Skip to content

G/G/1 Queue

A single-server queue model with general arrival and service behavior whose waiting results depend on declared sequencing and analysis assumptions.

Version
v1 · 2026-10-17 · History
Domain-specific #
14161
Domain group
Formal Sciences
Origin domain
Operations Research
Subdomain
Queueing Theory → Operations Research
Aliases
G/G/1 model

Core Idea

A G/G/1 queue is a model in which arrivals bring service requirements to one server, so an arrival may have to wait while that resource is occupied. The two Gs allow generally specified interarrival and service behavior; the 1 fixes the number of servers. The model links those inputs through an ordered queue state, rather than merely naming two distributions. Foss's exposition treats a stationary joint driving sequence and usually uses first-in-first-out (FIFO) service; it then studies independent identically distributed interarrival and service sequences as a more restricted GI/GI/1 case.[1]

The identity survives a change from packets to manufacturing lots because the same modeling question remains: when does work arrive, how much service does it require, and what happens when it encounters one busy resource? A packet-output model in a UC3M dissertation and a modeled workstation in a POLCA manufacturing study each answer that question with different carriers and objectives.[2][3] Neither use proves that its physical arrivals are independent renewal streams or that its numerical approximation is an exact law.

This is a model family, not a promise of FIFO service, a stationary wait, an infinite buffer, or a closed-form delay. A discipline, capacity rule, dependence structure, and performance question must be specified when a particular quantitative result needs them.[1][2]

Structural Signature

Four roles constitute the model. Two more roles are necessary for evaluating a quantitative application, but are not extra letters in the bare G/G/1 name.

  • Arrivals and interarrival process. Jobs, packets, or lots enter at times that establish the gaps between arrivals. Without an arrival process, there is no input to queue.[1][2][3]
  • Service requirements. Each arrival brings work for the resource. The requirements may vary; removing them removes the work whose congestion is modeled.[1][3]
  • One service resource. One modeled server processes work. Replacing it with two simultaneous servers changes the resource-count model, even when the arrival and service descriptions remain general.[1][3]
  • Queue state and sequence. Arrival times, service starts, and unfinished work are related in an ordered process. Two marginal distributions alone cannot tell which arriving job waits behind which work.[1]
  • Declared discipline and capacity, when solving a case. FIFO, priority, waiting-room limits, and loss rules affect a derived waiting result. A conditional FIFO recurrence cannot simply be carried into a differently scheduled application.[1][3]
  • Performance question and bounded output, when solving a case. A packet delay percentile and a station lead-time approximation ask different things of the same model form. The answer depends on the stated assumptions and method, not on G/G/1 notation alone.[2][3]

What It Is Not

  • Not identical to GI/GI/1. Foss's broad G/G/1 treatment starts with a stationary joint driving sequence; his GI/GI/1 subsection adds independent, identically distributed sequences and finite-mean conditions for particular results. Those restrictions cannot be read into every G/G/1 label.[1]
  • Not a waiting-time formula. Under Foss's FIFO indexing, Lindley's recurrence is W_(n+1) = max(0, W_n + s_n - t_n), where s_n is customer n's service time and t_n is the gap to the next arrival. It describes that conditional waiting sequence, not every priority or loss system.[1]
  • Not a Kingman result. A Kingman-style approximation can estimate a specified delay measure under additional conditions. UC3M computes a worked percentile estimate in §4.4.2 and separately tests theoretical estimators against simulations in §4.5; neither operation makes the model name an exact percentile law.[2]
  • Not an entire factory or transport network. A manufacturing shop can have several modeled workstation nodes, and a packet path can contain more than one resource. One G/G/1 node does not rename the whole network.[3]
  • Not G/G/2 or G/G/k. Those neighboring forms admit multiple servers. Their discussion in the frozen source member is contextual to this one-server model and does not supply a G/G/1 instance or establish a multi-server result here.

Scope of Application

The model is useful when an analyst can identify a single modeled bottleneck and wants to examine the waiting created as variable arrivals meet its service capacity. The UC3M dissertation models aggregated fronthaul packet flows at one output port of a 10 Gb/s switch as G/G/1. Packet timing supplies arrivals, serialization supplies service, and output-port contention supplies waiting.[2]

The POLCA article models each workstation node of a manufacturing shop as a single-class G/G/1 queue for an approximate performance analysis. Customer orders and aggregated lot movements supply arrivals; setup and processing contribute service requirements. The authors say that the one-server choice suits their application because most physical stations have one machine, and they note a possible multi-server extension. The complete shop remains a network of nodes.[3]

The recognition test is about the modeled resource, not the industry label. It does not certify that observed inputs satisfy independence, stationarity, FIFO, or a particular fitted distribution. Those are additional empirical or analytic commitments.[1][2][3]

Clarity

The notation separates two choices often compressed into one phrase. G/G/1 names general input and service behavior at one server. GI/GI/1, in Foss's restricted discussion, adds independent identically distributed sequences. A writer should state the dependence and stationarity conditions actually used rather than treating the two strings as interchangeable.[1]

It also separates a model from a result. Foss's FIFO recurrence has a particular indexing and sequencing convention. His finite-mean GI/GI/1 discussion analyzes a limiting wait under a load condition below one; that is not a universal stability theorem for every dependent, time-varying, capacity-limited G/G/1 application.[1]

Finally, distinguish source rounding from calculation. UC3M reports approximately 0.7 utilization for six flows of 1080 Mb/s on a 10 Gb/s output. The displayed rates alone divide to 0.648. Preserve the author's approximation and this arithmetic caveat; do not silently recalculate the dissertation's reported percentile from the rounded inputs.[2]

Manages Complexity

Packet traffic and factory production differ in units, timing, and operational control. G/G/1 compresses each modeled one-resource episode to arrivals, service work, and a waiting sequence. That compression lets an analyst ask which input summaries and scheduling assumptions matter before trying to carry over a delay method.[2][3]

It does not compress away variability for free. Both original applications use variability information in their approximate waiting analyses. Arrival or service variability can materially alter a modeled prediction, but its direction and magnitude belong to a stated method and regime; neither source supplies a universal numerical adjustment for all G/G/1 queues.[2][3]

The model also draws a boundary around a node. POLCA can use per-station approximations in planning a shop's lead time, but the resulting shop calculation must still respect that many nodes and release decisions interact. Isolating one server simplifies a question; it does not turn the network into one queue.[3]

Abstract Reasoning

First identify the candidate resource and test the four constitutive roles: timed arrivals, service requirements, one server, and an ordered waiting state. If two machines can serve at once, or the proposed unit is actually a network, the one-server model needs a narrower node boundary or a different form.[1][3]

Next specify what a requested answer requires. A waiting recurrence asks how customers are ordered; a stationary result asks about its driving sequence and hypotheses; an approximate percentile asks about its estimator and validation regime. A result justified for FIFO or an iid subsection does not follow from the broad G/G/1 label alone.[1][2]

This reasoning supports a practical intervention: before reusing a formula, compare the assumed discipline, capacity, dependence, and output measure with the modeled case. POLCA's due-date-based release and operator priority choices illustrate why an operational policy should not be mistaken for the FIFO sequence of a separate mathematical exposition. Release order alone does not establish the station's service law either.[3][1]

Knowledge Transfer

Within queueing analysis, the reusable move is to recognize and bound one server node. The packet-output case and the workstation case share that modeling move, even though one asks about a packet delay percentile and the other uses station waiting estimates in lead-time planning.[2][3]

What transfers is the question structure, not a numerical answer. A flow rate, packet size, service moment, or shop order policy must be re-established in the new case. UC3M's §4.5 simulation comparisons provide local evidence about its estimator, separate from the six-flow worked calculation in §4.4.2; POLCA's shop modeling does not validate that packet-tail estimate.[2][3]

Beyond queueing applications, one may analogize a shared resource and accumulating work, but the name G/G/1 belongs to a stochastic queue model with the constitutive roles above. A loose bottleneck metaphor is not another attested G/G/1 instance. Broader portability is a question for the live Queueing prime, not an automatic property of this domain entry.

Examples

Packet output at a 10 Gb/s switch. UC3M aggregates six fronthaul flows into one output-port model. The flows' packets and their interarrival timing fill arrivals; fixed 1500-byte payloads in the worked serialization calculation give service requirements of 1.2 microseconds of idealized serialization at that link rate; the output link is one server; packet contention yields the queue state and sequence. The source's case-specific switch/FCFS conventions supply the discipline used in analysis. Its output question is a conditional 90th-percentile delay estimate from a Kingman-style relation in §4.4.2. The dissertation separately compares its estimator with discrete-event simulation in §4.5. The thesis varies aggregate arrival variability and reports approximately 0.7 utilization, while six times 1080 Mb/s divided by 10 Gb/s is 0.648 from the stated rounded rates. Neither the approximation nor uncorrelated component flows prove iid aggregate arrivals or exact physical fit.[2]

A POLCA workstation model. In the manufacturing study, customer orders and aggregated manufacturing lots fill arrivals; setups and processing fill service requirements; a modeled workstation node is one server; lots awaiting that node supply the queue state and sequence. The authors summarize arrival and service moments and approximate waiting to inform station and shop lead-time planning, the output question. For the discipline role, distinguish assumptions made in a node calculation from physical release practice: due-date ordering and operator priorities are described for implementation, but do not by themselves prove FIFO service at every station. Most stations have one machine; the whole shop has multiple modeled nodes and is not one G/G/1 queue.[3]

Mapped back, each case has the same four constitutive roles and must declare its analytic choices. Packets and lots are different carriers, not different definitions of the queue.

Structural Tensions

Generality versus tractability. Broad arrival and service descriptions keep more candidate systems in view, while exact waiting claims require additional dependence, moment, or discipline hypotheses. Narrowing the model can permit a result but exclude the very behavior one meant to study; keeping it broad can leave only a bounded approximation. Diagnostic: which stated hypotheses actually justify this output?[1][2]

Node clarity versus network fidelity. One modeled server makes the packet output or workstation intelligible, but a shop or transport path may contain further resources and controls. A narrow node analysis can isolate a bottleneck; treating its result as a whole-network result can misattribute delay. Diagnostic: which waiting component belongs to this node, and which needs the surrounding network?[3]

Useful approximation versus tail accuracy. An estimated percentile or station wait can guide planning without a full exact solution. Across UC3M's tested loads, arrival-variability settings, and percentile levels, estimator-simulation gaps change; pushing that estimate into an untested regime may produce a misleading tail claim. Diagnostic: was the approximation checked against case-matched simulation or measurement at the requested regime?[2]

FIFO simplicity versus operational priority. A FIFO recurrence gives a transparent waiting sequence, while prioritized release or service may change the operative order. Choosing FIFO for calculational convenience can lose an important control feature; modeling every operational choice can sacrifice that simple analysis. Diagnostic: does the discipline assumed by the result match the modeled service sequence, rather than merely the release description?[1][3]

Structural–Framed Character

G/G/1 is strongly structural within its queueing frame. Its membership is fixed by the relationship among arrivals, service work, one resource, and a waiting sequence, rather than by an evaluator's preference for one outcome. That gives the notation real explanatory weight: the same four roles can be found in two unlike modeled applications.[1][2][3]

The entry is nevertheless framed by human modeling choices. Its formal research setting is queueing mathematics and operations research, visible in Foss's exposition and the UC3M and POLCA analyses; no named paper, university, or workplace is a constitutive role. Analysts choose a node boundary, decide which events count as arrivals and service, and select a discipline and output measure for a calculation. An institution may supply production orders or network traffic, but it does not create the mathematical one-server relation; conversely, that relation does not certify the institution's actual process. The vocabulary travels naturally between queueing applications, while a resemblance to waiting outside queueing is only an analogy unless the stochastic model roles are established.

Importing this model means supplying those roles and checking the case assumptions. Merely recognizing a busy resource is insufficient, and no performance result follows from the name alone. The portable skeleton is the broader Queueing concept of arrivals, service, and waiting, which would require its own full-signature test before any typed parent is approved. Its character: a structurally precise one-server model with domain-specific stochastic interpretation and assumption-bound quantitative use.

Structural Core vs. Domain Accent

The entry-level structural core consists of timed arrivals with general interarrival behavior (R1), generally described service requirements (R2), exactly one modeled server (R3), and an ordered queue state (R4). Removing the single-server condition changes this entry's identity, rather than merely changing its accent. These roles explain why packet output and a workstation can be mapped into the same model question. The domain accent is their queueing-specific carrier and case-variable analysis: packet serialization versus lot setup and processing, plus the declared discipline, capacity, and performance approximation when a calculation requires them.[1][2][3]

The named entry does not clear the Prime bar merely because one can recognize a bottleneck elsewhere. Its one-server stochastic notation and service semantics are specialized; dropping them leaves a broad analogy rather than this model. Gate 4 found no all-instance full-signature or ancestor proof for a current live parent. A more general queue-model genus is a future research question; the approved unparented placement is provisional and does not imply this specialized model is a Prime.

The G/G/1 queue has no broader abstraction in the encyclopedia yet. Queueing is the closest general abstraction: the model concerns arrivals, service, and waiting. The relationship is conceptual, though: Queueing's full definition, and those of the abstractions above it, do not hold for every purely formal G/G/1 model, so the G/G/1 queue is not counted as a kind of Queueing.

The Lindley equation concerns a reflected waiting recurrence. It can analyze a FIFO sequence under suitable indexing and conditions, but is not this entire model. Kingman's formula concerns a conditional waiting approximation, also narrower than the model. A layered queueing network concerns interacting queues; its network structure cannot be substituted for the one-server resource count.[1][2][3]

Neighborhood in Abstraction Space

G/G/1 Queue sits in a sparse region of the domain-specific corpus (90th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2721 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-17

Not to Be Confused With

  • GI/GI/1: ask whether iid and independence restrictions have actually been imposed. They are extra assumptions in Foss's subsection, not the definition of every G/G/1 case.[1]
  • G/G/2 or G/G/k: ask how many resources can serve simultaneously. The frozen original member’s multi-server paragraphs are contextual neighboring-model material outside this entry; they are not positive instances of this one-server model.
  • A FIFO or priority policy: ask how the actual service sequence is controlled. A policy can change a result without necessarily changing the broad one-server model designation.[1][3]
  • Lindley's recurrence or a Kingman-style estimate: ask whether the statement describes the modeled system or a conditional analysis of it. Neither formula is the whole G/G/1 identity.[1][2]
  • A manufacturing shop network: ask whether the claimed unit is one station or the arrangement of many stations. POLCA uses G/G/1 at modeled nodes, not as the name of the whole plant.[3]

References

[1] Foss, Sergey (undated). “The G/G/1 queue”. Original seven-page author-hosted mathematical exposition, PDF p. 1 (§1 and opening paragraphs), with general-case discussion later in the document. Author PDF. No publication year or DOI is inferred. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w

[2] Otero Pérez, Gabriel (2020). “Design and Analysis of Ultra-low Latency Fronthaul and Backhaul Networks for 5G”. Doctoral dissertation, Universidad Carlos III de Madrid, §2.7.2, printed p. 36 (physical PDF index 61, FCFS switch convention), and Chapter 4 §§4.4.2–4.5, printed pp. 62–65 (physical PDF indices 87–90). Institutional PDF. The cover says December 2020; the internal title page is dated 6 November 2020. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t

[3] Vandaele, Nico; Van Nieuwenhuyse, Inneke; Claerhout, Diederik; and Cremmery, Rony (2008). “Load-Based POLCA: An Integrated Material Control System for Multiproduct, Multimachine Job Shops.” Manufacturing & Service Operations Management 10(2): 181–197. DOI: 10.1287/msom.1070.0174. See §§3.2.1–3.2.2, printed pp. 186–188, and §4, printed pp. 191, 193–194. Original journal-layout PDF copy. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w ↩x