Cell Relay¶

Switch statistically multiplexed fixed-length cells through a broadband network, trading bounded serialization and regular cell forwarding for per-cell overhead plus segmentation-and-reassembly state.

Version: v2 · 2026-09-06 · History
Domain-specific #: 1448
Origin domain: computer networking
Subdomain: broadband cell switching
Aliases: Cell switching, Fixed-length cell relay

Core Idea¶

Cell relay is a network-transfer discipline in which traffic is divided into small fixed-length cells, statistically interleaved on shared links, and forwarded cell by cell. Its best-known realization is Asynchronous Transfer Mode (ATM). ITU-T Recommendation I.150 defines ATM as a packet-oriented, asynchronous time-division method using fixed-size cells; the five-octet ATM header identifies the virtual channel and accompanies a 48-octet information field.^[1] I.361 specifies that ATM cell structure and header coding.^[2]

The abstraction is not “ATM” by another name. ATM is the dominant connection-oriented member: signaling or management establishes virtual-path/virtual-channel state, and the VPI/VCI label lets switches forward each cell along that association. But historically the cell-relay family also included Switched Multi-megabit Data Service (SMDS), which RFC 1209 describes as a connectionless public packet-switched service.^[3] The family invariant is therefore fixed-cell relay with statistical multiplexing, not universal connection orientation. A connection or forwarding association is load-bearing in ATM; a connectionless cell service can instead carry sufficient per-datagram addressing in its service machinery.

Because most user protocol data units are not 48 bytes, an adaptation function segments them at ingress and reassembles them at egress. In ATM Adaptation Layer 5 (AAL5), a higher-layer protocol data unit is padded, given an eight-octet trailer, split into 48-octet payloads, and carried in successive ATM cells. RFC 2684 states the 53-octet cell composition and the AAL5 segmentation/reassembly role for variable-length bridged or routed protocol data units.^[4] Recommendation I.363.5 defines the AAL5 convergence and segmentation-and-reassembly functions.^[5]

Fixed length regularizes switch-fabric work and bounds the time for which one already-transmitting unit can block another at a given link rate. It does not make delay or jitter constant. Cells still wait in queues; traffic contracts, scheduling, policing, congestion, and loss determine observed cell transfer delay and cell delay variation. I.356 defines those ATM performance quantities and QoS classes, while I.371 specifies traffic and congestion-control concepts including traffic contracts and service categories.^[6]^[7]

The design pays for regularity with overhead and state. Every 48 payload octets carry a five-octet cell header; partially filled final cells add padding; ingress and egress must preserve PDU boundaries; and loss of one cell can make an entire reassembled AAL5 PDU unusable. ATM provides header error control but does not generally perform information-field error control at the ATM layer, so cell relay must not be mistaken for end-to-end reliability.^[1]

Structural Signature¶

Sig role-phrases:

the offered traffic units — streams, packets, frames, or service data units that must share a broadband transport
the fixed cell format — one invariant transfer-unit length with a small forwarding/control header and a bounded payload field
the cellization rule — the adaptation procedure that maps variable-length or continuous-source traffic into payload-sized pieces, adds boundary/control information, and pads as required
the traffic association or addressing context — connection state and short path/channel labels in ATM, or the connectionless service context needed by another family member
the statistical multiplexer — the scheduler that interleaves cells from active sources rather than reserving every source a permanently recurring slot
the cell-by-cell forwarding path — switches that examine the cell header, select an output, and translate or preserve the local label as the protocol requires
the queue and traffic-control regime — buffers, service categories, traffic descriptors, conformance checks, policing, shaping, and congestion behavior
the reassembly contract — the egress procedure that recognizes the end of the higher-layer unit, checks adaptation-layer integrity, discards invalid reconstructions, and delivers a PDU or stream
the performance-and-overhead ledger — cell transfer delay, delay variation, loss, cellization delay, padding, header tax, reassembly failure, and useful-payload efficiency

The locked recognition test requires the fixed cell format, statistical sharing, cell-by-cell forwarding, and an explicit mapping between offered traffic and cells. A technology is Cell Relay when fixed-length cells are the network's operative relay units, not merely an internal memory size or an incidental link encoding. For an ATM instance, virtual path/channel labels and connection state must also be present; for a claimed non-ATM member, the addressing and service contract must show how cell units are relayed without importing ATM's connection model.

Diagnostics follow those roles. Uneven latency with compliant traffic points first to queues and scheduling, not to a violation of fixed cell size. Excessive bandwidth tax points to cell headers, padding, or a small-PDU workload. Whole-PDU loss after one missing cell points to reassembly dependence. Misdelivery or reassembly mixing points to label/context handling. A “Cell Relay” design that forwards variable-length frames end to end has crossed into frame or packet relay even if hardware internally stores fixed-size chunks.

Interventions are role-specific: measure offered PDU-size distribution before choosing encapsulation; calculate cell and padding overhead; select the adaptation layer and multiplexing mode; provision traffic contracts and queues; shape bursty sources; verify label/context isolation; test cell-loss and reassembly behavior; monitor transfer delay, delay variation, and loss separately; and prefer a PDU-oriented transport when preserving cells across an interworking boundary adds cost without preserving a needed cell-level service.

What It Is Not¶

Not ATM alone. ATM is the principal connection-oriented realization, but the broader historical category can include connectionless fixed-cell service such as SMDS.
Not generic packet switching. Packets may be variable in length and carry end-to-end addresses; Cell Relay makes the fixed cell the operative network relay unit.
Not Frame Relay. Frame Relay switches variable-length frames and has different overhead, fragmentation, and delay behavior.
Not synchronous time-division multiplexing. Statistical multiplexing sends cells for active sources as demand occurs; it does not permanently reserve every source a slot in every cycle.
Not circuit switching. An ATM virtual circuit is a forwarding association, but its cells contend statistically for link and buffer resources rather than occupying a continuously reserved physical circuit by definition.
Not mere chunking. Segmentation is one role. The full identity also includes a cell header, shared-link multiplexing, cell forwarding, queueing, and egress reassembly or delivery.
Not guaranteed low latency or jitter. Fixed size bounds serialization granularity; queues, scheduling, traffic conformance, and congestion determine actual delay distributions.
Not end-to-end reliability. Header checks, an AAL5 CRC, and discard behavior can detect corruption or incomplete reassembly; they do not necessarily recover lost user data.
Not a QoS class. Cell Relay is the transfer architecture within which multiple traffic contracts and service categories can be offered.
Not “any network using fixed internal blocks.” Private implementation cells do not establish Cell Relay unless the forwarding and service contract exposes cells as the relay units.

Scope of Application¶

Cell Relay belongs principally to broadband telecommunications and wide-area network architecture.

ATM public and private networks. Virtual paths and channels carry voice, video, circuit emulation, and data over common cell-switching infrastructure.
Packet and frame carriage over AAL5. Routed protocol packets and bridged LAN frames are encapsulated, segmented into ATM cells, and reconstructed at the destination; RFC 2684 specifies common encapsulations.^[4]
Constant- and variable-rate services. ATM traffic categories and contracts distinguish service requirements and source behavior rather than assuming all multiplexed traffic has the same delay or loss needs.^[7]
Cellular interworking and pseudowires. Legacy ATM cell streams can be transported over packet networks when preserving cell-level behavior is necessary. RFC 4717 specifies ATM pseudowire encapsulations and distinguishes cell-relay modes from more efficient AAL5 PDU modes.^[8]
Historical connectionless broadband service. SMDS used connectionless datagram service over fixed-size cell-like transfer machinery, establishing the reason the family boundary cannot be reduced to ATM connection state.^[3]
Performance engineering. Cell transfer delay, cell delay variation, cell loss, cell misinsertion, and severely errored cell blocks provide a cell-level measurement vocabulary for service objectives.^[6]

The abstraction's literal scope ends when variable-length frames or packets remain the network relay units, when fixed chunks are invisible implementation details, or when no statistical multiplexing of cell traffic occurs. Modern Ethernet, IP/MPLS, and optical systems may reuse lessons about quantized forwarding and traffic engineering without thereby becoming Cell Relay networks.

Clarity¶

Three lengths must remain distinct: user PDU length, cell information-field length, and complete cell length. For ATM these are variable, 48 octets, and 53 octets respectively. Quoting “48-byte cells” silently drops the five-octet header; quoting “53 bytes of payload” is simply wrong.^[2]

Next distinguish a virtual circuit from a physical circuit. An ATM VCC or VPC supplies connection-oriented forwarding context. Multiple such associations are statistically multiplexed over the same physical link and may compete for queues. The word “circuit” therefore describes logical connection state, not exclusive continuous ownership of the line.

Then distinguish cell transfer performance from end-to-end application performance. I.356's delay and loss measures apply at ATM boundaries. Application response time can additionally include cellization, reassembly, higher-layer retransmission, processing, and endpoint queues.^[6] A small cell reduces serialization granularity but does not by itself satisfy an end-to-end deadline.

Finally, state the adaptation layer and encapsulation. AAL5 identifies the end of the PDU and supplies length and CRC information in its trailer. LLC encapsulation and VC multiplexing attach different higher-layer identification overheads under RFC 2684.^[4] Overhead calculations and demultiplexing claims are uninterpretable without that choice.

Manages Complexity¶

Cell Relay replaces a mixed population of long data frames and timing-sensitive short messages with one scheduling quantum. A switch fabric can process one size, buffers can count cells, and a high-priority cell need wait at most for the current cell's serialization at a non-preemptive output rather than for an arbitrarily long frame. The same output link can carry many associations because inactive sources consume no data-cell slots.

That regularity moves complexity rather than eliminating it. Variable-length data must be segmented and its boundary recovered. A short message can occupy a partly empty final cell. A burst creates many cells and buffer demand. Since cells from different PDUs interleave, each must preserve sufficient forwarding context. Since a lost cell damages a larger reconstruction, loss accounting must connect cell-level events to PDU-level consequences.

Traffic contracts make the moved complexity governable. Peak and sustainable cell rates, burst tolerance, conformance checking, service category, queue scheduling, and admission decisions collectively determine whether statistical sharing meets a service objective.^[7] The architecture thus compresses switching into a regular primitive while expanding the control plane, adaptation layer, and performance ledger.

The key complexity-management question is not “Are cells simpler than packets?” It is “Does a uniform relay quantum reduce forwarding and multiplexing complexity enough to justify cell headers, padding, connection/context state, and reassembly?” Workloads with many long packets often answer differently from workloads in which fine interleaving across mixed traffic classes is valuable.

Abstract Reasoning¶

Overhead accounting. If a higher-layer encapsulation contributes H octets, the AAL5 trailer contributes eight, and the user PDU contributes B, the cell count is ceil((B+H+8)/48). Raw ATM bytes are 53 times that count. Separate cell-header, encapsulation, trailer, and padding contributions rather than reporting one unexplained efficiency percentage. RFC 2381 gives the corresponding AAL5 cell-count arithmetic for IP over ATM.^[9]

Serialization reasoning. A complete ATM cell occupies 53*8/R seconds on a link of bit rate R, before physical-layer framing effects. This bounds one-cell serialization granularity; it is not a bound on queueing or end-to-end delay.

Loss amplification. Map a cell-loss process to the probability that at least one cell in an N-cell PDU is lost. Under an explicitly independent approximation with cell-loss probability p, PDU survival is (1-p)^N; real loss may be bursty, so measurement should replace the independence assumption where possible.

Traffic-contract audit. Compare the source's peak, sustainable rate, and burst behavior with the contract and queue allocation. Separate nonconforming discard from congestion loss and separate policing from scheduling.

Boundary test. Ask which unit the network switch parses and forwards. If the operative unit is a 53-octet ATM cell, Cell Relay is literal. If a variable-length Ethernet frame is forwarded while an ASIC privately divides memory into fixed slots, “cell relay” is only an internal implementation analogy.

Interworking decision. Preserve cells when cell-level timing, OAM, or service semantics must survive a packet-network crossing. Otherwise terminate AAL5 and carry the reconstructed PDU; RFC 4717 notes the efficiency advantage of PDU mode for appropriate traffic.^[8]

Knowledge Transfer¶

Within networking, the role map transfers across ATM access, backbones, circuit emulation, packet carriage, and pseudowires. In each case identify the offered unit, fixed cell, cellization rule, multiplexing and forwarding context, queues, reassembly contract, and performance ledger. What changes is the adaptation layer, service category, label scope, and interworking boundary.

The overhead method transfers directly to other quantized transports: divide a variable object into fixed payload quanta, add per-quantum metadata, account for final-fragment padding, and measure the cost of damage to one fragment. But that transfer yields a fragmentation or chunking analysis, not a new Cell Relay instance unless fixed units are actually switched as network relay units.

The latency lesson also transfers: smaller non-preemptive service quanta bound serialization blocking while queue discipline governs waiting. Packet schedulers and storage I/O can exhibit the same mathematical trade-off, yet they do not inherit ATM vocabulary or Cell Relay identity automatically.

The design-review workflow is portable: state the operative forwarding unit; distinguish reserved capacity from statistical sharing; identify where segmentation and reassembly occur; calculate overhead under the real size distribution; and trace a unit-loss event to the larger service object. The specialist identity remains network-bound because “cell,” VPI/VCI, AAL, VCC/VPC, and ATM traffic contracts refer to an engineered protocol stack.

Examples¶

Canonical¶

A 1,500-octet IP packet over AAL5 with LLC/SNAP encapsulation. Take a 1,500-octet IP packet, eight octets of LLC/SNAP identification, and the eight-octet AAL5 trailer. Before padding, the AAL5 payload stream is 1500+8+8=1516 octets. It requires ceil(1516/48)=32 cells. Their information fields provide 1,536 octets, so final padding is 20 octets. The link carries 32*53=1696 ATM octets. Relative to the 1,500 user octets, the 196-octet difference consists of 160 cell-header octets, eight LLC/SNAP octets, eight AAL5 trailer octets, and 20 padding octets. Useful-payload efficiency at the ATM-cell boundary is about 88.44 percent. RFC 2684 supplies the cell and AAL5 encapsulation facts; RFC 2381 supplies the cell-count form.^[4]^[9]

Mapped back: the IP packet is the offered traffic unit; ATM's 5+48 format is the fixed cell format; AAL5 plus LLC/SNAP is the cellization rule; the VCC and VPI/VCI labels provide the traffic association; 32 separately forwarded cells traverse the statistical multiplexer and the cell-by-cell forwarding path; the end marker, length, and CRC govern the reassembly contract; and the 196-octet decomposition is the performance-and-overhead ledger.

Applied / In Practice¶

Mixed delay-sensitive and bursty traffic on an ATM trunk. Suppose a provider carries a paced circuit-emulation stream and intermittent AAL5 data over one ATM output. Both become 53-octet cells. The stream receives an appropriate real-time service contract and queue treatment; the data association uses a burst-tolerant contract. Cells are statistically interleaved rather than assigning all capacity permanently to one source. The fixed quantum prevents one long data frame from occupying the transmitter intact ahead of a stream cell, but a data burst can still fill queues, violate its contract, or cause loss. The operator therefore checks traffic conformance under I.371 and measures delay, delay variation, and loss using I.356 concepts rather than treating fixed cell size as a QoS guarantee.^[7]^[6]

Mapped back: the two services supply the offered traffic units; their AAL mappings perform the cellization rule; their virtual associations carry distinct labels; the shared output implements the statistical multiplexer; traffic descriptors, policing, and queues form the queue and traffic-control regime; receiver adaptation restores each service; and class-specific delay, variation, loss, and overhead complete the performance-and-overhead ledger.

Structural Tensions¶

T1: Fine serialization granularity vs. per-cell overhead. Small fixed cells limit the time one cell occupies an output, but the five-octet header recurs every 48 payload octets. Diagnostic: plot useful-payload efficiency by offered PDU size and compare it with the maximum tolerable serialization blocking at the link rate.
T2: Regular switching vs. adaptation complexity. One cell size simplifies relay machinery while pushing variable-length boundary, padding, CRC, and reassembly work to adaptation endpoints. Diagnostic: separate switch processing cost from ingress/egress SAR state, memory, and failure rates.
T3: Statistical utilization vs. queue uncertainty. Sharing idle capacity improves utilization, but simultaneous bursts create waiting and loss. Diagnostic: compare source traffic descriptors and correlated bursts with queue allocation and measured cell delay variation.
T4: Short labels vs. connection/context state. Compact local forwarding labels reduce per-cell address cost, while ATM requires association setup and state consistency. Diagnostic: audit label scope, translation tables, setup failure, stale state, and restoration behavior rather than assuming a short header is stateless.
T5: Cell-level isolation vs. PDU-level loss amplification. A fault may affect one small cell, yet incomplete AAL5 reassembly can discard the whole higher-layer PDU. Diagnostic: report both cell loss and resulting PDU discard by PDU cell count.
T6: Family breadth vs. ATM dominance. Treating every Cell Relay system as ATM gives a crisp connection-oriented story but excludes connectionless SMDS; broadening the family too far admits any fixed internal block. Diagnostic: require fixed units to be the network relay units and state the actual addressing or connection contract.
T7: Preservation vs. termination at interworking boundaries. Cell-relay pseudowires preserve legacy cell semantics but may carry more overhead than reconstructing and transporting AAL5 PDUs. Diagnostic: list the cell-level timing, OAM, or service properties that would be lost by termination; if none is required, compare PDU-mode efficiency.
T8: Autonomy vs. reduction. Cell Relay is autonomously useful because fixed-cell format, cellization, statistical interleaving, forwarding context, queue control, and reassembly form one repeatable network architecture. Reduction to multiplexing preserves shared-channel allocation but loses the cell-size contract, SAR boundary, header tax, label semantics, and cell-to-PDU failure mapping. Diagnostic: if those cell-specific roles do no explanatory work, classify the case under the broader parent rather than retaining Cell Relay.

Structural–Framed Character¶

Cell Relay is mixed-structural, with the structural side dominant. Its mechanism can be recognized objectively: fixed-size relay units, statistical interleaving, cell-by-cell forwarding, segmentation/reassembly, and measurable delay/loss/overhead consequences. The architectural trade-offs follow from sizes, rates, queues, and failure propagation.

Evaluative weight: low. A configuration either uses fixed cells as relay units or does not; “good QoS” is an evaluated outcome, not part of identity.

Human-practice-bound: moderate. Literal instances are engineered telecommunications systems with defined endpoints, service contracts, and operational responsibilities; the abstraction is not a naturally recurring phenomenon independent of network design.

Institutional origin: moderate. ATM's 53-octet format, AALs, headers, performance definitions, and traffic categories exist through standards and interoperable implementation. Standards choose the concrete contract, although the resulting overhead and queueing effects are not matters of institutional opinion.

Vocabulary travels: partly. “Cell,” “segmentation and reassembly,” “VCC,” “VPI/VCI,” and “AAL5” remain specialist. Fixed-quantum multiplexing and overhead accounting travel as mathematics, but applying only those roles elsewhere does not make the target a Cell Relay network.

Import vs. recognize: chiefly recognize within networking, import outside it. Network engineers can identify the mechanism from forwarding units and protocol state. Calling a factory, bureaucracy, or biological system “cell relay” would import a telecommunications frame rather than discover the literal protocol identity.

The portable skeleton is: quantize offered objects into one fixed service unit, statistically share capacity at that unit, preserve enough context to forward and reconstruct, and account for recurring metadata plus fragment-loss amplification. Its character: an engineered but structurally analyzable fixed-unit multiplexing architecture whose specialist protocol contract is indispensable to literal identity.

Structural Core vs. Domain Accent¶

Skeletal structural core. Fixed-length quanta, statistical multiplexing, per-quantum forwarding context, queues, segmentation/reassembly, and an overhead/failure ledger form the reusable mechanism. Rate, size, and probability reasoning explain its principal trade-offs.

Domain-bound accent. ATM/SMDS family boundaries, 53-octet ATM cells, 48-octet information fields, VPI/VCI labels, VCC/VPC state, AAL5 trailers, traffic contracts, service categories, cell transfer delay, and cell delay variation are telecommunications semantics. Removing them prevents one from deciding whether a real protocol is Cell Relay, calculating its actual cost, or diagnosing its failures.

Why it is not a prime. Generalizing to “fixed-size units share a resource” collapses the candidate into existing primes such as Multiplexing, Chunking, Queueing, and Latency. What remains novel is precisely the network-specific composition in which cells are forwarded, labeled, queued, and reassembled. That specialist composition recurs across services and interworking settings but does not literally recur across unrelated substrates.

prime:multiplexing — instantiates. Cell Relay is a strict specialist form of statistical time sharing: cells from many traffic associations share one channel. This is the proposed minimal direct DAC parent.
prime:queueing — related mechanism. Output cells wait according to arrival processes, service rates, priorities, and buffer policies. Queueing explains delay variation and loss but does not supply the fixed-cell or adaptation identity.
prime:latency — related performance dimension. Cellization, serialization, waiting, propagation, and reassembly contribute delay. Latency is measured, not the transfer method.
prime:chunking — related internal role. Segmentation maps larger PDUs to cell payloads. Chunking alone lacks cell switching, statistical multiplexing, and a forwarding context.
prime:network — related substrate. A Cell Relay system is a network architecture; Network is too broad to serve as an informative additional direct parent.

The proposed graph therefore uses only prime:multiplexing as a direct parent. The remaining primes are explanatory neighbors, avoiding redundant multi-parent inheritance.

Relationships to Other Abstractions¶

Current abstraction Cell Relay Domain-specific

Parents (1) — more general patterns this builds on

Cell Relay is a kind of Multiplexing Prime

prime:multiplexing — instantiates. Cell Relay is a strict specialist form of statistical time sharing: cells from many traffic associations share one channel.

Hierarchy paths (2) — routes to 2 parentless roots

Cell Relay → Multiplexing → Scarcity → Constraint

Show alternative path (1)

Neighborhood in Abstraction Space¶

Cell Relay sits in a sparse region of the domain-specific corpus (86^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (1565 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Asynchronous Transfer Mode (ATM). ATM is the dominant connection-oriented Cell Relay technology, not the entire family. Tell: ask whether the claim is about the general fixed-cell relay discipline or specifically the 53-octet ATM protocol and its VPI/VCI connection state.
SMDS. SMDS is a historical connectionless service that demonstrates a family boundary; it is not an alias for all Cell Relay. Tell: its service transfers independent datagrams without prior connection setup.^[3]
Frame Relay. Both are wide-area relay technologies. Tell: Frame Relay's operative unit is a variable-length frame; Cell Relay's is fixed-length.
Packet switching. Cell Relay is packet-oriented in a broad sense. Tell: generic packet switching does not require one fixed relay-unit length or adaptation-layer segmentation into cells.
Circuit switching. ATM uses virtual circuits. Tell: Cell Relay cells are statistically multiplexed and buffered; a physical circuit normally reserves a continuous resource path.
Time-division multiplexing. Both divide link use in time. Tell: synchronous TDM assigns recurring slots; Cell Relay schedules labeled cells as sources have traffic.
Cellular network relay. “Cell” can mean a radio coverage area or mobile base-station region. Tell: Cell Relay's cell is a fixed-length data transfer unit, not a geographic radio cell.
Switch-fabric cell. Routers may internally segment packets into proprietary fixed blocks. Tell: an internal block is not the network service's forwarded relay unit and need not have Cell Relay adaptation or interoperability semantics.

References¶

[1] International Telecommunication Union, ITU-T Recommendation I.150, B-ISDN Asynchronous Transfer Mode Functional Characteristics (1999), official recommendation record. registry ↩a ↩b

[2] International Telecommunication Union, ITU-T Recommendation I.361, B-ISDN ATM Layer Specification (1999), official recommendation record. registry ↩a ↩b

[3] Piscitello and Lawrence, RFC 1209, The Transmission of IP Datagrams over the SMDS Service (1991), RFC Editor. registry ↩a ↩b ↩c

[4] D. Grossman and J. Heinanen, RFC 2684, Multiprotocol Encapsulation over ATM Adaptation Layer 5 (1999), RFC Editor. registry ↩a ↩b ↩c ↩d

[5] International Telecommunication Union, ITU-T Recommendation I.363.5, B-ISDN ATM Adaptation Layer Specification: Type 5 AAL (1996), official recommendation record. registry ↩

[6] International Telecommunication Union, ITU-T Recommendation I.356, B-ISDN ATM Layer Cell Transfer Performance (2000), official recommendation summary. registry ↩a ↩b ↩c ↩d

[7] International Telecommunication Union, ITU-T Recommendation I.371, Traffic Control and Congestion Control in B-ISDN (2004), official recommendation record. registry ↩a ↩b ↩c ↩d

[8] L. Martini et al., RFC 4717, Encapsulation Methods for Transport of Asynchronous Transfer Mode (ATM) over MPLS Networks (2006), RFC Editor. registry ↩a ↩b

[9] M. Laubach, RFC 2381, Interoperation of Controlled-Load Service and Guaranteed Service with ATM (1998), RFC Editor. registry ↩a ↩b