
BeanheimNet
Valheim 1.0 netcode fixes with telemetry: RTT-sized send window, Steam rate, fair priority, fauna throttle, deadbands, item merge/report. Server + clients.BeanheimNet
Valheim 1.0.x netcode improvements for the Beanheim dedicated server and its clients, with the telemetry to prove (or disprove) each one. One DLL, runs on the server and on clients. No ServerSync, no version lock: a client without it connects and gets vanilla behaviour, because nothing here changes the wire format.
The research behind it (decompiled 1.0.15 server assembly, prior-art survey, corrections to
earlier assumptions) is in the profile repo at docs/networking.md.
What vanilla does, in one paragraph
Every ZDO update is sent reliable on one Steam lane. ZDOMan.SendZDOs keeps at most 10 KB in
flight per peer (unacked bytes count), so throughput is 10 KB / round-trip-time — 68 KB/s at
150 ms — and Steam's send rate is pinned at 150 KB/s. The same limit applies to each client's
uploads. Send order puts every building piece ahead of every creature, cart and drop. Owners write
position on any exact change, so a bobbing or jittering object talks every frame, and ~95 % of
relayed updates on this server were ambient fauna.
Features
Everything that is invisible to players ships on: F1, F2, F3, F4, F5, F6 and telemetry. The two
that change what players see ship off: F8 Cleanup (deletes far-off floating junk) and F9
ItemMerge (merges nearby piles). Nothing here adds a version lock, so a player without the mod
still connects; they simply keep vanilla behaviour on their side.
Toggles
| Section | What it does | Where |
|---|---|---|
F1 SendWindow |
Window per peer = RTT × target rate × BDP factor, clamped [MinWindowBytes, 65536]; falls back to vanilla when no reading exists. MinWindowBytes matters for LAN peers: a 0 ms ping gets no RTT sizing, and at this server's ~10 send rounds/s the vanilla 10 KB window is a ~100 KB/s cap on its own. Queue-drain compatibility for ServerSync/Jotunn-style waiters; raises Jotunn's MaximumSendQueueSize while on. Transpiler with an IL shape check that refuses if SendZDOs changed. |
both |
F2 Scheduler |
Serve every connected peer each interval under a frame budget instead of one peer per frame with an idle arming frame. The dedicated server runs a hard 30 fps, so vanilla's lap is ~10 Hz per peer at two players; ServerTargetFrameRate (0 = vanilla 30) raises the frame cap at ~2× main-thread CPU. |
both |
F3 SteamTransport |
SendRateMin and SendRateMax to SendRateKBps (Steam requires them equal), globally and on live connections, read back. Enable FIRST; without it F1 only builds queue. |
both |
F4 RelayThrottle |
Per class (fish / creature / item) and distance band, cap how often Default-type objects are relayed. Distance is measured from the receiver on the server and from the nearest loaded player on the client uplink, so nothing anyone is looking at is slowed; defaults (near/far/beyond, Hz): fish 6/3/2, creature ∞/10/5, item ∞/5/3, bands at 40 m and 100 m. Receivers extrapolate motion between updates. Nothing is dropped; exempt: never-sent, owner changes, force-sends, players/ships/carts, non-Default types. | both |
F5 Priority |
Age-fair send order below the foreign-owned-Prioritized and Terrain tiers; live x/z reference position from the character ZDO. | server |
F6 Deadband |
Owner-side: skip position/rotation/velocity writes inside a deadband of the last written value; one exact zero velocity when an object comes to rest. Never players/ships/carts/Prioritized. | clients |
F7 Telemetry |
Per peer every 10 s: Steam RTT vs app RTT (the gap is bufferbloat), window, queue, in-flight, rounds, refused, candidates, sent, KB/s, deferred, throttled, ms/call, top prefabs; receive tally; boot census of every ZDO by prefab/type. Also appended to BepInEx/BeanheimNet-stats.log, which survives restarts. |
both |
F8 Items |
Server: periodic ItemDrop report (by prefab, age, in water, far from players). Cleanup=true deletes items older than MaxAgeHours of played time, >ProtectRadiusMeters from every player, not near a base piece, (only in water by default), not protected/piece. Hard rule no setting relaxes: an item with custom data (every EpicLoot enchantment), a quality above 1, or a max stack of 1 (all gear) is never deleted. Every deletion is logged. |
server |
F9 ItemMerge |
Client side: merge same-name, same-quality, same-variant plain stackable drops within MergeRadiusMeters (5) into one stack, every MergeIntervalSeconds (30). Vanilla's own auto-stack does this at 4 m, once per item, only while 200+ drops are loaded. Same hard exclusions as F8: enchanted, quality, variant or gear items are never merged. |
clients |
Console: bnet stats | census | items | cleanup dryrun|run | reload | config (in-game console; the
dedicated server binary never reads stdin, so on the server use the cfg instead: set one of
[10. Triggers] Census / ItemReport / ItemCleanupOnce / Stats = true, it runs on the next reload
(within 5 s) and resets itself to false).
Build and deploy
dotnet build BeanheimNet/BeanheimNet.csproj -c Release # build
dotnet build BeanheimNet/BeanheimNet.csproj -c Release -p:DeployToProfile=true # + copy into the Gale profile
References resolve from the real game and Gale's BepInEx install; no binaries are committed.
The profile's tools/vsync.sh diff then shows + BeanheimNet.dll; the server owner stops the
server, runs apply, checks the RE-VERIFY block is all-zero, and restarts. Friends receive the
DLL through the normal Gale profile sync.
Reading the numbers
deferred= candidates that did not fit the window this round. Sustained non-zero at 2+ peers means the window is the wall. At ≤ 1 peer it is structurally 0 (nothing to relay).refused= rounds that sent nothing because the queue was above the window. Highrefusedwith highrttis the 10 KB/RTT limit.appRTT ≫rtt(Steam) = data is queueing behind other data (bufferbloat). Raise F3 before F1.throttled= F4 removals this interval (deferred by rate, not lost).frame avg 33.3mson the server = vanilla's 30 fps cap;sched frames/served/budget-breaksappear once F2 is on.steam-status ok N not-ok N last <EResult>appears when Steam's connection-status call fails; without it F1 stays at vanilla for that peer and F3 cannot re-apply to live connections.- Vanilla's own
Connections N ZDOS:X sent:Y recv:Zline stays as the cross-check; both counters are meaningless at ≤ 1 peer.
Measuring a change
To attribute a change, turn a toggle off in the cfg (live; no restart), play ≥ 10 min, compare the
stats lines, turn it back on. The vanilla baseline for Beanheim was recorded on 2026-09-21 (see the
profile's docs/networking.md): a LAN client pinned at ~130 KB/s upload with thousands of updates
deferred per 10 s, and a base arrival relayed at 40–60 KB/s with 160,000+ deferred.
Rollback
[1. General] Enabled = false (live) makes every patch behave like vanilla on its next call. Full
removal: delete BepInEx/plugins/BeanheimNet/ locally, vsync.sh diff shows - BeanheimNet.dll,
stop, apply, restart. Nothing persists in the world except F8 deletions, and only when
Cleanup = true.
Fail-soft rules
Every private vanilla member is bound once at startup (Bindings.cs); a missing member disables
the features that need it with a logged reason. Every patch body is wrapped: an exception logs once
per minute and that call falls back to vanilla. The plugin refuses to install anything but
telemetry if Smoothbrain Network, ValheimTune, NetworkPerformanceSystem, ReturnToSender,
BetterNetworking or VBNetTweaks is loaded, and lists every other Harmony owner of the methods it
touches at boot ([owners] lines).

