Tensions in Practice: Fixed slot assignment in tension with borrowing idle shares¶
Two logical streams · four transmission slots
A channel has four slots per interval. Reserving two for each stream leaves half the channel idle when only one stream requests four slots. Letting the active stream borrow the idle share serves all four. But when both request four together, the channel still has only four slots. Under the declared equal-sharing rule, each receives two; the extra service seen while alone was conditional on the other stream being quiet.
Keep slot assignment simple
Repeat a fixed two-slot partition without detecting and reallocating each interval’s unused share.
Use otherwise idle shares
Let one active stream consume capacity the other does not currently need.
Why these aims pull against each other
Borrowing improves use of noncoincident demand without increasing physical capacity. A peak rate observed in isolation is not an unconditional promise to every stream at once. Borrowing also requires demand-aware assignment machinery; the two-slot joint-demand service is identical in both arrangements.
Choose an arrangement to see what changes and what remains difficult.
Finite illustrative comparisons. Labels carry the meaning; color does not establish a preference or measured effect.
What this choice protects
What it costs
When it fits
Compare the arrangements
Reserve two each
Assign two slots per stream, leaving unused reservations idle. The toy has no carryover between intervals.
| Requested A / B | Served A / B | Idle slots | |
|---|---|---|---|
| A alone | 4 / 0 | 2 / 0 | 2 |
| B alone | 0 / 4 | 0 / 2 | 2 |
| Both peak | 4 / 4 | 2 / 2 | 0 |
- What it protects
- A fixed schedule supplies two slots per stream without demand-aware reassignment.
- What it costs
- One stream’s unmet requests coexist with idle slots reserved for the quiet stream.
- When it fits
- Fits when simple repeated slot ownership is preferable to the added assignment machinery needed to recover idle reservations.
Illustration note: This does not create separate physical failure domains; both streams still share the same channel.
Borrow idle slots
Allow the sole active stream to use all four slots. When both request four, alternate slots so each receives two.
| Requested A / B | Served A / B | Idle slots | |
|---|---|---|---|
| A alone | 4 / 0 | 4 / 0 | 0 |
| B alone | 0 / 4 | 0 / 4 | 0 |
| Both peak | 4 / 4 | 2 / 2Four unserved | 0 |
- What it protects
- The single-active scenarios use all four slots instead of leaving two idle.
- What it costs
- The receiver and scheduler must identify active demand and assign borrowed slots while preserving stream recovery. This adds control and bookkeeping work; its overhead is not included in the four-slot arithmetic.
- When it fits
- Fits when recovering idle slots justifies demand-aware assignment and its implementation can preserve the declared two-slot joint-demand service.
Illustration note: Equal division during joint demand is stipulated, not supplied by multiplexing alone. The table does not estimate how often either demand pattern occurs. The added tracking and reassignment are implementation requirements inferred from this demand-dependent rule; T5 supplies the general machinery trade.
What this illustration does—and does not—establish
Multiplexing: Statistical (oversubscribed) multiplexing wins on average but fails under correlated demand supplies the correlated-demand boundary. The finite allocation keeps capacity fixed and makes conditional borrowing visible without claiming additional capacity. Multiplexing: Multiplexing relocates complexity rather than removing it, and the new locus can be costlier supplies the added scheduling and recovery machinery; fixed versus demand-aware assignment specializes that general source trade.
- No demand independence or measured utilization gain is assumed; the scenarios show why coincidence matters.
- The unserved requests require a queue, rejection or later-slot policy outside this single-interval model.
- Correct stream labeling and demultiplexing are assumed; extra slots do not remove protocol overhead.
Source entries
Multiplexing
Multiplexing: Statistical (oversubscribed) multiplexing wins on average but fails under correlated demand supplies the conflict examined here.
Statistical (oversubscribed) multiplexing wins on average but fails under correlated demand
Many practical systems multiplex more streams than the substrate could carry if all were maximally active, betting that demand is bursty and uncorrelated so the average load fits. This statistical multiplexing yields far higher utilization than rigid slot-per-stream allocation. But the bet fails precisely when demand becomes correlated — everyone transmits at once, every process wakes together, every car arrives at the green — and the oversubscribed substrate collapses into congestion exactly when it is needed most.
What It Is Not
The substrate's total bandwidth, throughput, or duty cycle still bounds the sum of the streams; multiplexing reallocates a fixed capacity among many users rather than enlarging it.
Multiplexing relocates complexity rather than removing it, and the new locus can be costlier
Replacing N physical channels with one channel plus a partition rule reduces hardware but introduces scheduling, synchronization, addressing, and demultiplexing machinery. For some workloads this trade is decisively favorable; for others the protocol overhead, latency variance, and synchronization fragility cost more than the hardware they save.