fs: the NC8400-4TH 4-slot chassis as a stated 4x NC8400-32C fill (128x 100G QSFP28) #1719

Merged
jaylouisw merged 1 commit from catalogue/1116-1955 into main 2026-11-16 19:59:54 +00:00
Owner

Fixes #757

The 2026-11-10 pass left this open because FS publishes a slot count and a module set (a bay, not a port). Option (a) in that comment was "you name the module set for the SKU you want" - and FS's own data sheet names it for us: "The NC8400-4TH switch can be loaded with line card NC8400-32C and supports a maximum of 128x 40G/100G high-density full line rate ports." So the chassis is modelled as that fill, exactly the way NC8200-4TD is modelled as 4x NC8200-24BC (#1690), and the label says which fill it is: NC8400-4TH (4x NC8400-32C).

What the spec says

field value source
line-card slots 4 data sheet Table 1
ports 128x HundredGigabitEthernet<slot>/<port>, QSFP28, 100G (40G capable) Table 1: "100G port density 128", "40G port density 128"; Table 2: 32C = 32x 40G / 32x 100G
console 1 (type: console) Table 1: "Console port 1"
fabric 12,800,000 Mbps (12.8 Tbps per direction of the stated 25.6 Tbps) Table 4: "Switching capacity 25.6 Tbps", "Forwarding rate 19050 Mpps"
PoE none, no budget Table 1 declares no PoE; the 32C is a QSFP28 card

Port names follow FS's own NC8400 CLI reference guide, whose examples are interface HundredGigabitEthernet 1/1 and int hundredGigabitEthernet 2/49; the 4-slot frame numbers its bays 1-4. Slot 1 ports 1-32 -> HundredGigabitEthernet1/1 ... HundredGigabitEthernet4/32.

Sources

Discrepancies, stated rather than smoothed over

  1. 25.6 Tbps is duplex. 128x 100G is 12.8 Tbps each way, so fabric.capacityMbps is 12,800,000 - the same per-direction convention NC8200-4TD uses, because the fabric audit compares the figure against the sum of the ports. A duplex 25,600,000 would have claimed a full 32C load was half empty.
  2. Forwarding rate changes units between FS's two data sheets: the older prints "19.05 Bpps", the current one "19050 Mpps". Same number, so no contradiction - but 19.05 Bpps would have been 19050 Mpps only if "B" means billion, which the current sheet says outright.
  3. The current PDF is headed "FS NC8400 SERIES SWITCHES DATA SHEET" on its cover and "FS NC8200 SERIES SWITCHES DATA SHEET" on several interior page headers - a copy-paste artefact in FS's document. Every table under those headers carries NC8400 P/Ns (NC8400-4TH / -32C / -16CD), so the tables are NC8400 tables.
  4. Breakout is documented twice, inconsistently, and is not modelled. The port note says "QSFP28 can be used for 40G/100G or 4x 10G/4x 25G connection", while Table 2 reports no 40G/100G port density with breakout cable for the 32C, and the breakout section says ports 1-4 are one group of which only one supports breakout - "the other three ports in the same port group will be unavailable". Breakout therefore costs ports, which this spec's flat port list cannot express, so speedCaps is 40G/100G only.
  5. IEEE 802.3af appears in the L2 protocol list of the older data sheet, but no port can source power and no budget is stated, so no poe-pse and no wattage were added.

Not modelled, deliberately

  • The 4x NC8400-16CD fill (64x 40/100G + 16x 400G QSFP-DD, per Table 2 and the product page's "64x 40/100G with 16x 400G"): a different box with a different port census, and it needs its own spec rather than a second model name here. Mixed fills are anywhere between the two.
  • The management port (out-of-band, no rate stated) and the USB 2.0 port (not a network port).

Gates

  • node scripts/syntax-check.js - 358 files, all OK
  • node scripts/audit-port-types.js - exit 0, "No port contradicts its declared type"; the 128 new QSFP28 ports are declared with a connector, so they add nothing to the undeclared-hardware lists
  • node scripts/audit-devices.js - "No hard defects"
  • test/hardware-invariants.test.js 26/26, test/fabric-agreement.test.js 15/15, test/faceplate-catalogue.test.js 14/14, test/catalogue-hygiene.test.js 25/25, test/import-devices-capabilities.test.js 34/34, test/doc-catalog-counts.test.js 43/43
  • node scripts/regen-registry-doc.js - 2,328 -> 2,329 models, 947 -> 948 port shapes (DEVICE_REGISTRY, CONNECTORS, README, wiki Home, build-wiki template)
Fixes #757 The 2026-11-10 pass left this open because FS publishes a *slot count* and a *module set* (a bay, not a port). Option (a) in that comment was "you name the module set for the SKU you want" - and FS's own data sheet names it for us: "The NC8400-4TH switch can be loaded with line card NC8400-32C and supports a maximum of 128x 40G/100G high-density full line rate ports." So the chassis is modelled as that fill, exactly the way `NC8200-4TD` is modelled as 4x `NC8200-24BC` (#1690), and the label says which fill it is: **NC8400-4TH (4x NC8400-32C)**. ## What the spec says | field | value | source | |---|---|---| | line-card slots | 4 | data sheet Table 1 | | ports | 128x `HundredGigabitEthernet<slot>/<port>`, QSFP28, 100G (40G capable) | Table 1: "100G port density 128", "40G port density 128"; Table 2: 32C = 32x 40G / 32x 100G | | console | 1 (`type: console`) | Table 1: "Console port 1" | | fabric | 12,800,000 Mbps (12.8 Tbps per direction of the stated 25.6 Tbps) | Table 4: "Switching capacity 25.6 Tbps", "Forwarding rate 19050 Mpps" | | PoE | none, no budget | Table 1 declares no PoE; the 32C is a QSFP28 card | Port names follow FS's own NC8400 CLI reference guide, whose examples are `interface HundredGigabitEthernet 1/1` and `int hundredGigabitEthernet 2/49`; the 4-slot frame numbers its bays 1-4. Slot 1 ports 1-32 -> `HundredGigabitEthernet1/1` ... `HundredGigabitEthernet4/32`. ## Sources - https://resource.fs.com/mall/resource/nc8400-series-switches-data-sheet.pdf (Tables 1-4, line cards, breakout restrictions) - https://img-en.fs.com/file/datasheet/nc8400-series-switch-datasheet.pdf (the earlier NC8400 data sheet - agrees on 4 slots, 128x 40G/100G, 1 management / 1 console / 1 USB, 25.6 Tbps, <1950 W) - https://img-en.fs.com/file/user_manual/nc8400-series-switches-cli-reference-guide.pdf (interface naming) - https://www.fs.com/products/110486.html (product page; line-card SKUs and the 400G fill) ## Discrepancies, stated rather than smoothed over 1. **25.6 Tbps is duplex.** 128x 100G is 12.8 Tbps each way, so `fabric.capacityMbps` is 12,800,000 - the same per-direction convention `NC8200-4TD` uses, because the fabric audit compares the figure against the sum of the ports. A duplex 25,600,000 would have claimed a full 32C load was half empty. 2. **Forwarding rate changes units between FS's two data sheets**: the older prints "19.05 Bpps", the current one "19050 Mpps". Same number, so no contradiction - but 19.05 *B*pps would have been 19050 Mpps only if "B" means billion, which the current sheet says outright. 3. **The current PDF is headed "FS NC8400 SERIES SWITCHES DATA SHEET" on its cover and "FS NC8200 SERIES SWITCHES DATA SHEET" on several interior page headers** - a copy-paste artefact in FS's document. Every table under those headers carries NC8400 P/Ns (NC8400-4TH / -32C / -16CD), so the tables are NC8400 tables. 4. **Breakout is documented twice, inconsistently, and is not modelled.** The port note says "QSFP28 can be used for 40G/100G or 4x 10G/4x 25G connection", while Table 2 reports **no** 40G/100G port density with breakout cable for the 32C, and the breakout section says ports 1-4 are one group of which only one supports breakout - "the other three ports in the same port group will be unavailable". Breakout therefore costs ports, which this spec's flat port list cannot express, so `speedCaps` is 40G/100G only. 5. **IEEE 802.3af appears in the L2 protocol list** of the older data sheet, but no port can source power and no budget is stated, so no `poe-pse` and no wattage were added. ## Not modelled, deliberately - **The 4x NC8400-16CD fill** (64x 40/100G + 16x 400G QSFP-DD, per Table 2 and the product page's "64x 40/100G with 16x 400G"): a different box with a different port census, and it needs its own spec rather than a second model name here. Mixed fills are anywhere between the two. - **The management port** (out-of-band, no rate stated) and the **USB 2.0** port (not a network port). ## Gates - `node scripts/syntax-check.js` - 358 files, all OK - `node scripts/audit-port-types.js` - exit 0, "No port contradicts its declared type"; the 128 new QSFP28 ports are declared with a connector, so they add nothing to the undeclared-hardware lists - `node scripts/audit-devices.js` - "No hard defects" - `test/hardware-invariants.test.js` 26/26, `test/fabric-agreement.test.js` 15/15, `test/faceplate-catalogue.test.js` 14/14, `test/catalogue-hygiene.test.js` 25/25, `test/import-devices-capabilities.test.js` 34/34, `test/doc-catalog-counts.test.js` 43/43 - `node scripts/regen-registry-doc.js` - 2,328 -> **2,329 models**, 947 -> 948 port shapes (DEVICE_REGISTRY, CONNECTORS, README, wiki Home, build-wiki template)
fs: the NC8400-4TH 4-slot chassis as a stated 4x NC8400-32C fill (128x 100G QSFP28)
All checks were successful
CI / Classify the change (pull_request) Successful in 14s
PR rebase / Rebase onto main and arm auto-merge (pull_request) Successful in 18s
CI / Server smoke test (pull_request) Successful in 20s
CI / Docker build (pull_request) Successful in 33s
CI / Test (Node 20) (pull_request) Successful in 1m0s
CI / Test (Node 18) (pull_request) Successful in 1m4s
CI / Test (Node 22) (pull_request) Successful in 1m5s
3c2bfe87de
Fixes #757. The request was blocked on a named module set; FS's own data
sheet leads with the 32C load ("can be loaded with line card NC8400-32C and
supports a maximum of 128x 40G/100G"), so the chassis is modelled as that
fill, the way NC8200-4TD is modelled as 4x NC8200-24BC.

Ports: 128x HundredGigabitEthernet<slot>/<port> QSFP28 at 100G/40G (slot 1-4
per the NC8400 CLI reference guide), one console port. Fabric 12.8 Tbps per
direction of the stated 25.6 Tbps duplex. No PoE. The 16CD fill (400G) is a
different box and needs its own spec.
jaylouisw scheduled this pull request to auto merge when all checks succeed 2026-11-16 19:59:43 +00:00
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jaylouisw/packetforge!1719
No description provided.