Hotplug Coded Caching

• Hotplug coded caching is a caching model where placement is finalized before the actual active users are revealed, preserving coded multicasting gains under user uncertainty.
• It leverages array-based formulations like HpPDA to systematically design placement and delivery strategies that adapt to dynamic active sets and helper associations.
• Key scheme families, including MAN-HpPDA and t-design based methods, optimize memory-rate tradeoffs while addressing extensions such as privacy, secrecy, and hierarchical network applications.

Searching arXiv for recent and foundational papers on hotplug coded caching and related dynamic/shared-cache variants. Hotplug coded caching denotes a family of coded-caching models in which cache placement is completed before the delivery-time participation pattern is known. In the literature, the term most often refers to a dedicated-cache system with $K$ placed users but only $K'\le K$ active users in delivery, so placement knows the value of $K'$ but not the active set $\mathcal I\in\binom{[K]}{K'}$. Closely related work uses an analogous late-binding model in shared-cache networks, where user-to-helper attachment is unknown during placement and revealed only afterward. In both interpretations, the central problem is to preserve coded-multicasting gain under uncertainty in who is online, which helper serves whom, or which caches remain accessible, without redoing placement after the actual delivery topology is revealed (Rajput et al., 2023, Chinnapadamala et al., 2024, Peter et al., 2022).

1. Canonical model and scope

In the dedicated-cache hotplug model, the server stores $N$ files, serves $K$ users over a shared link, and performs placement knowing that exactly $K'$ users will later be active, but not which users those will be. Delivery begins after a subset $\mathcal I\in\binom{[K]}{K'}$ appears and reveals demands. This differs from the original Maddah-Ali–Niesen setting, which assumes that all $K$ users are active and that the server waits for all demands before transmission. Later privacy-oriented work keeps the same hotplug structure and denotes the active set by $\mathcal A\in\Omega_{[K]}^{K'}$, with the remaining users offline (Chinnapadamala et al., 2024, Ma et al., 2023).

The literature also uses a broader hotplug interpretation for topological uncertainty. In the shared-cache model with helper nodes, one server serves $K'\le K$0 users through $K'\le K$1 helper caches, each user attaches to exactly one helper, and the unknown quantity is the user-to-helper association rather than the active user subset. The case in which the association is unknown during placement and revealed only afterward is explicitly identified as the closest formalization of hotplug or dynamic attachment after placement in that setting (Peter et al., 2022).

This scope matters because not every dynamic coded-caching problem is hotplug coded caching in the same sense. Asynchronous-request models keep the user population fixed and let requests arrive at different times; dynamic centralized designs preserve incumbent users’ caches while onboarding new users across rounds; and adaptive-delivery schemes may assume arbitrary cache states at delivery time without modeling user arrival or departure during an episode. These are adjacent problems, but they solve different parts of the broader dynamical-caching design space (Ghasemi et al., 2019, Zhang et al., 2019, NaderiAlizadeh et al., 2019).

2. Array-based formulations

The dominant combinatorial abstraction for dedicated-cache hotplug systems is the hotplug placement delivery array (HpPDA). An HpPDA is a pair $K'\le K$2 where $K'\le K$3 is an $K'\le K$4 array of stars and nulls, with $K'\le K$5 stars per column, and $K'\le K$6 is a $K'\le K$7-PDA. The defining property is that for every active set $K'\le K$8, $K'\le K$9, there exists a row subset $K'$0, $K'$1, such that the restricted subarray $K'$2 has the same star pattern as $K'$3. This gives a $K'$4 hotplug coded-caching scheme with subpacketization $K'$5, memory ratio $K'$6, and rate $K'$7 by first splitting each file into $K'$8 parts, then applying an $K'$9 MDS code, placing coded subfiles according to $\mathcal I\in\binom{[K]}{K'}$0, and running PDA delivery according to $\mathcal I\in\binom{[K]}{K'}$1 on the realized active set (Rajput et al., 2023).

A later refinement keeps the same HpPDA pair but changes the MDS dimension. For any $\mathcal I\in\binom{[K]}{K'}$2-HpPDA, the scheme in (Chinnapadamala et al., 2024) encodes each file from $\mathcal I\in\binom{[K]}{K'}$3 source subfiles into $\mathcal I\in\binom{[K]}{K'}$4 coded subfiles and achieves

$\mathcal I\in\binom{[K]}{K'}$5

The reason is structural: each active user already holds $\mathcal I\in\binom{[K]}{K'}$6 coded symbols from placement and receives $\mathcal I\in\binom{[K]}{K'}$7 additional coded symbols from delivery, so it accumulates $\mathcal I\in\binom{[K]}{K'}$8 coded symbols, sufficient for MDS decoding. For the same HpPDA, this shifts the operating point to a lower cache-memory ratio than the earlier $\mathcal I\in\binom{[K]}{K'}$9 normalization (Chinnapadamala et al., 2024).

The same array methodology has been generalized in two directions. In two-layer hierarchical networks with mirror caches and user caches, the hierarchical hotplug placement delivery array (HHPDA) uses one array for mirror placement and local user structure, and one PDA for the active-user projection. The resulting scheme has mirror memory ratio $N$0, user memory ratio $N$1, first-layer load $N$2, and a second-layer load $N$3 determined by the local symbol sets induced at each mirror (T. et al., 1 Jul 2025). In combinatorial multi-access hotplug networks, a generalized HpPDA uses a cache-placement array $N$4, a derived user-access array $N$5, and a family of PDAs $N$6 indexed by the number of online caches accessible to a user. Under the decodability condition $N$7, it yields

$N$8

for the $N$9 combinatorial multi-access hotplug model (Singh et al., 15 Jan 2026).

3. Dedicated-cache scheme families

A first major family is the MAN-HpPDA family, which shows that the Ma–Tuninetti hotplug construction can be represented as an HpPDA. For $K$0, the parameters are

$K$1

This produces the hotplug memory-rate point

$K$2

before repeated-demand reduction. The same paper observed that when $K$3, the standard delivery transmits excess coded symbols per active user. By deleting a set of PDA labels $K$4 such that $K$5 in every column, it reduced the MAN-HpPDA rate to

$K$6

This is the first systematic rate-improvement mechanism internal to the HpPDA framework (Rajput et al., 2023).

A second family constructs HpPDAs from $K$7-designs. For a $K$8-$K$9 design with point-incidence parameter $K'$0, choose integers $K'$1, $K'$2, and define

$K'$3

The resulting $K'$4-scheme achieves

$K'$5

and the same label-deletion argument yields an improved rate

$K'$6

For $K'$7, the improved $K'$8-scheme reaches the cut-set line $K'$9 at specific high-memory points arising from $\mathcal I\in\binom{[K]}{K'}$0-design families, and the paper shows that this optimal regime starts at a smaller memory ratio than the baseline hotplug scheme (Rajput et al., 2023).

The MDS reuse of existing HpPDAs sharpened these tradeoffs further. For the $\mathcal I\in\binom{[K]}{K'}$1 hotplug system derived from a $\mathcal I\in\binom{[K]}{K'}$2-$\mathcal I\in\binom{[K]}{K'}$3 design, the new scheme in (Chinnapadamala et al., 2024) turns two earlier HpPDAs into the points $\mathcal I\in\binom{[K]}{K'}$4 and $\mathcal I\in\binom{[K]}{K'}$5, and the paper reports that these points outperform the compared schemes in the memory range $\mathcal I\in\binom{[K]}{K'}$6. It also reports outperforming ranges $\mathcal I\in\binom{[K]}{K'}$7 for $\mathcal I\in\binom{[K]}{K'}$8 and $\mathcal I\in\binom{[K]}{K'}$9 for $K$0, while explicitly noting that the improvement comes with increased subpacketization (Chinnapadamala et al., 2024).

4. Shared caches, helper attachment, and topological generalizations

Hotplug ideas also appear in shared-cache systems, where the uncertainty concerns attachment rather than merely activity. In the shared-cache model with private user caches, the network contains one server, $K$1 users, $K$2 helper caches of size $K$3, and one private cache of size $K$4 per user. The hotplug-like case is the one in which the user-to-helper association is unknown during placement and revealed only before delivery. The proposed scheme splits each file into a helper-cache part of fraction $K$5 and a private-cache part of fraction $K$6, and achieves

$K$7

with

$K$8

The helper and private components are placed independently, and the delivery is optimal for that placement via an index-coding converse. This captures late binding of users to helpers after prefetching, but it still assumes fixed $K$9, fixed $\mathcal A\in\Omega_{[K]}^{K'}$0, exactly one helper per user, centralized placement, and static attachment during delivery (Peter et al., 2022).

When the helper association is known in advance, the same paper shows that significantly stronger coupling is possible. Its joint shared-plus-private coding scheme achieves exact optimality in the regime

$\mathcal A\in\Omega_{[K]}^{K'}$1

with

$\mathcal A\in\Omega_{[K]}^{K'}$2

This benchmark quantifies the price of not knowing the attachment structure during placement (Peter et al., 2022).

Two other topological extensions make hotplug behavior more structural. In shared-cache networks, generalized PDAs replicate cache columns according to the realized user-to-cache profile, so placement is association-oblivious and delivery adapts after association is revealed; this gives low-subpacketization schemes and recovers the optimal Parrinello–Ünsal–Elia shared-cache scheme as a special case when the starting PDA is the MAN PDA (Peter et al., 2021). In a different direction, a matrix-and-design-based shared-cache scheme fixes subpacketization at $\mathcal A\in\Omega_{[K]}^{K'}$3, does not make it depend directly on the number of users or caches, and explicitly allows new caches to be added without redoing the placement of the existing caches, provided the enlarged matrix preserves the required circuit structure. This is a genuine cache-hotplug property, although delivery must still be recomputed for the expanded system (Das et al., 2022).

Hierarchical and multi-access hotplug models extend the same principle to more complex topologies. HHPDA-based schemes handle two-layer networks with cache-aided mirrors and offline users, generalizing earlier work restricted to zero mirror memory (T. et al., 1 Jul 2025). Combinatorial multi-access hotplug networks allow each user to access an $\mathcal A\in\Omega_{[K]}^{K'}$4-subset of $\mathcal A\in\Omega_{[K]}^{K'}$5 caches, declare a user active whenever $\mathcal A\in\Omega_{[K]}^{K'}$6 for the online-cache set $\mathcal A\in\Omega_{[K]}^{K'}$7, and use $\mathcal A\in\Omega_{[K]}^{K'}$8-design-based generalized HpPDAs to obtain flexible rate-memory-subpacketization tradeoffs in which redundant multicast transmissions can be removed by parameter selection (Singh et al., 15 Jan 2026).

5. Privacy, secrecy, and adaptive delivery

Hotplug coded caching has also developed privacy-constrained variants. In demand-private hotplug caching, only $\mathcal A\in\Omega_{[K]}^{K'}$9 users are active, but each active user requests a scalar linear combination of the $K'\le K$00 files, represented by a vector $K'\le K$01. The privacy condition requires

$K'\le K$02

Two MDS-based schemes are proposed. The first achieves

$K'\le K$03

with subpacketization $K'\le K$04, and matches the foresight benchmark in the small-memory regime

$K'\le K$05

A second scheme covers the large-memory endpoint $K'\le K$06 with rate $K'\le K$07 (Ma et al., 2023).

A stronger secretive hotplug coded-caching model requires both cache secrecy and delivery secrecy. Its conditions include

$K'\le K$08

For MAN-HpPDAs, the proposed secretive scheme achieves

$K'\le K$09

with the MAN-HpPDA parameters $K'\le K$10, $K'\le K$11, $K'\le K$12, $K'\le K$13, and $K'\le K$14. In the special case $K'\le K$15, it attains the secrecy lower bound $K'\le K$16 at

$K'\le K$17

A second secretive construction based on $K'\le K$18-design HpPDAs achieves

$K'\le K$19

and both secretive schemes outperform a baseline adaptation of classical secretive PDA coding in certain memory regions (Chinnapadamala et al., 18 Jul 2025).

A distinct but adjacent direction is adaptive delivery for arbitrary realized cache states. A deep reinforcement learning approach treats caches as arbitrary, uses a full cache/request state representation of dimension $K'\le K$20, optimizes delivery only, and reports worst-case inference complexity $K'\le K$21 versus SACM’s $K'\le K$22. This is relevant to hotplug-like arbitrary side information at the start of delivery, but it does not model user join/leave during an episode and keeps $K'\le K$23 fixed (NaderiAlizadeh et al., 2019).

6. Dynamic interpretations, limitations, and open problems

Two earlier dynamic lines of work clarify what hotplug coded caching does and does not include. Dynamic centralized coded caching across multiple rounds considers a persistent set of fixed users and a set of mobile users joining later, keeps the fixed users’ caches unchanged, fills only the new users’ caches, and then couples both groups through concatenating placement and saturating-matching delivery. For the $K'\le K$24 model with $K'\le K$25, it derives an achievable rate depending on whether $K'\le K$26 or $K'\le K$27, and proves a constant-factor order-optimality guarantee

$K'\le K$28

via comparison to an order-optimal heterogeneous-cache benchmark (Zhang et al., 2019). Asynchronous coded caching, by contrast, keeps a fixed pre-cached user population, lets requests arrive at different times with deadlines, formulates an offline linear program over time intervals and user groups, and proposes an online heuristic that succeeds with probability at least $K'\le K$29 when it does not declare infeasibility. Its dynamicity is in demand arrivals, not in late user admission without preloaded caches (Ghasemi et al., 2019).

Across the literature, several limitations recur. Many hotplug models fix the total number of users or caches and vary only which subset is active. Dedicated-cache formulations usually assume that the value of $K'\le K$30 is known during placement, though the active identities are not. Shared-cache late-attachment formulations typically keep $K'\le K$31 and $K'\le K$32 fixed, require one-helper-per-user attachment, and do not model mid-delivery reassociation. Hierarchical and multi-access formulations likewise assume a fixed number of active users or online caches, not a variable-size active population within one universal placement (Peter et al., 2022, T. et al., 1 Jul 2025, Singh et al., 15 Jan 2026).

This suggests a sharp distinction between current hotplug coded caching and a more general online topology problem. Current work is strong on robustness to one kind of late uncertainty—offline users, unknown helper attachment, unknown online-cache subset, or new caches appended under a preplanned algebraic structure—but it does not yet provide a single theory for simultaneous user arrivals, user departures, helper failures, helper additions, multi-helper attachment changes during delivery, and cache-state evolution under one universal placement. A plausible implication is that future progress will require combining the array-based universality of HpPDA-type designs with explicit update-complexity, signaling, and state-migration metrics, which are largely absent from the present formulations (Das et al., 2022, Rajput et al., 2023).