 * DCOM
 * DDE
 * SOAP