Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/kenzz55/ue5-iocp-mmo-server/llms.txt

Use this file to discover all available pages before exploring further.

MoveLoadTester (ServerSolution/Tools/MoveLoadTester/) is a .NET console application that spawns N concurrent clients, each executing the complete game connection flow: HTTP login and server selection, TCP EnterGameReq, UDP hello handshake, and then a continuous stream of C_MoveReq UDP packets. Warm-up traffic is excluded from the final statistics, and the tool writes a structured JSON result file that the PerformanceReport tool can consume directly.

​
MoveLoadTester flags

All options are parsed by a switch statement in Program.cs. Every flag takes a single value argument.
FlagDefaultDescription
--auth-urlhttp://localhost:5000Base URL of the Auth HTTP service
--clients10Number of concurrent simulated clients
--duration60Measurement duration in seconds (after warm-up)
--warmup0Warm-up period in seconds; this traffic is excluded from RTT stats
--connect-rate0Maximum new TCP connections per second; 0 means connect all clients immediately
--move-hz20Movement input packets per second per client
--echo-hz1TCP echo request packets per second per client
--account-prefixloadAccount name prefix; accounts are named {prefix}{index:D5}
--account-start0First account index
--password12345678Password used for all test accounts
--server-id1Game server ID sent to /auth/select-server
--game-host(from select-server)Override the game server IP returned by the auth service
--game-port(from select-server)Override the game server TCP port
--output(none)Path for the final JSON result file
--series-output(none)Path for the per-second time-series CSV
--run-id(auto-generated)Identifier embedded in the result JSON
--scenario(auto-generated)Scenario label; auto-derived from --move-hz and --echo-hz if omitted

​
Example invocation

dotnet run --project ServerSolution/Tools/MoveLoadTester -- \
  --auth-url http://localhost:5000 \
  --clients 100 \
  --duration 60 \
  --warmup 10 \
  --connect-rate 20 \
  --move-hz 20 \
  --echo-hz 1 \
  --output Performance/run-001.json \
  --series-output Performance/run-001-series.csv

​
What each client does

Each simulated client runs the following sequence in order:
1

Register (best-effort)

POST /auth/register with the account credentials. Errors are silently ignored — the account may already exist from a previous run.
2

Login

POST /auth/login to obtain an accessToken.
3

Select server

POST /auth/select-server with the accessToken and --server-id. The response provides gameServerIp, gameServerPort, enterToken, and characterId.
4

TCP connect

Opens a TCP connection to the game server and sends EnterGameReq (packet ID 0x0201) containing the enterToken and characterId. Waits for EnterGameRes (packet ID 0x0202) which carries a udpToken and UDP port.
5

UDP hello

Sends UdpHelloReq (packet ID 0x0204) over UDP carrying the udpToken, accountId, and characterId. The server binds the UDP endpoint for this session.
6

Move loop

At --move-hz packets per second, sends C_MoveReq (packet ID 0x0211) with a rotating direction vector. The client cycles through four cardinal directions every 120 packets. Latency is measured from the send timestamp to the arrival of the corresponding S_MoveNotify or S_MoveBatchNotify that echoes back lastProcessedInputSequence.
7

Echo loop (TCP)

Concurrently, at --echo-hz packets per second, sends EchoReq (packet ID 0x0001) over TCP. Echo RTT is tracked separately from movement RTT.

​
LoadMetrics and final report

The LoadMetrics class records all events via lock-free atomics and a LatencyHistogram with 0.25 ms buckets capped at 60 000 ms. During the warm-up phase no latency samples are recorded and the measured send/receive counters are not incremented. After the run ends, BuildFinal produces a LoadTestResult record that is written to --output: 
{
  "runId": "load-20240101T120000Z-...",
  "scenario": "move-20hz-echo-1hz",
  "clients": 100,
  "warmupSeconds": 10,
  "durationSeconds": 60,
  "moveHz": 20,
  "echoHz": 1,
  "connected": 100,
  "entered": 100,
  "udpHello": 100,
  "connectFailed": 0,
  "enterFailed": 0,
  "clientFailures": 0,
  "measuredMoveSent": 120000,
  "measuredMoveReceived": 119850,
  "moveRtt": { "samples": 119850, "averageMs": 8.5, "p50Ms": 7.75, "p95Ms": 18.0, "p99Ms": 28.5, "maxMs": 120.0 },
  "echoRtt": { "samples": 5990, "averageMs": 0.5, "p50Ms": 0.5, "p95Ms": 0.75, "p99Ms": 1.0, "maxMs": 4.0 }
}
Real-time progress is printed to the console every second in the format: 
[Load] phase=warmup clients=100 connected=84 entered=80 udpHello=80 moveSendPps=1600.0 moveRecvPps=1590.0 moveRttP99Ms=0.00 ...
[Load] phase=measure clients=100 connected=100 entered=100 udpHello=100 moveSendPps=2000.0 moveRecvPps=1995.0 moveRttP99Ms=22.50 ...

​
PowerShell test scripts

ServerSolution/Tools/ includes lightweight PowerShell scripts for smoke-testing individual flows without running the full load tester. 
# Register, login, list servers, and select-server
# Defaults: -BaseUrl https://localhost:5001 -Account test -Password 12345678 -Nickname Tester -ServerId 1
.\TestAuthFlow.ps1 -BaseUrl http://localhost:5000 -Account myuser -Password mypassword
Use TestAuthFlow.ps1 first to obtain an EnterToken, then pass it to TestEnterGame.ps1 to verify the full TCP game-entry path: 
# Step 1 – get a token (copy enterToken from the output)
.\TestAuthFlow.ps1 -BaseUrl http://localhost:5000

# Step 2 – enter the game with that token
.\TestEnterGame.ps1 -EnterToken "abc123..." -CharacterId 1

​
Load-test spawn spread

When running many clients on a single server instance, all spawned characters would land on the same map position by default, placing every player in every other player’s Area of Interest and generating worst-case fanout. The flag EnableLoadTestSpawnSpread avoids this: 
constexpr bool  EnableLoadTestSpawnSpread = true;
constexpr float LoadTestSpawnSpacing      = MovementAoiViewDistance + 1000.0f; // 4000 UU
constexpr int   LoadTestSpawnGridWidth    = 10;
With these defaults, players are placed on a 10 × 10 grid with 4000 Unreal Units between each grid cell. Because MovementAoiViewDistance is 3000 UU, adjacent cells are just outside each other’s AoI, preventing cross-client movement notifications and giving a cleaner measurement of per-client throughput rather than AoI fanout overhead.
Always run load tests against a Release build. The Debug build carries bounds-checking and assertion overhead that can make tick-time measurements several times higher than they would be in production, leading to misleading baseline values.
MoveLoadTester calls POST /auth/register for every client index at startup. All test accounts (load00000, load00001, …) are persisted to the database. Clean them up after testing — either by truncating the accounts table or by using a dedicated test database instance that is discarded after each run.

Build docs developers (and LLMs) love

Get started for freeTalk to us