Level Up Your Meetings: What Esports Taught Us About QoS and how to implement.

hanro440
Nov 11
4 min read

Why esports?

In esports, milliseconds decide matches. Teams separate real‑time traffic, prioritise it, and keep buffers short so inputs and comms stay snappy. Your Teams/Zoom traffic has the same needs: low latency, low jitter, and predictable loss. Borrowing the playbook turns a chaotic office network into a disciplined arena where voices get a clear lane.

TL;DR (3 Takeaways)

Prioritise real‑time > bulk. Voice/video should beat file syncs and backups to the front of the queue.
Tiny QoS tweaks = big wins. A few DSCP tags and queues can remove most “you’re breaking up” moments.
Measure jitter, not just speed. Throughput is fine; jitter/MOS tells you if people sound like robots.

1) Map your traffic: real‑time vs bulk

Before touching configs, inventory your traffic.

Real‑time (latency/jitter‑sensitive):

VoIP (SBC/SIP), Teams/Zoom audio & video, Google Meet, Webex
Live screen‑share, interactive RDP/VDI

Bulk/elastic (can wait a bit):

OneDrive/SharePoint/Drive sync, Dropbox
Software updates, EDR/AV definition pulls, backups
Large file transfers (CAD, media), email bursts

How to map quickly

Export top talkers from your firewall for a normal workday.
List SaaS domains and update services (Microsoft/Apple/Adobe) that spike at 8–10am.
Note WAN types per site (fibre, wireless, LTE/5G) and upstream speeds—that’s where contention hurts.

Goal: Know what deserves a VIP lane (voice/video) and what can be politely slowed for a few seconds (sync/updates).

2) Put voice & video on their own VLAN

A voice/video VLAN makes identification and control easy.

Design tips

Separate subnets: e.g. VLAN 30 for VoIP phones, VLAN 40 for conferencing devices/meeting rooms.
DHCP options for IP phones (option 66/150 if you host call control).
mDNS/Bonjour: gate carefully—don’t flood the whole campus; use mDNS gateways for room devices.
Wi‑Fi: create an SSID for meeting devices/boardrooms with WMM enabled; disable legacy rates.

South Africa note: On smaller branches with LTE/5G failover, keep the voice/video VLAN pinned to the primary link if possible. If you must fail over, cap bulk traffic to protect the limited mobile uplink.

3) QoS queues & DSCP basics (the fast lane)

You don’t have to be a Expert. You just need the right tags and three clear queues.

DSCP cheat‑sheet

EF (46) → Voice (audio)
AF41 (34) → Video
AF21 (18) → Screen‑share / data
CS0 (0) → Best effort (most traffic)

Quick‑start for Teams/Zoom

Mark audio = EF (46), video = AF41 (34), screenshare = AF21 (18) at the edge (client or access switch/AP). Many apps can mark, but don’t rely on them—trust your network to remark if missing.
Create at least three queues on WAN routers/APs: Low‑latency (strict priority for EF), Assured (AF41/AF21 with bandwidth guarantees), Best Effort.
On Wi‑Fi, ensure WMM is on, mapping voice/video to higher access categories.

Shaping vs policing

Shape your WAN egress to just under your real upstream (e.g., shape to 95% of 100 Mbps). This prevents provider buffers from bloating.
Police only egregious classes (e.g., bulk updates) to stop them from bursting over the shaper.

4) Beat bufferbloat (router & ISP)

Bufferbloat makes speech stutter even when speed tests look great. Fix the queues where they actually fill.

Steps

Test: Run a latency‑under‑load test (idle vs upload/download). You want minimal additional latency (<30–40 ms) when saturating.
Enable smart queue management (SQM) on your edge (e.g., FQ‑CoDel/Cake). Set egress/ingress a bit below the real line rate.
Ask your ISP about upstream congestion management—some last‑mile links in ZA still have deep buffers. If they can’t help, keep your own shaper conservative.

Tell‑tales of bufferbloat

Calls degrade when someone sends big email attachments or starts a cloud backup.
Uploading to Teams during a meeting causes everyone to sound metallic.

5) Monitor what matters: MOS & jitter

Speed ≠ quality. Watch these:

Jitter (ms): variation in packet arrival. Keep < 30 ms for solid calls; <15 ms is great.
Packet loss (%): keep < 1% sustained; short spikes hurt audio more than video.
One‑way latency (ms): <150 ms is comfortable for conversation.
MOS (Mean Opinion Score): aim ≥ 4.1 for business‑grade; if it dips <3.6, users will complain.

Where to get the numbers

Microsoft Teams admin, Zoom dashboards, SBCs, and many next‑gen firewalls expose per‑call MOS/jitter.
For branches, a cheap synthetic probe (a micro PC or capable router) can send small UDP streams EF/AF41 to your DC or a cloud probe and log jitter/loss.

6) Go-live checklist (copy/paste)

Before

Voice/video VLANs created; DHCP scopes & routes in place
WMM on; meeting-room SSID tested
DSCP marking/remarking verified with a packet capture (EF/AF41/AF21)
WAN shaper set to ~95% of real up/down; SQM (FQ-CoDel/Cake) enabled
Queues built: Priority (EF), Assured (AF41/AF21), Best Effort; bulk/update class rate-limited
ISP tested under load; added latency acceptable

During pilot

Run real Teams/Zoom calls at peak hour across sites
Watch MOS, jitter, loss on live dashboards
Ask users: “Any robot voices? Video freezing?”

After

Lock configs; template for other branches
Alerts: MOS <3.8, jitter >25 ms, loss >1% for >2 min
Document an escalation path (site → ISP → platform)

Troubleshooting quick hits

Good bandwidth, bad calls → suspect bufferbloat or wrong queue mapping.
Great audio, flaky video → AF41 not guaranteed; increase assured bandwidth; check WMM.
Calls die during backups → rate-limit backup job class; stagger schedules.
Branch on LTE/5G → cap bulk to a tiny slice; shape hard; prefer audio-first.