ARC.

NO. 21ATLASCUSTOMER STORYRIFTBORN GAMES

Riftborn shipped a worldwide launch without a queue.

They rebuilt their realtime netcode on the Arc mesh and took a flagship multiplayer title from a fragile three-region rollout to a single, simultaneous global launch. Players felt the signal — not the seams.

Live wavefront
SIGNAL CAPTURE · LAUNCH WINDOW 2026-03-14
  1. R-01PEAK

    +318%

    concurrent playerspeak at global launch — zero queue formed
  2. R-02p99

    11ms

    p99 net tickdown from 74ms on the legacy regional mesh
  3. R-0390d

    0

    rollback incidentsacross the first 90 launch-window days
  4. R-04T+0

    6wk

    to first matchkickoff to shipping authoritative netcode

The problem

02 / SIGNAL DIAGNOSTIC

p99 net tick
74ms
topology
3 regional brokers
signal view
none

The mesh worked in three regions. It would not survive one launch.

A soft launch is forgiving. A worldwide simultaneous release is not — it meant 10× the concurrency, every region failing in its own dialect, and not one place to watch a player’s input actually travel.

Riftborn’s previous architecture stitched together regional brokers with hand-tuned routing. It held for a soft launch in three territories, but the math broke at scale: a 74ms p99 net tick is enough to turn a clean headshot into a support ticket, and the queue formed the moment two regions crested at once.

Worse, every region had its own failure mode and no shared timeline. There was no single view of how a packet propagated from client input to authoritative tick — so every incident opened the same way: an hour of engineers guessing where the signal stalled before anyone could fix why.

LEGACY SIGNAL PATH — PRE-ARCUNINSTRUMENTED
  1. +0msclient input
  2. +22msregional broker
  3. +?relay (unwatched)
  4. +?relay (unwatched)
  5. 74msauthoritative tick

Two relays added latency no dashboard was watching. The path was a guess, not a readout.

The solution / signal path

One signal axis, predicted forward and reconciled on the crest.

No new regions, no new brokers. We collapsed the patchwork into a single deterministic path, instrumented every hop, and rehearsed the launch until the wavefront held under load on its own.

  1. STAGE 01 — TRACE3 HIDDEN RELAYS

    Mapped the signal path

    We traced every packet from client input to authoritative tick and found three relays adding latency no one was watching. The lattice made the propagation legible.

  2. STAGE 02 — REROUTE1 SIGNAL AXIS

    Re-routed the mesh

    A deterministic rollback layer replaced the patchwork of regional brokers. Player state now travels one diagonal — predict forward, reconcile on the crest, never on the player.

  3. STAGE 03 — SHIP2.4M SESSIONS

    Shipped under load

    We launched into a worldwide simultaneous release with synthetic traffic shadowing real demand. The wavefront held; the queue never formed.

The proof — before / after

The wavefront arrives. The numbers light up with it.

Same title, same launch window, same player demand. One column is the legacy regional mesh on its last good day. The other is the Arc mesh under a worldwide simultaneous release. Watch the signal cross the gutter.

BeforeLegacy regional mesh3-region ceiling · hand-tuned routing
AfterArc meshglobal · single release · predicted forward
p99 net tick74msa headshot turns into a complaint
p99 net tick11msthe player feels the signal, not the seams
peak concurrencyRegional
queues
players parked behind a spinner at launch
peak concurrencyZero
queue
+318% concurrent players, nobody waited
rollback incidents9/ qtrevery incident began with an hour of guessing
rollback incidents0/ 90 ddeterministic reconcile, nothing to roll back
release footprint3 regionsstaggered rollout, hoping the seams held
release footprint1 launchone simultaneous worldwide go-live

Riftborn Games · flagship multiplayer · launch window 2026-03-14 → 2026-06-12 · 2.4M day-one sessions

We watched the wavefront hold under a load we used to fear. Launch night, the dashboards just sat there flat and green — and for the first time in my career, that was the entire point.

Mara VelascoVP Engineering · Riftborn Games
peak concurrency
2.4M
rollbacks
0

By the numbers · end to end

The arc, complete.

One signal, propagated from the first packet to the last region. The figures that survived launch night — measured, not promised.

DAY 01
2.4M
launch-day sessionsserved on day one — peak 412k concurrentsingle global release
SLA
99.99%
match availabilityrolling 90-day window, all regions52 min budget / yr
UNIT COST
−71%
infra spendper concurrent player vs legacy meshat 3.4× the load

SIGNAL PROPAGATED — END TO END2.4M · 99.99% · −71%

Your launch, next

See what the Arc mesh does under your peak.

Bring your concurrency targets and your worst launch-night story. We map your signal path on a live mesh and show you the wavefront under synthetic peak — before you commit a single region.

Session
45 min
Live mesh
your load
Deliverable
signal map
Map my signal pathSee how it works

No deck. An SRE on the call. p99 figures, not promises.