Retrial Queue¶
A queueing model in which blocked requests enter an orbit and later retry admission to a finite-capacity service facility under an explicit retrial policy.
Core Idea¶
A retrial queue models a finite-capacity primary service facility in which a blocked request enters an external orbit and later attempts admission again. Fresh and returning attempts compete for capacity. A primary facility may have one or multiple servers and can include finite waiting places; the zero-waiting-room case is not universal. The orbit differs from an ordered line because an orbiting request must reattempt access rather than necessarily retaining the next service position.[ref-8c785b09e4ee][ref-1c59411451ca]
The retrial policy is decisive. Independent exponential clocks of individual rate \(\nu\) give an aggregate \(n\nu\) retry intensity for \(n\) orbit members. A constant-rate policy instead gives the nonempty orbit an aggregate attempt rate \(\nu\) independent of \(n\). Those policies can yield different backlog and stability behavior. The frozen Wikipedia article is a short discovery stub; these qualifications come from original research.[ref-7f18898de6cf][ref-1c59411451ca]
Scope of Application¶
Original models address telephone/call-center redial and carrier-sensing medium access. Aguir and colleagues analyze a multi-server call center with retrials plus balking and impatience; Avrachenkov, Nain and Yechiali analyze two input classes with separate retrial orbits and class-specific Poisson re-dispatch rates.[ref-b6894377532f][ref-f550a8831cb9] The model does not assert that all callers retry, all network failures are collisions or all orbits retain requests forever.
Clarity¶
An external arrival and a retrial attempt are different events: one request can make several attempts before service. A blocked request may be lost in a pure loss system, hold an ordered position in a conventional waiting line, or leave to retry in a retrial queue. The last choice creates feedback from prior blocking into future attempted load.[^ref-8c785b09e4ee]
There is no universal stability inequality. Liang and Kulkarni report \(\lambda/\mu<1\) as sufficient for a specified \(M/G/1/1\) model with exponential retrial times and give a counterexample to generalizing it to arbitrary single-server retrial queues. In a different \(M/M/1/1\) constant aggregate retrial-rate case, Avrachenkov and Morozov's Markovian criterion specializes to \(\lambda(\lambda+\nu)<\mu\nu\), where \(\lambda\) is fresh arrival rate, \(\mu\) service rate and \(\nu\) the nonempty orbit's aggregate attempt rate. These are model-specific statements, not competing universal laws.[ref-8c785b09e4ee][ref-1c59411451ca]
Manages Complexity¶
Separating primary occupancy from orbit population reveals why blocking is not always final loss. A request outside the facility can later return, and the attempt stream experienced by the facility is therefore affected by its past congestion. Explicitly stating the retry discipline prevents a constant-rate orbit from being mistaken for independent customer redial: doubling orbit size doubles \(n\nu\) under the simple independent exponential model but leaves the aggregate \(\nu\) clock unchanged in the constant model.[ref-7f18898de6cf][ref-1c59411451ca]
Abstract Reasoning¶
Identify primary capacity, fresh arrivals, the blocking rule, where blocked requests reside and how they reattempt. Ask whether retryers keep an ordered waiting place, whether they may abandon, and whether the retry rate is per request or per orbit. Only then select a stability or delay result whose arrival, service, capacity and retrial assumptions match the system.[ref-8c785b09e4ee][ref-b6894377532f]
Knowledge Transfer¶
In the call-center case, the bounded resource is a group of agents, fresh calls enter, unsuccessful callers may return, and service or abandonment removes them. In the medium-access model, a shared server receives two input classes and blocked jobs enter class-specific orbits that redispatch later. The preserved structure is finite admission → external orbit → repeat attempt → service or continued blocking, not a shared numerical formula or a claim that packets are identical to callers.[ref-b6894377532f][ref-f550a8831cb9]
The live Queueing prime is a close neighbor but currently defines waiting-line accumulation and a discipline. Because independent retrial orbits need not reserve positions in such a line, this author draft proposes no strict parent pending review. Its stochastic orbit-feedback core is a domain-specific queueing model, not a domain-free prime.
[^ref-8c785b09e4ee]: Huei-Mei Liang and V. G. Kulkarni, “Stability condition for a single-server retrial queue”, Advances in Applied Probability 25 (1993), pp. 690–701, original publisher abstract. [^ref-7f18898de6cf]: Tewfik Kernane, “Conditions for stability and instability of retrial queueing systems with general retrial times”, original author manuscript, Introduction and retrial-policy definitions. [^ref-1c59411451ca]: Konstantin Avrachenkov and Evsey Morozov, “Stability Analysis of GI/G/c/K Retrial Queue with Constant Retrial Rate”, original author report (2010), abstract and §3.1 Eq. (38), specialized here only to \(M/M/1/1\). [^ref-b6894377532f]: Mohamed Salah Aguir, Fikri Karaesmen, O. Zeynep Aksin and Fabrice Chauvet, “The impact of retrials on call center performance”, OR Spectrum 26 (2004), original abstract. [^ref-f550a8831cb9]: Konstantin Avrachenkov, Philippe Nain and Uri Yechiali, “A retrial system with two input streams and two orbit queues”, original author manuscript (2012), abstract and Introduction.
Neighborhood in Abstraction Space¶
Retrial Queue sits in a sparse region of the domain-specific corpus (83rd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Unclustered & Miscellaneous (2551 abstractions)
Nearest neighbors
- Volunteer-Management Overload — 0.82
- Sequence number — 0.82
- Byzantine Generals Problem — 0.81
- Convergence Failure — 0.81
- Parking Orbit — 0.81
Computed from structural-signature embeddings · 2026-10-08