The rated speed is one link in a chain — the transfer moves at the speed of the slowest.
Latency in its parts: propagation is distance ÷ signal speed in the medium (light in fiber travels about 200,000 km/s, about 5 µs per km; copper runs at a velocity factor of roughly 0.66–0.85 of c by cable type, planned here at 0.7; free space is c itself). Serialization is packet size ÷ link rate — the time to put one packet on the wire: a 1,500-byte packet takes 12 µs at 1 Gbps and 1.2 ms at 10 Mbps. Processing and queuing are the variable remainder, entered as an allowance. RTT is the round trip; one-way figures are kept distinct and labelled. The bandwidth-delay product (rate × RTT) is the data that must be in flight to fill the path, and a single TCP flow cannot exceed window ÷ RTT — window scaling raises the window, and parallel flows multiply the ceiling until another constraint binds. Jitter is the variation in delay, not the delay: the mean deviation between packet spacing at arrival and at sending (RFC 3550, in plain words). Real-time media absorbs jitter with a buffer set to at least the jitter, and the buffer costs latency against the one-way voice budget of about 150 ms (ITU-T G.114); jitter beyond the buffer arrives as loss, not slowness. Direction is a quantity: a symmetrical service rates both directions alike, an asymmetrical one does not, and a saturated upload also throttles downloads, because TCP acknowledgements travel upstream. The effective rate of a download is the minimum of its chain — the access rate in that direction, the server’s per-connection cap, the server’s uplink, the window ceiling, and what remains after the overhead allowance — and the tool names the link that binds. Figures are computed from the entered values; a measured path adds routing and equipment delays the allowance fields stand in for.
One node, or a transmission line? The bound is stated before the interconnect is designed.
The criterion, with its tiers as the books state them. A structure is treated as one lumped node — no phase variation along it — only while it stays short against the wavelength at the highest frequency it carries, λ = vprop ÷ f. Within a hundredth of a wavelength (L ≤ 0.01λ) the lumped treatment is highly accurate; within a tenth (L ≤ 0.1λ) it stands as the general boundary; between the two lies a transition band (Book One A). In time the same boundary reads tpd ≤ tr/2 — the propagation delay within half the rise time (Book One B §1.4.3) — equivalently, L ≥ vp·tr/2 marks the distributed side (Book Nine). Book Five states the boundary qualitatively for interconnects: L much less than λ, the length against the clock wavelength. Past the boundary the structure is a transmission line, and the quantities above — propagation, serialization, the window ceiling — are the ones that govern it.
A different jitter from the network jitter above: not the variation in packet delay, but the converter’s own sampling instant — the bound is stated before the digitizer is designed.
The bound, as the book states it: an analog-to-digital converter’s sampling instant is uncertain by σt, and a full-scale signal at fin then carries a jitter-limited signal-to-noise ratio; the book’s relation inverts it — σt ≤ 10(−SNR/20)/(2πfin) (Book Three, Line 4049, Ch 23, in an analog/mixed-signal VLSI treatment). At the book’s stated state — SNR 74 dB, fin 1 GHz — the bound computes to 31.76 fs; the book prints ≈ 31.7 fs, and the derivation registry records the instance exact. Line b653 of the same book states the same relation in its SNR form at its own Ch 23; the worked instance, and this panel’s ground, remains Line 4049’s. The jitter named here is not the jitter of the first section: that is the variation in packet delay across a network; this is the converter’s aperture uncertainty — the timing substrate beneath its noise figure, standing at the digitizer of a receiver chain. The same chapter’s quantization metrics — step size, SQNR, ENOB — are a different measurable family and are not computed here.
A signal does not travel at c once it is inside a material: the substrate it travels through sets its speed. Book Nine’s interconnect treatment states the relation (registry Entry 32): the propagation velocity in a dielectric, vp = c/√(εrμr), and its reciprocal reading, the propagation delay per unit length. State the medium — its relative permittivity εr and relative permeability μr — and the panel returns the velocity, the delay per unit length, and the delay across a stated length; state a frequency and it returns the wavelength in the medium, the form Book Forty-One computes in code. Every result carries the medium state it stands in. The panel opens on Book Forty-One’s stated medium state entire — standard FR-4 at εr = 4.4, with μr at the entry’s stated 1; the length and the frequency open at round visitor values, each labeled a visitor input at its field.
The same medium read at a frequency: the wavelength shortens by the same root factor that slows the wave.
The relation stands in Book Nine, Ch 7 (the interconnect treatment; the Book Nine census register, B9-R4), and the derivation registry carries it as Entry 32: vp = c/√(εrμr); delay per unit length = 1/vp. The entry’s attested medium state is FR-4 at εr ≈ 4.2 with μr = 1 in the passage’s reading, returning vp ≈ 1.46 × 108 m/s — exact at the book’s stated precision — and a delay of ≈ 6.85 ps/mm read from the book’s rounded velocity (the unrounded path at the passage’s figures gives 6.83 ps/mm; the registry records both, the delay figure following the book’s rounded velocity state). Book Forty-One states the delay member outright in dielectric form — tpd = L·√εr/c — with c = 2.99792458 × 108 m/s and εr = 4.4 for standard FR-4, and computes the wavelength in the medium in code, λ = c/(f·√εr), at the bound seats of its router treatment (the Book Forty-One deduction, D13). This panel computes with c at Book Forty-One’s stated value entire, so at Book Nine’s state its unrounded delay reads 6.836 ps/mm beside the book’s two recorded readings — the difference is the c state, recorded here, neither side corrected. Book Thirty-Three carries the entry in reduced state: the delay member in worked use inside a proved time bound, its velocity element glossed at the seat as vprop ≈ 2 × 108 m/s (Lemma 39.1, §39.6.2; the Book Thirty-Three deduction, D4) — a correspondence standing, recorded as such; it is a gloss inside another book’s apparatus, not a medium state, and no preset is built on it. The entry’s relation to the sections above is its own: they compute propagation over link distances, and this member states the velocity in the dielectric itself, at board level — delay as a property of the material the signal travels through, where the host sections treat it as a property of distance. Within SolveForce’s collective work the entry corresponds to the Connectivity and Networks & Data Centers pillars (board- and fabric-level interconnect), and serves no finder directly — recorded.
The two FR-4 states stand as two books’ states, neither corrected into the other: Book Nine’s εr ≈ 4.2 (approximate at its statement) and Book Forty-One’s εr = 4.4 (stated for standard FR-4). μr stands in no form in Book Forty-One; the field opens at the entry’s stated μr = 1 from Book Nine’s passage and is labeled an input wherever it stands. The length and frequency fields open at round visitor values and are visitor inputs entire — no book in the entry’s ground states a length or a frequency for this relation, and no opening value was adjusted to reproduce any printed figure. The λ/10 bound at which Book Forty-One’s code checks the wavelength against a conductor length belongs to the lumped–distributed criterion’s seat (Entry 3), not to this entry, and is not computed here. Every default that is a book’s is named as the book’s at its field; everything else is the visitor’s to state.