Server meshing

Benoit Beausejour says the game is still being turned into an MMO and content and features are outpacing the guardrails and cross-system validation, so the team needs to add stronger validation.

Beausejour says the cause isn't pinned down yet, but some server meshes run below the target frame rate, and the team is tracking six to eight optimizations, including lock contention in the area component, for the hotfix channel.

Beausejour says shard lock is the price of keeping shards fluid instead of a classic MMO setup, that a 4.8 bug stopped some characters from being stowed and has been fixed, and that players should be patient before using character repair.

A copy of every land claim exists on all shards in a region so none can be claimed twice, shop inventories are shared, and raids happen on one dynamically chosen shard where attackers and defenders get transferred.

3.18 was critically unstable, mostly because of a massive backend problem rather than server crashes; the score is skewed by local crash loops, so territory-based scoring is being discussed.

Static server meshing has run for months, with the configuration changed every patch, hybrid performance and binding improvements helping desync, and lessons about load distribution informing dynamic server meshing.

No date is given, but dynamic server meshing work is under way on moving streaming groups between servers, enabling arbitrary subdivision and instancing across the mesh, with higher player counts not a current priority.

Crushing trees with large ships needs unfinished gameplay design plus promotion, entity management and persistence work, and it isn't planned or guaranteed, though the system should support more complex behaviour.

The developer says server meshing will help, because each server simulates fewer elements and performs better, which should speed up development and patch delivery, give more performance buffer, and provide finer control.

Dynamic meshing needs less development than static meshing, which already covered the core technology; the developer hopes it comes quickly, with some dynamic elements already in 4.0, and doesn't expect multi-month development.

The replication layer lets AI write data like players do, while moving AI between servers will require replicating their decision-making context and restoring correct states, such as dead or unconscious, so players don't notice switches.

They take the joke well, explaining that standing on chairs and T-poses come from several different bugs, mostly streaming-related, and deliberately aren't hidden, since they flag problems that need proper fixing.

Jared Huckaby says he didn't cry, and tells how Paul Reindell's visual server meshing demo was challenged to run live, showing the team doesn't take the easy, low-risk route.

Roccucci admits PU AI isn't where he wants it, citing unfinished systems, behaviors awaiting replacement and cross-team dependencies, but says he's proud of the direction and expects improvements.

Persistence stays per server before server meshing, but items stowed in inventory can be carried to another server; no matchmaking changes are planned for this version, though crashed players get a reserved slot for a limited time.

The team is working on server frame rate, synchronization bugs, calculation priorities, and faster queries, mostly without needing server meshing, though heavy NPC counts on one server remain a problem.

Unstowed ships persist in their shard until stowed, but a landed ship that logs out is planned to be auto-stowed if no players are nearby, though one left behind and destroyed could mean waking in a med bed.

No, assets can't cross shards unless stowed in an inventory first, and within a shard, entities like missiles can span server nodes while bullets are spawned separately on each client and node.

Players won't be permanently tied to shards, and a matchmaking system in development with server meshing is planned to use factors like reputation to keep each shard semi-diverse.

Server meshing is described as very advanced, with the prerequisite streaming, authority and replication work largely done, a static mesh next, and a dynamic mesh still to come after that.

Server performance affects AI quite a bit, since calculations are spread over time and slow updates make social reactions, vision and ship aiming sluggish or off; server meshing is expected to help.

They want hackers for hire to clear others' CrimeStats but can't yet, because meshed servers and persistence make editing offline players hard; the new hacking database lets you hack individual crimes.

Better hardware and cloud servers made walkable planets possible and enable "server meshing", where many servers talk to each other so far more players can share one world.

Roberts says the tech for thousands of players in one world is well along, and that cloud-hosted clients and servers would greatly reduce lag and cheating.

Todd Papy doesn't confirm server-side OCS for 3.6, saying such work happens behind the scenes and feature teams switch systems over as they go, with vehicles needing more than mining.

Two systems are planned to run as separate server instances, and they're thinking, undecided, of using server migration to move a ship's server bubble between systems, which first requires building the server meshing tech and solving coordinate system collisions.

Ping is an unbreakable barrier, but they plan to attempt a single shard, possibly using geographically central or layered servers and latency hiding; they have no definite answers yet, and Arena Commander and Star Marine will probably stay region-based.

They can't predict the player cap or fleet-battle size for 3.0 because networking code is still being refactored; they want fleet fights as big as possible, but multicrew syncing is a challenge.

Instancing is nearly possible now, but they are instead pursuing one seamless unified world meshed across servers, with the low-level networking rewrite aimed at 3.0 and the unified model later with no date, and gating for crowded spots.

There is no preset limit on players or ships on a planet; it depends on optimization and performance, and the numbers are expected to keep growing, with networking work hopefully helping.

Chris Roberts says the UK visits focus on foundational engineering, including a new cloud-based back-end for scaling game servers and AI work for Squadron 42 and the Persistent Universe, plus in-person collaboration.

Players are only moved to another server when an area gets too crowded, so scouts can usually go ahead and the fleet join them, with slots kept open for friends, though how oversized fleets are handled is undecided.

Roberts says they want capital ship battles but it's early; recent server work, including 64-bit global entity IDs, enables flexible server handover and multiple servers per client, supporting higher player counts.

There are currently no plans for an offline Star Citizen, as the simulated universe needs servers; a private-server option is planned after launch, and a simpler offline version isn't ruled out later.

A thousand players per server was never promised, with the target being 50 to 100 and 25 currently; landing pads may be instanced so players can usually land when slots exist.

Chris Roberts says Squadron 42 will be playable offline without co-op, and smaller private servers for friends are still planned, though with less scale and slower content updates than the main universe.

Chris says players will make up at most a fraction of the population, instance sizes are limited, and matchmaking plus PvE/PvP choices should stop large groups dominating areas, though organisations should still affect the economy.

They'd like 50 to 100 players per server instance but are still a way off, with optimization and net code work needed; a star system can hold far more players across instances, though no giant single battles are promised.

The PvP slider is one matchmaking input, alongside ping, friends and skill, that biases whether AI or real players stream into your instance; it can't be changed on the fly, and no-holds-barred areas still exist.

Locking down a whole system is hard because of multiple instances, and they don't want that; groups could influence or monopolize a local area, but the exact design is still undecided.