ClientServer Postmortem

1
Client/Server Postmortem

• Mark Brockington Scott Greig

2
45 Programmer Years In 45 Minutes

• User-Created Content
• Multiplayer
• Multiplayer vs. User-Created Content
• Neverwinter Nights Toolset
• Team Organization
• Artificial Intelligence
• Conclusions Questions

3
User-Created Content

• Why We Did it?
• DD nostalgia
• Creation is one of DDs four fundamentals
• Tap into Creation Desires
• Even I could do better
• Creation is fun
• Communities Keep Games Alive

4
User-Created Content

• They would if they could
• Most content is created by 1 of 1
• Barriers to Entry
• No instant gratification
• No Art Skills
• No Programming Skills
• Only one person
• Our goal was to reach to the 1s

5
User-Created Content

• What we did
• The tile painting system
• Easy to use
• Fun to use
• Limited the quality
• Hard to add custom content

6
User-Created Content

• What we did
• Technical constraints
• No forced downloads for client
• All custom module data sent at run time
• Limited customization on any client data
• No preprocessed shadow solution
• Custom content cant break the install
• Isolated custom content into modules
• Isolated hacking and custom art into HakPaks

7
User-Created Content

• What we learned
• Supporting user created content pays off
• You still need to support both the 1ers and the
  1 of 1ers
• The easier you make it use, the harder it is to
  add custom content
• To make content creation accessible, you will
  have to make sacrifices

8
Multiplayer

• Baldurs Gate ignored bandwidth as a valuable
  resource.
• Our Goal For NWN A secure server running 8
  players on a 56k modem.
• Sending 100K to one client would take 3 minutes.

9
Multiplayer450 bytes/second?

• Tiles create low-bandwidth areas.
• 10K for the largest area to download
• At most 10 bytes per tile, including all lighting
  information.
• Up to three lights in each tile and one animation
  per tile fit within 10 bytes
• Used zlib to force sounds and map pin info into
  the area message.

10
MultiplayerUpdating Objects

• Adding/Deleting/Modifying visible objects in area
  was carefully metered.
• Follows the Tribes network model
• Server has view of each clients objects
• Maintains a priority list of objects to update
• Fills message with differences between that
  clients view and the servers view.
• Never re-send the same data.

11
MultiplayerClient/Server Model

• Client is treated as insecure.
• All input from the client are requests to do
  things that must be validated by the server.
• Never tell the client what it doesnt need to
  know.
• Spend extra effort validating commands for
  super-users (Dungeon Masters and Server
  Administrators)

12
MultiplayerClient/Server Model

• What Went Wrong?
• Design from 30 2-hour missions to 3 20-hour
  chapters.
• I can run a persistent world on my machine!
• Overzealous bandwidth reduction requires lots of
  RAM
• Increased our minimum specification a LOT!

13
Feature Checklist

• Every feature had to meet all of the following
  criteria
• Does it improve the players fun?
• Is it faithful to the Dungeons Dragons rules?
• Can it be implemented with little bandwidth and
  not break the server?
• Does this feature break user-designed content?

14
Custom Contentversus Multiplayer

• Characters are the only resource a client adds to
  a server.
• How do I verify your player character is not
  hacked?
• If you have access to your character, we cant
  prove anything.
• Server versus local vaults.

15
Custom Contentversus Multiplayer

• Items contain executable code (The Sims)
• Importing executable code (scripts) violates
  every rule of server security.
• No scripts are allowed to be imported on to a
  server (characters or their items).
• How to implement custom item properties?
• Modules implement them, and the custom abilities
  dont transfer.

16
Custom Contentversus Multiplayer

• User community evolution took us by surprise.
• Each server has its own unique HAK pack.
• Leaving game to install HAK packs is the 1
  reason amongst our users for not playing
  multiplayer!
• Were fixing this in an upcoming release.

17
Neverwinter Toolset

• The good stuff
• Easy to use
• Not quite even your grandma can use
• Basic functionality within the reach of novice
  users
• Made our development faster
• Reduced training time
• Made good use of nested levels of difficulty
• Shared Source Code with the game

18
Neverwinter Toolset

• The bad stuff
• Seven times the programmers
• Incompatibilities between development
  environments
• Was developed during the content creation phase
• Out of order implementation
• Posting of fixes without adequate testing
• Required several GUI rewrites

19
Neverwinter Toolset

• The surprising stuff
• Focus Testing Results
• The expected barriers were there but
• Too many basic concepts were assumed
• What to do when they just dont get it
• Not everyone thinks like a developer
• We were required to add step by step wizards
• We were required to post tutorials
• The Neverwinter Toolset is great for evaluating
  designer applicants

20
Team Organization

• Our most programmer heavy project
• Baldurs Gate only required seven

21
Team Organization

• Neverwinter required 25 programmers
• The Rule Of Four

22
Team Organization

• Dividing between Client and Server
• Seemed like a good idea
• Promoted good networking
• Caused constant delays and communication overhead
• Dividing between Game and Tools
• Worked out well
• The Generic File Format

23
Artificial Intelligence

• Wasnt a top priority.
• Overengineered with two rationales
• Wouldnt this be cool?
• What does the end user need?

24
Reputation System

• Based on a matrix of faction to faction
  reputation values.
• I have to enter how much data?
• Where did the orcs go?
• Did the toolset make them disappear?
• Did the game destroy them?

• The Deer Ate Them!

25
Pathfinding

• All pathfinding run from the server.
• Hierarchical pathfinding.
• Inter-tile pathfinding used to plan path between
  tiles.
• Intra-tile pathfinding used to resolve how to
  traverse each tile.

26
Pathfinding
27
Pathfinding 31 Flavours
28
Why Did The Pathfinding Suck? (Part 1)

• We cant pre-compute mesh around placeable
  objects!
• itll cost too much in bandwidth
• they can be spawned in dynamically.
• Up-close pathfinding versus long range
  pathfinding.

29
Why Did The Pathfinding Suck?(Part 2)

• Bumping your followers into unwalkable locations.
• Re-entrant pathfinding.

30
Conclusions

• In June 2002, we feel we accomplished our goals
• A true multiplayer DD experience for both
  Dungeon Masters and players, with a solid
  single-player game
• 2100 user-created adventures posted on
  Neverwinter Vault (nwvault.ign.com)
• 25000 players enjoying multiplayer each weekend.
• 850,000 user-created accounts on our community
  site.

31
Any Questions?

• Mark Brockington markb_at_bioware.com
• Scott Greig - scott_at_bioware.com
• For more information on Neverwinter Nights, see
  http//www.neverwinternights.com/