THE GAME SERVER EXPERIMENT

Connection test results

Review saved tests and compare connections over time.

CLIENT-MEASURED PERFORMANCEReal packets · Saved runs · Repeatable profiles

— saved runsD1 Durable experiment history

ONE TEST · THREE WORKLOADS

Inspect a measured stage

Choose one workload to inspect and compare consistently across saved tests.

Result stage

02 / THE MEASUREMENT

PAIRED DATA CHANNELS

Your browserReal client network

ADirect internet

BCloudflare Spectrum

Game originOne game origin

Ready for an experiment

Start a browser test to collect a measurement.

—REPLIES / PATH

THE APPLICATION WORKLOAD

Game actions

PLAYER STATE UPDATE

The client sends player-state commands. The origin applies accepted mutations to authoritative in-memory state and returns tick-paced acknowledgments. An acknowledgment does not represent a durable database transaction.

DIRECT · ACKNOWLEDGMENT P95—ms

No command measurements yet

SPECTRUM · ACKNOWLEDGMENT P95—ms

Application acknowledgments

Command metric	Direct	Spectrum

Success counts applied and duplicate acknowledgments. ACK latency includes all received outcomes; stale, rejected, and timed-out commands remain separate. Supporting message timing appears below.

THE APPLICATION WORKLOAD

100 KB object delivery

CHUNKED UDP

All transfer chunks are required for a complete object.

DIRECT · COMPLETION P95—ms

No object measurements yet

SPECTRUM · COMPLETION P95—ms

All chunks are required

Object metric	Direct	Spectrum

Object completion is an application workload measurement. Transfer-chunk arrival timing appears below as supporting evidence.

DIRECT / P95

—msWaiting for a measurement

SECOND PATH / P95

—msSame client, same time window

P95 DIFFERENCE

—msDescriptive, direct minus second path

03 / OBSERVE

Round-trip latency

DirectSecond path

Let the network tell its story.

Launch an experiment or open a saved run.

Application delay · Measured in the client · No measured data yet

04 / INSPECT

Path performance

NO RUN SELECTED

Metric	Direct	Second path

Application RTT, loss, jitter, and deadlines describe this client and time window. Repeat across client networks and times before drawing a broader conclusion.

BROWSER MEASUREMENTS

Data channel traffic

WEBRTC DATA CHANNEL

Recorded field	Direct	Cloudflare

Actual data-channel application bytes and messages, including this stage’s warm-up. These counters exclude IP, UDP, DTLS, and SCTP overhead. WebRTC connectivity-check RTT is separate from the game-message timings.

PRIVATE NETWORK PROVENANCE

Client addresses & locations

RECORDED WITH THE TEST

Browser initiator · HTTPS

Historical client · HTTPS

Origin-observed UDP peersSelected stage

Recorded field	Direct	Spectrum

Browser and native-probe locations describe their HTTPS requests. UDP peer metadata comes from the origin socket or a PROXY v2 header, with header trust shown per path. GeoIP enriches socket peers or trusted headers only when their client address exactly matches a recorded browser or native-probe HTTPS IP. The matching source is named per path.

AT THE ORIGIN

Server health & delivery

ORIGIN-WIDE

Origin-wide aggregate. Includes warm-up and other clients; these values cannot be attributed to either route.

Origin telemetry will appear when a run includes server snapshots.

Server telemetry complements the client measurements.

05 / LEARN ACROSS RUNS

The experiment record

BROWSER · WEBRTC DATA CHANNEL

Select for comparison	Experiment	Client / started	Direct p95	Second path p95	Status	Open result

Completed measurements will appear here. Results are saved in Cloudflare D1.

OVER TIME

P95 latency across runs

DirectSecond path

Your next run adds another data point.

At least two matching runs are needed for a trend.

Respects the profile and client filters above · Each point is one saved experiment

SIDE BY SIDE

Compare selected measurements

0 SELECTED

Select two or more saved tests to compare the selected stage. Legacy runs appear only when they match this workload.

Supporting UDP packet metrics describe the labeled client and profile. Baseline repeats test the direct route twice. Throughput is UDP payload at the configured workload rate; useful goodput excludes 88 protocol bytes per reply. These measurements do not test maximum link capacity. Scroll horizontally for delivery and server timing metrics.

Experiment / route	p50 RTT	p95 RTT	p99 RTT	Jitter	Loss	Deadline miss	Offered Kbps	Received Kbps	Goodput Kbps	Send pps	Receive pps	Loss bursts / max	Reply gap p95 / max	Server process p95	Network estimate p95