1. Nov 27, Fall 2006IAT 4101 Networks Network Protocols Peer-to-peer Client-Server Configurations Trust

2. Nov 27, Fall 2006IAT 4102 Networks  Required for multiplayer games  3 Standard technologies –Modems –Ethernet –Internet

3. Nov 27, Fall 2006IAT 4103 Internet  The greatest thing since sliced bread  The savior of humanity  Will increase freedom and democracy –Around the world –In your neighborhood

4. Nov 27, Fall 2006IAT 4104 Internet User Protocols  TCP –Connection –Reliable –Bytes arrive in order they were sent –Collects small packets and transmits them together –Stream of bytes  UDP –Connectionless –Unreliable –Arbitrary arrival order

5. Nov 27, Fall 2006IAT 4105 TCP  Reliable stream of bytes –Implies the need for a “connection”  Connection sets up data structures –Hold incoming packets –Hold outgoing packets –Handle retransmits

6. Nov 27, Fall 2006IAT 4106 TCP Reliability  Each packet does Send-Receive- Acknowledge  Sender holds sent packet until ACK is received  Sender retransmits if ACK takes too long Sender Receive Receiver Send Acknowledge

7. Nov 27, Fall 2006IAT 4107 TCP  One Send-Receive-Ack takes time  Overlay Sends and Acks  Maintain a queue in sender and receiver 0 1 2 3 0 1 2 3 Ack 0 1 2 3 Send Sender Receiver

8. Nov 27, Fall 2006IAT 4108 TCP Circular Queue -- Sender  Sends data and Puts it in send queue  Sets timer on this queue item  If timer expires, and no ACK, re-send data –Set another, longer timer –Exponentially increasing time  When ACK received –If this queue slot is the oldest, Free the slot for new data  If no queue space avail, sender app waits!

9. Nov 27, Fall 2006IAT 4109 TCP Receive Queue  Receiver maintains a queue the same size as the sender’s  When a packet arrives, send ACK  If the packet is next in sequence –Give it to application  Else Keep it in queue –Another, earlier packet is on its way

10. Nov 27, Fall 2006IAT 41010 TCP  If no ACKS arrive for a long enough time –Temporarily gives up –Sends test packets  When test packets get through –Starts slow, builds up

11. Nov 27, Fall 2006IAT 41011 TCP Wrap-up  Connection sets up sequencing and queues –Reliable arrival: Retransmit –Reliable order:Sequence numbers  TCP bunches up data on 200ms intervals –Minimizes overhead for small chunks of data –This option can be turned off  TCP Has an “emergency” channel –OOB Out Of Band

12. Nov 27, Fall 2006IAT 41012 UDP  Connectionless! –No underlying data to maintain  Unreliable transmission –If you lose a packet, it’s gone –Network software must handle this  Out-of-order arrival –Network software must handle that, too!  Fast –When the port gets the data, the app gets it

13. Nov 27, Fall 2006IAT 41013 UDP  Packets will drop! –1 in 5 over non-local connection  Have to do your own re-send  Some packets are time sensitive –Care little about the past ship location  No header compression –May end up with greater overhead than TCP with PPP

14. Nov 27, Fall 2006IAT 41014 Game Architectures  Peer-to-peer  Client/Server –One server per game  Floating server –One client is also a server  Distributed server –Multiple servers for large world

15. Nov 27, Fall 2006IAT 41015 Peer-to-Peer  Simple version: Lockstep –eg. Doom –Each client transmits to other –Wait for everyone to get data –Proceed to next step

16. Nov 27, Fall 2006IAT 41016 Peer-to-Peer  Advantages –Simple –Nobody has to provide a server Including the Game’s authors! –Good for turn-based games with low bandwidth –TCP  Disadvantages –Frame rate is that of Slowest machine Worst connection –Hackable –Not good for real-time games

17. Nov 27, Fall 2006IAT 41017 Client/Server  Server per game –MUDs, Fireteam, NetTrek –Someone must provide server ($$$) Possibly the game’s authors –Less hackable –Single point of failure –Server must be big & well-connected

18. Nov 27, Fall 2006IAT 41018 Floating Server  Peer-to-peer  Server resolves the action  One peer is the server –Unreal One player elects to be the server –X-Wing vs Tie-Fighter: First player to enter session –Starcraft Player with the CD

19. Nov 27, Fall 2006IAT 41019 Multiple Server  Many machines coordinate service –Ultima Online, Everquest, AOL  Used for large virtual worlds  Everquest –One server per game-geographic region –Freeze on handoff affects game play

20. Nov 27, Fall 2006IAT 41020 What Data to Send?  Sending entire world state is usually too much  Can send just user actions –Simulation engine does the same thing at each client –Pseudo-random numbers from same seed

21. Nov 27, Fall 2006IAT 41021 Sending User Actions-- Problems  Any error in engine –Divergence in worlds –Small error can lead to big divergence  X-Wing vs Tie Fighter –Created a resynchronize protocol –Causes jumps Wrote smoothing algorithm for resynchs  Sim City 2000 Network Edition –Send checksums for world state each turn

22. Nov 27, Fall 2006IAT 41022 Prediction  Eg. Unreal  Waiting for user inputs is too slow  Client does prediction –Motion prediction  Server corrects things if client is wrong

23. Nov 27, Fall 2006IAT 41023 Prediction: Dead Reckoning  Eg. SIMNET (US Army Tank Simulator)  Each vehicle simulates own tank  Sends data every 5 seconds, updating –Position, Speed, Acceleration –Expected path –Prediction violation criteria  Receiver simulates own tank –AND simulates local copy of other tanks

24. Nov 27, Fall 2006IAT 41024 Dead Recokoning  Receiver gets latest 5-second update  Updates own copy of other tanks  Predicts other tanks –Using prediction data –Until new data arrives  Each simulator also sends update –When own prediction violates own criteria  Assumes latencies < 500ms

25. Nov 27, Fall 2006IAT 41025 Dead Reckoning Sim A A’s Predicted Path B’s Predicted Path Sim B Sim A A’s Predicted Path B’s Predicted Path Sim B B Exceeds prediction: predict again and transmit Transmit new prediction every 5 seconds Predict B Predict A

26. Nov 27, Fall 2006IAT 41026 Dead Reckoning: Requirements  Data structures for other entities  Model of entity behavior –Vehicle speed, acceleration range, turn radius –Responsiveness to commands  Situation parameters –Following a road –Precomputed path (NPCs)

27. Nov 27, Fall 2006IAT 41027 Multiple Copies  Maintain 2 Data sets  Now –Accurate self –Predicted others –“Zero” latency for self  Ground Truth –Accurate everybody –Large latency for almost everybody –200-500ms ago

28. Nov 27, Fall 2006IAT 41028 Latency Issues  When latencies get high –Prediction gets worse and worse  Correcting prediction errors may cause visual jumps –Easy to notice!  If jumps are large enough –Temporarily interpolate between wrong prediction and the new correction

29. Nov 27, Fall 2006IAT 41029 Prediction Interpolation Real Interpolated Response Predicted

30. Nov 27, Fall 2006IAT 41030  Some games may allow distributed ownership –Ballistic simulation –Shooter fires bullet –Intended target receives the simulation  Sports - eg. Tennis –Player A hits ball –Player B gets simulation token –B simulates ball path from A’s racket Token Ownership

31. Nov 27, Fall 2006IAT 41031 Trust  “Never trust the client”  Data on the user’s hard drive is insecure –Diablo utility to modify character data –Wrote patch to prevent hacking Throws out your stuff if there’s a time inconsistency –Daylight savings nuked my stuff!

32. Nov 27, Fall 2006IAT 41032 Trust  Network communications are insecure  NetTrek communications are encrypted  NetTrek also requires “blessed” client –Servers have different policies on requiring a blessed client –Prevents robot players or assistants

33. Nov 27, Fall 2006IAT 41033 Trust -- Checksums  First line of defense: –Checksum of all packets –Include header in checksum! –Stops casual tampering  Hash function –Hard to compute source value from result –MD5

34. Nov 27, Fall 2006IAT 41034 Checksums  Not immune to: –Code disassembly –Packet replay  Packet replay attack: –Capture a legal packet, and re-send it more frequently than allowed –Client can restrict send frequency –Server cannot reject high-frequency packets Internet bunch-ups are source of OK bunch-ups

35. Nov 27, Fall 2006IAT 41035 Combating Replay  Each new packet client sends is different –Add a pseudo-random number to each packet –Not just sequence number! –Client & Server match pseudo-random numbers  Random numbers –Seeds must match! –Dropped packets: include sequence number!

36. Nov 27, Fall 2006IAT 41036 Combating Replay  XOR each packet with a pseudo-random bit pattern –Make sure the bit patterns are in sync! –Based on previous synchronized pseudo- random numbers  Add junk – Confuse length analysis

37. Nov 27, Fall 2006IAT 41037 Reverse Engineering  Remove symbols  Put encryption code in with rest of network stuff  Compute magic numbers: –At runtime –In server  Encrypt from the start!

38. Nov 27, Fall 2006IAT 41038 Lists Of Servers  Denial of service: –Send a packet to server-server saying “I’m a server” –Fake the IP return address with a random IP# –Server-server adds “new server” to list –Server may run out of memory storing hundreds of thousands of fake servers

39. Nov 27, Fall 2006IAT 41039 List of Servers  Require a dialog –Server-list server responds with Password Keepalive interval –Password must be given by attacker at the correct time –Works OK if client is not better connected!