showstack

GitHub
Free forever·no account, no tracking, no third-party requests ·open data, free API
showstack / network

Converged network planner

Several protocols on one switch fabric, each with its own idea of priority. Tick what runs on your network: the plan below shows which queue each traffic class belongs in, names the collisions before they cost you a show, and adds the bandwidth up against real link sizes. Every value carries its source.

QoS priority plan

Queue shape assumes a managed switch with four egress queues and strict priority on the top one, which is how AV-preset switches (Netgear M4250 AV profiles, Luminex GigaCore presets) ship. DSCP sources: Audinate, Q-SYS.

Will it fit the link?

Planning ceiling is 75% sustained: the rest is burst headroom, and NDI in particular is all bursts. Figures: Dante flows per Glensound's Dante note (a flow carries 4 channels and costs the same even part-filled), sACN worst case with ETC's measured typical far lower, NDI per the NDI white paper, Q-LAN per QSC's Q-LAN notes.

Why converged networks fall over

Clock starvation. PTP is tiny but timing-critical. If heavy media shares its queue (the Dante-audio vs Q-LAN-clock collision above), clock packets arrive late, devices drift, and audio pops or mutes "randomly" under load. Strict-priority the clock queue and keep it clean.
Multicast flooding. sACN and Dante multicast need IGMP snooping with an active querier. Without one, every multicast packet hits every port, and the moment universes or flows scale up the whole switch chokes. One querier per VLAN, snooping on, per Audinate's network admin guide.
Energy Efficient Ethernet. EEE (802.3az, "green Ethernet") powers the link down between packets and adds wake latency that wrecks PTP. Audinate and QSC both say the same thing: turn it off on AV ports.
Microbursts. NDI and file traffic arrive as bursts that fill shallow switch buffers even when average load looks fine. That is why the plan above puts them in best effort with bandwidth headroom instead of priority.
AVB is different on purpose. Milan/AVB does not trust DSCP at all: the switch reserves bandwidth per stream (SRP), which is why every switch in an AVB path must support it, and why a Milan rig will not cross a non-AVB core.

Deeper reading lives on the protocol pages: Dante, Q-LAN, AES67, AVB/Milan, sACN, PTP. Capture and check with the monitoring tools in the index.

Running a stack this page gets wrong?

Say so - every DSCP value and bandwidth figure here is data with a source, and correcting one fixes the planner for everyone.