← all systems · see it in The System
OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200)
The purpose of this Reference Model of Open Systems Interconnection is to provide a common basis for the coordination of standards development for the purpose of systems interconnection, while allowing existing standards to be placed into perspective within the overall Reference Model. ... The resulting architecture with seven layers, illustrated in Figure 11, obeys the principles in 6.2 a) and b).
Properties published scienceour proposal
Sourced fields, and our own readings.
| born at | 1978 (provisional model, SC16/N34, March 1978) |
| born precision | month |
| boundary | Inside: the abstract architecture (layers, entities, services, service access points, protocols, data units, management and the principles for drawing layers) as written in ISO/IEC 7498 / X.200 and its parts, and as people hold and use it. Outside: the real systems and their internal working ('OSI is concerned only with the interconnection of systems'), the service and protocol standards written to it (which can be implemented; the model 'cannot be implemented'), and the rival models. |
| nested level | Built from seven layers and the elements of layer operation; embedded in ISO's and ITU-T's standards work and in the field's teaching. The highest of three levels of abstraction in the OSI scheme, above service and protocol specifications. |
| scale | technological |
Measures (24) published science
| layers in the reference model | 7 layers |
| principles used to choose the layers | 13 principles (1980 list) |
| drafting time of the model in SC16 | 18 months |
| experts contributing to the model | more than 100 experts |
| characteristic clock: initial development of the reference model | 2.5 years |
| characteristic clock: Draft Proposed Standards from SC16 | around 2 per year (1983) |
| the wait: a whole service data unit before passing | one complete SDU per connection |
| revision interval: IS 7498 to the revised joint text | 1983 to 1 July 1994 dates |
| US federal OSI procurement mandate in force | August 1990 (GOSIP v1 published August 1988); abandoned by NIST in 1994 dates |
| IETF rough consensus (the rival process) | 80 to 90 percent agreement |
| IETF participants protesting the IAB's CLNP proposal, July 1992 | approximately 700 participants |
| layers in the ARPANET Reference Model (rival) | 3 (one mandatory) layers |
| payload share of a TCP ack carried as IP over ATM | 40 of 106 bytes (the source: '37% payload (TCP Ack size)') |
| IP over ATM overhead on a DS3 line | about 31 percent of 44.736 Mbps |
| payload share of a full Ethernet frame after TCP/IPv4 headers | 1460 of 1500 bytes of the Ethernet data field (the IP datagram), not of the frame |
| OSI connection-mode transport classes | 5 (TP0-TP4) classes |
| EvoArch hourglass resemblance H at round 100 (median) | almost 95 percent |
| span over which the protocol hourglass emerged | 30-40 years |
| change in quiz score after analogical role-play | +8 percentage points (n = 10, single group, eight questions) |
| DS3 physical-layer toll: PLCP overhead | 4.032 Mbps of 44.736 |
| DS3 cell-layer toll: per-cell headers | 3.840 Mbps of 44.736 |
| DS3 adaptation-layer toll: SNAP header and padding | 5.900 Mbps of 44.736 |
| first revision interval: IS 7498 to Addendum 1 (connectionless) | spring 1983 to 15 July 1987 dates |
| EvoArch layers for the TCP/IP stack | 6 layers (application protocols at 5, applications at 6) |
Open contentions (6) open contention
Where sources or readings disagree. Both stay on the map.
- C-1373 Do networks really have seven layers?
booked: The map uses the seven-layer OSI model as the field teaches it, and keeps the Internet's four-layer view, which folds two of them into the applications, alongside.
the other reading: the Internet's four layers fold them into applications and TCP (RFC 1122); the layer died but its notation lived: ASN.1 (the presentation layer's encoding rules, src:wp:osi-model) encodes every X.509 certificate the Internet uses ('ASN.1 distinguished encoding rules (DER)', src:rfc5280) - C-1542 Why did TCP/IP win over OSI?
booked: The map lists the sourced reasons (OSI's committee process, TCP/IP being free, running code first) and rules none.
the other reading: see research/CONTENTIONS.md C-1542; OSI's wide waist (two network services, five transport classes, then profiles to choose among them) and conformance without interworking (X.290: 'conformance to a test suite alone cannot guarantee interworking'); the specialist's reading, partly sourced here - C-3829 Should network layers be kept strictly apart, or does real engineering need to reach across them?
booked: The map keeps the competing readings; none is ruled.
the other reading: layered architecture 'optimal up to an order'; unbridled cross-layer design leads to 'spaghetti design' (Kawadia and Kumar 2005); 'Layering Considered Harmful': layering raises complexity and cost (RFC 3439); an 'avalanche' of cross-layer proposals with open opportunities (Srivastava and Motani 2005); 'Cross-layer functions are the norm, in practice' (encyclopedia) - C-3848 When did the US government give up its requirement to buy OSI networking: 1994 or 1995?
booked: The map keeps the competing readings; none is ruled.
the other reading: NIST proposed dropping the OSI requirement in 1994; adopted into FIPS (FIPS 146-2) the following year (encyclopedia, Protocol Wars) - C-3870 Why seven layers, and is seven the right number for a network model?
booked: The map keeps the competing readings; none is ruled.
the other reading: three broad layers (ARPANET Reference Model, RFC 871); four layers (RFC 1122); a fixed five-layer internet model (as RINA's authors describe it); one recursive layer type repeated as needed (RINA); 'many alternative decompositions, leading to a choice of different layering architectures' (Chiang et al. 2007); Cohen and Postel's mockery of the number seven (via Russell 2006); EvoArch parameterizes the TCP/IP stack with six layers (src:akhshabi-dovrolis-2011); one of the model's own authors (Day) later co-authored the recursive rival (RINA) - C-3888 Who designed the seven-layer network model?
booked: The map keeps the competing readings; none is ruled.
the other reading: the seven-layer concept from Charles Bachman's work at Honeywell (encyclopedia, citing J.A.N. Lee); Zimmermann as 'the real driving force' (Davies, quoted in the encyclopedia)
Cross-scale patterns seen here (21) our proposal
Shapes that recur when the same kind of problem is solved at different scales, and how each shows in OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200). Each links to every other system that shows it.
- namespace completeness (the map predicts its gaps) (structural)
The model showed missing standards before they existed: 'if A is to be standardized and B is to be standardized then C must be too'; in 1980 'No standard exists and no proposal has been currently available' for the session layer, and the session protocol was a Draft Proposal by 1983. X.200 states it - nested ladder (structural)
Three levels of abstraction, each a tighter constraint: 'The OSI Reference Model, Services, and Protocols are successively more detailed and therefore more constraining specifications'; anything satisfying a protocol specification satisfies the service specification and the model. Within the stack, - same role, different builds (structural)
The same job, reliable end-to-end transfer between hosts, done by TP4 and by TCP ('Class 4 is closest to TCP'); the same layered division done by SNA, AppleTalk, IPX, UMTS and SS7 in the encyclopedia's comparison table. - same toolkit, independent builds (structural)
The hourglass with an old, conserved waist arises in protocol stacks by competition ('not designed but it emerged through the competition between protocols') and in animal and plant development ('the oldest genes concentrated at the hourglass waist'); the authors trace both to one principle, increas - templating (structural)
One template for every layer: entities, services, SAPs, protocols and PDUs are defined once for an '(N)-layer' and reused at all seven ('the general principles described in clauses 4 and 5 hold for all layers'); a layer can be swapped out whole while keeping its services (principle f). The field's o - The carry (structural)
Overflow passed on: a message too large for a link is split into fragments at the network layer, and transport cuts segments to fit the path's MTU (1460 bytes on Ethernet). The wait: 'Data may be held within a connection until a complete service-data-unit is put into the connection', and in strict l - The Payment to Entropy (structural)
Signal/material form: every hand-off down the stack adds protocol-control information, so each layer pays a header toll that compounds with depth. Numbers: a 40-byte TCP ack took 106 bytes as IP over ATM ('37% payload'), total overhead about 31% of a DS3 line; a full Ethernet frame keeps 1460 of 150 - the whole sets the local reference (structural)
Every OSI standard measures itself against the model ('shall be consistent'; 'identical definitions and terminology'), and each reads true locally, yet local conformance does not close into a working whole: 'this product conforms to the OSI Reference Model' does not imply the ability to interwork wi - trade one property for another (structural)
Layering buys independence and longevity and pays in performance ('the optimization of each layer has to be done separately'); cross-layer design buys performance and pays in architecture ('spaghetti design'). Kinds settle along the dial: the five transport classes, from TP0 ('the fewest features', - Umwelt: each system carves its own slice of one shared medium (structural)
Each layer reads only its own slice of the same data unit: 'Only those capabilities that can be seen from the layer above are services'; the transport layer is like a post office that 'inspects only the outer envelope'. The model's own Umwelt (its channels) is narrow by design: 'OSI is concerned onl - who opens the door (structural)
The service access point is a port opened from above ahead of time: an (N)-SAP is where the (N)-entity serves an (N+1)-entity, bindings are kept in an (N)-directory, and a connection needs the (N-1)-connection already available ('The same consideration applies downwards until an available connection - barrier-gates-complexity (partial)
Layer boundaries let each layer be redesigned alone; the without-the-barrier case is argued in Kawadia and Kumar: 'unbridled cross-layer design can lead to spaghetti design, which can stifle further innovation'. RFC 3439 argues the opposite (more layering, more complexity), so the reading is contest - borrowed-then-held (partial)
Lower layers were borrowed from existing CCITT and ISO standards ('The only work to be done will consist of clearly relating those standards to the OSI Architecture model'); session and presentation were built by SC16 itself. Both phases shown, but for different layers, not one job moving inside. - causal ceiling (partial)
By design no layer can see below the service it is given: the network layer 'makes invisible to transport-entities how underlying resources such as data-link-connections are used'. A limit set by the architecture's structure, not by a layer's means; but it is a designed rule rather than a causal str - conservation law plus closure (partial)
The model leaves the slot open on purpose: it 'does not specify services and protocols' and 'cannot be implemented'; service and protocol specifications must close it before anything can be built. Not a balance law, so partial. - lightning (partial)
July 1992: the IAB's draft to use CLNP leaked ('everything accelerated. Some journalists got the news'), and about 700 IETF participants revolted; the discharge settled who would standardize the Internet's protocols. A social discharge, read from one history. - partner-becomes-part (partial)
ISO's and CCITT's separate documents were merged in 1983 into one text, published identically by both since; the two bodies kept their own identities and memberships, so the fusion is of the text only. - runaway from an unstable balance (partial)
Suites 'coexisted throughout the late 1970s and 1980s'; then 'The market for network protocols had tipped in favor of TCP/IP'. EvoArch gives a mechanism: once one protocol at the waist gains value, its competitors die and the cascade runs. The balance point itself is not measured. - same shape, opposite sign (partial)
One (N)-protocol run both ways: the sender's (N)-entity adds protocol-control information to the SDU, the receiver's peer entity reads and strips it ('successively stripped from each layer's header or footer'). Same protocol, direction reversed; partial because sender and receiver are two entity inv - seed (partial)
The model opened a space of standards to fill layer by layer ('around two a year' in 1983) and practical subsets 'defined from optional functions'; whether that is vast new combinatorics or ordinary planning is unclear. - symmetry breaking (name the cut) (partial)
All open systems have the same seven layers; the cut is named at the transport layer: relaying happens 'within the Network Layer or below', while transport protocols 'operate only between OSI end open systems'. Each system takes the end or relay role per instance of communication.
Valence: what it trades, and with whom (9) published science
Each exchange with a neighbour, read from both sides: who gains, who pays, and what carries it.
- Human (Homo sapiens) via sys:sensory-environment · OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200) gain
- ARPANET (the first carrier network) via sys:written-record · OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200) gain
- Internet via sys:mental-representation · OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200) mixed
- The OSI protocol suite (the rival) via sys:written-record · OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200) mixed
- Internet protocol suite (TCP/IP): the layered datagram protocols every host and router runs, the RFC record that specifies them, and the community that keeps them via sys:mental-representation · OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200) mixed
- Information security via sys:written-record · OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200) gain
- CCITT (International Telegraph and Telephone Consultative Committee) via sys:written-record · OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200) gain
- Curriculum via sys:human-society · OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200) gain
- ISO (ISO/IEC JTC 1; earlier TC97/SC16) via sys:written-record · OSI model (the Basic Reference Model for Open Systems Interconnection, ISO/IEC 7498-1 | ITU-T X.200) gain
Partners and parts published science
Built from / builds (10)
- presentation layer
- session layer
- Reference models for describing networks: seven layers against three, four, five and recursive
- Physical layer
- Data link layer
- Network layer
- Transport layer
- Application layer
- Service access point (SAP)
- Protocol data unit (PDU)
Sits inside / holds (3)
- CCITT (International Telegraph and Telephone Consultative Committee)
- Curriculum
- ISO (ISO/IEC JTC 1; earlier TC97/SC16)
expressed by (18)
- Generation/Synthesis
- Recombination/Composition
- Scaffolding/Enablement (Meta-Tooling)
- Templating / Standardization
- Quality Improvement/Refinement
- Interoperability/Transduction (Bridging)
- Encapsulation & Modularity (Abstraction)
- Symbiosis/Co-Creation (Collaborative Specialization)
- Evolvability/Plasticity Engineering (Design for Future Variation)
- Distributed Autonomy/Decentralized Execution (Subsidiarity)
- Internal Coordination/Regulation
- Mapping/Modeling
- Navigation/Pathfinding
- Proactive Defense
- Self-Monitoring
- Validation of Identity/Intention
- Quality Control & Error Correction
- Positional Identity (Body Map)
floor filled by (20)
- Electromagnetic field
- Human (Homo sapiens)
- Internet protocol suite (TCP/IP): the layered datagram protocols every host and router runs, the RFC record that specifies them, and the community that keeps them
- x25
- The medium or field that carries it
- Information security
- ConnectionLess Network Protocol (OSI)
- Curriculum
- Osi transport protocol
- IEEE 802 reference model (MAC and LLC sublayers)
- X500
- ISO (ISO/IEC JTC 1; earlier TC97/SC16)
- asn1
- hdlc
- is is
- iso ballot
- iso tc97 sc16
- layer 8
- osi session protocol
- x400
represents (2)
Kinds (10)
- Internet protocol suite (TCP/IP): the layered datagram protocols every host and router runs, the RFC record that specifies them, and the community that keeps them
- ARPANET Reference Model
- Five-layer internet model
- Recursive InterNetwork Architecture (RINA)
- OSI Reference Model, first edition (ISO 7498:1984 / X.200 1984)
- OSI security architecture (ISO 7498-2 / X.800)
- OSI naming and addressing (ISO 7498-3)
- OSI management framework (ISO 7498-4)
- OSI conformance testing methodology and framework (X.290-X.296 | ISO/IEC 9646)
- OSI service conventions (X.210 | ISO/IEC 10731)
Notes published science
- The standard admits its own number is a choice: 'It may be difficult to prove that any particular layering selected is the best possible solution' (X.200). Its rivals mocked the seven, wondering whether Zoroastrianism, celestial beings or the seven deadly sins 'shaped the choice of Seven' (Cohen and Postel, via src:russell-2006).
- Competition alone grows an hourglass: in a simulation where Internet protocols live or die by the value of what is built on them, the stack narrows to a waist of old, hard-to-replace protocols (mostly IPv4, TCP and UDP). The same team then built a model of embryo development that predicts the same hourglass, with the oldest genes at the waist, and found fly and cress gene data consistent with it and with what Domazet-Loso and Tautz had measured (src:akhshabi-dovrolis-2011, src:akhshabi-2014-hourglass).
- Every layer charges a toll: a 40-byte TCP acknowledgement needed 106 bytes when carried as IP over ATM, and about 31% of a DS3 line went to overhead (src:rfc3439).
- One of the OSI model's own authors later argued against its fixed stack: John Day co-wrote the 1983 reference-model paper, and is an author on the Recursive InterNetwork Architecture, where one kind of layer repeats instead of seven different ones (src:day-zimmermann-1983, src:vrijders-2020-rina).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
mental representation: the seven-layer model as engineers, teachers and learners hold and use it
- L1: Networks as people meet them: invisible, layered processes
- L2: Learning the seven layers: lecture, analogy (the postal letter), role-play
- L3: A shared vocabulary: protocols of other suites named by OSI layer (TCP and UDP 'layer 4')
- L4: Working by layer: troubleshooting, and cross-layer designs that reach across layers
- L5: Security architecture by layer
- L6: Extensions above the seven: the user (layer 8), money and politics (layers 8-9)
- L7: The reference for teaching and documentation, outliving its protocols
the OSI reference model as its users draw it: seven layers from the physical medium to the application (reader's view)
- L1: Physical layer: the mechanical, electrical, functional and procedural means for bit transmission over physical media (the media themselves are sometimes called a (0) layer)
- L2: Data link layer: data-link connections over physical connections; detects and possibly corrects physical-layer errors
- L3: Network layer: routing and relaying across subnetworks in tandem or parallel; may split messages too large for a link
- L4: Transport layer: transparent, end-to-end transfer between end systems, optimizing the network service at minimum cost
- L5: Session layer: organizing and synchronizing dialogue between presentation entities
- L6: Presentation layer: common representation of data, syntax independence for applications
- L7: Application layer: application entities, the sole access of application processes (people, programs, machines) to the OSI environment
the boardroom (shared floors)
- L5: ISO TC97/SC16 (Open Systems Interconnection) and its working group on architecture, with the CCITT rapporteur's group: where contributions, comments and liaison texts converge
- L6: The reference model text: ISO 7498 / X.200 and its parts (security, naming and addressing, management), the definitions every OSI standard must use
- L7: The field that holds and uses the model: standards developers, governments, network engineers, teachers and security architects; the shared reference for teaching and documentation
way in: CCITT's work (public data networks, X.25, the rapporteur's group on a layered model)
- L1: Public data networks adopting X.25 in the late 1970s
- L2: Existing CCITT physical and network-level standards (X.21, V.24, V.35; X.25 level 3)
- L3: The CCITT Rapporteur's Group on a layered model for public data network services recognises the model
- L4: Collaborative drafting with ISO/IEC to one identical text
way in: Critiques and rival theories of layering
- L1: Implementers find strict layering imperfect and break it creatively
- L2: Written critiques: 'That violates Layering!' (1982); 'Layering Considered Harmful' (2002)
- L3: Named principles about where functions belong: the end-to-end argument
- L4: Cross-layer design and its cautions
- L5: Rival architectures: recursive layers (RINA) against a fixed five-layer internet model
- L6: Formal theories of layering: optimization decomposition; evolutionary models (EvoArch)
way in: Data units arriving at an open system (the flow the model describes)
- L1: Bits over physical connections
- L2: Frames checked: errors detected and possibly corrected
- L3: Network: routed and relayed across subnetworks
- L4: Transport: end to end between end systems
- L5: Session: dialogue organized and synchronized
- L6: Presentation: common representation of data
- L7: Application entity: the application process's only access
way in: Existing networks and network architectures (ARPANET, CYCLADES, NPL, EIN; vendors' SNA and DECnet; X.25)
- L1: Running packet networks and vendors' installed networks
- L2: Each maker's own 'network architecture' of conventions
- L3: Published proposals entered as SC16 documents (INWG 96.1 transport as SC16/N24; IFIP virtual terminal as SC16/N23)
- L4: Principle e): boundaries where past experience has shown success
way in: Learners and practitioners (classroom, troubleshooting, the 'layer 8' talk)
- L1: Networks whose processes are invisible to people
- L2: A lecture and a role-play seen and heard by students
- L3: The postal-service analogy maps the seven layers to sending a letter
- L4: Understanding measured by quiz
- L5: Layer numbers used in practitioners' talk and troubleshooting
- L6: The model extended in jokes and politics: layer 8 user, layer 9 politics
way in: National member bodies and their experts (contributions, comments, ballots)
- L1: More than 100 experts from many countries and international organizations
- L2: Written contributions merged into the text
- L3: National positions: delegates acting for national firms
- L4: Ballots and comment rounds on DP and DIS
way in: New transmission technology and modes (fibre, LANs, datagram and connectionless service, ISDN)
- L1: New physical media such as fibre optics
- L2: Local area networks on broadcast media
- L3: Connectionless (datagram) operation
- L4: Outstanding architectural issues listed for further work
way in: Security needs and threats on the communication path
- L1: The communications path between end systems
- L2: Commercial users' need to transfer data securely
- L3: Security services (peer entity authentication, confidentiality, integrity, non-repudiation, traffic flow confidentiality)
- L4: Security layering principles decide placement
way in: The rival's running practice: TCP/IP networks, their running code and the market
- L1: TCP/IP in widespread use on multi-vendor networks
- L2: Running code: multiple interoperable implementations before standardization
- L3: A descriptive rival reference model (the ARPANET Reference Model, three layers)
- L4: Comparisons: the US National Research Council study of TCP against TP-4 (asked 1983, reported 1985)
- L5: Procurement policy met by administrators who kept TCP/IP
- L6: The market tips (the World Wide Web)
way out: Coordinating standards: layer service definitions and protocols written to the model
- L1: Published documents (ITU-T X-series; ISO/IEC standards)
- L4: Layer standards: e.g. the connection-mode transport protocol in five classes, TP0 to TP4 (X.224)
- L5: TC97 adopts SC16's recommendations to start protocol projects (end of 1979)
- L6: Compliance and consistency statements required of other standards
way out: Designs placed by layer (and cross-layer designs)
- L1: Implemented on links and hosts
- L4: Protocols placed in a layer; cross-layer schemes reach across
- L7: Designers reason by layer
way out: Government profiles and procurement mandates (GOSIP)
- L1: Purchases by federal agencies
- L5: GOSIP version 1 published August 1988; required from August 1990
- L7: Governments endorse OSI
way out: Layer talk: naming problems and devices by layer number
- L1: Spoken among people
- L4: Troubleshooting placed by layer (a 'layer 8 issue')
- L7: Layer numbers as the shared vocabulary
way out: Security placed by layer (X.800 / ISO 7498-2)
- L1: Protected traffic on the medium
- L3: Encryption at the network layer (IPsec); MACsec at the data link layer (layer 2) is booked beside it
- L6: Security services allocated to layers
way out: Teaching the seven layers
- L1: Students enact the layers as sender and receiver groups
- L3: A lecture and the postal analogy
- L7: The model in curricula
way out: The OSI protocol suite on the wires (X.400, X.500, IS-IS lasting)
- L1: Signals on the medium
- L3: IS-IS routing at the network layer
- L7: Application protocols X.400 (messaging) and X.500 (directory)
written record: the reference model as a standard (ISO 7498 / ITU-T X.200 and its parts)
- L1: Contributions and working documents from experts and member bodies (SC16 N-documents)
- L2: SC16 and its architecture working group draft the model: 18 months, seven layers, transmitted to TC97 in July 1979
- L3: Ballots: Draft Proposal (late 1980), Draft International Standard (spring 1982) after two rounds of comments
- L4: International Standard ISO 7498 (spring 1983); merged with CCITT's text and published 1984 as ISO 7498 and X.200; revised joint text approved 1 July 1994
- L5: The family of parts: basic model, security architecture, naming and addressing, management
- L6: Three levels of abstraction and compliance: architecture, then service and protocol specifications that must be consistent with it
- L7: Purpose: a common basis for coordinating standards development, and a frame within which teams work independently