summaryrefslogblamecommitdiff
path: root/docs/architecture.txt
blob: 7cd54d53465b36de629545606282d0358a9ef523 (plain) (tree)
1
2
3
4
5
6
7
8
9
10
11
12











                   


                              
 
 

               




                                                                              



                                                                         
                                                     

                                    





                                                                              
                                                                        




                                                                               



                                                                          
 





                                                                              
 













                                                                               


































                                                                                




                                                                              
                                               
 


                                                                               
 

                                                                              


                                                                               
 
                                                                          


                                                                              






                                                                               
 
 

           













                                                                              






                                                                      

                                                                            

 


                                                                             

                                                                            
 
 

               

                                                                            
























                                                                                
 























                                                                                
-------------------
SERVER ARCHITECTURE
-------------------

1. INTRODUCTION
2. RO REVIEW
3. EATHENA MODEL
4. TMW SERVER
5. TCP - UDP
6. SECURITY
7. DATABASE
8. REGISTRATION
9. SERVER CONTROL
10. OPERATING SYSTEM
11. ADVANCED DATA TRANSMISSION


1. INTRODUCTION

One of the most important thing in a online game is the architecture of the
server system, which reflects in performances (lag and denial of service),
scalability, fault tolerance. In this article we will examine the pre-existing
model and we will evaluate a way to improve it and to add custom features.


2. RO REVIEW

Let's start by taking as reference the current server system used to play
Ragnarok Online on the euRO server. (www.euro-ro.net)
RO works by using 4 kinds of server:

 - Login Server (L): takes care of verifying accounts with username-password
   system, allows also encrypted login.

 - Char Server (C): saves every player status (stats, items, equipment, skills
   and so on.

 - Map Server (M): the real game server, work as interconnection between
   clients, manage chat, monster AI, damage calculations and everything you can
   see in game.

 - Inter Server (I): probably manages the messages between the other type of
   servers.

In euRO there are 1 login server, 1 char server, 1 inter server and 14 map
servers.


3. EATHENA MODEL

The eAthena system mirrors the way used by official RO servers. eAthena
implements 3 servers: login, char and map. It is allowed to have more than one
map server at a time. Every server communicates with all the others.


4. TMW SERVER

The basic idea of TMW server architecture mainly is the same as the one used by
eAthena. Since the login and char server don't have heavy traffic they could be
melt togheter.

        C       M
          \   /
        C - L - M
          /   \
        C       M
        
The login server manages new connections and stores all the informations about
the player. The map server is pretty the same as the eAthena one. The login
server manages also connections to a new map server when changing map. Having 
only one server (L) to manage all the connections between clients and map
servers is a bad point: if the login server crashes players won't be able to
play anymore. In fact new connecting players won't be able to connect to login
server, while map server will disconnect every player, since they can't save
their infos.
The solutions are:
    
    - implementing a distributed login server which can manage crashes and
      redirect new connections to another login server. This way means a more
      complex implementation and probably the need to other computers since we
      want the login servers to be independent from each other crashes at all.
      (Remember this is a free project so we should limit the number of
      computers to act as servers)
      
    - RALS (Redundant Array of Login Servers): we can have two login servers,
      only one of them is active at a time and they share the same database
      where to store infos about players. If one of the map servers loose the
      connection with the login server enables the other, while the crashed one
      restarts. Even if it restarted is it now considered disabled. The new
      active login server will send messages to every map server to inform them.
      Every time a client connects check both of the login server and connect to
      the active one. The bad points of this system are that will imply a lot of
      data consistency checks: map servers should resend last data since it 
      was lost when the login server crashed, the new login server should check
      if some of the data was already stored before the crash, what if both of
      them crash?
      
    - waiting it's the only solution! Let's design the login server as much
      simple and stable as possible and create a smart restarting system.
      This way we will have less frequent crashes and a very low restarting
      time. Obiouvlsy this is the easiest and less expensive solution.    


5. Network protocol: TCP

RO is TCP based, mainly because you use the mouse to move the player. When you
want to reach a point on the map, you click on it and the client send a packet
to the server with destination coordinates. The server replies with an
agreement if there's a path to that way. Using this way you don't need high
speed nor a lot of packets, so TCP it's enough.

With our custom server we want to achieve pixel movements, by that we mean that
the player is not always positioned in the center of the tile, but will have
fine coordinates.

Asking the server if every destination coordinates is walkable means a lot of
traffic from and to the server and probably will result in lag. Using UDP will
probably help avoiding this problem, but having a reliable protocol such as TCP
will make things easier and more stable. We just need to design an efficient
prediction system (probably also a linear one will suffice).

An idea could be using the system I used for a racing game where speed was
fundamental. When you press a key to move, the client sends a packet with
starting coordinates, direction and client time. When you release the key (or
change direction) the client sends another packet with current coordinates and
client time. According to the player speed and the difference of client times
the server check if the coordinates sent by the client are right, if not reply
with a packet with the correct position. The server also check if the time
interval sent by the client is right: if not it means that the values have been
hacked or the lag corrupted them. We can set a tolerance value: if the time
interval exceeds the tolerance then the whole movement is discarded and the
server send a packet to tell the client to place the player at starting coords.


6. SECURITY

To certificate authenticity of the player we can use a system based on digital
signature and hash (encrypted login).

Better security can be provided by encrypting payload of packets using RSA
algorytm. When logging in the client generates both its public and private key
and will send the public one to the server. The server acts the same: when it
starts it creates both the key and replies to login with its public key.
Using encryption will reduce client/server performances because this will need
a lot more calculations.
Furthermore if using digital signature will introduce a lot of overhead.
So there's still the need to discuss if we need to use encryption not only in
the login part.

Solutions to keep the server working and avoid unfair players:

 - DoS attack:
        * Account activation.
        * Limit number of accounts to 1 per email address.
        
 - Cheating/Botting:
        * First of all just keep every calculation done by the server.
        
Also we need the possibility to warn/kick/ban players/accounts/ip addresses.


7. DATABASE

Player data should be stored using databases, probably MySQL. This way player
infos could be easily accessed trough web and used to show stats or required
infos on the website/forum.


8. REGISTRATION

Still to decide if we want to use a dialog (client registration) or to use a
web based interface (web registration).
Registration should ask for less details as possible. This way it will be less
annoying. Required infos will be:
    
    - username
    - password
    - email (to limit account number and to be used to activate account)
    
More infos could be added later for security problem (such as activation codes
and so on).
In RO you also have to choose the sex of your player. It will be better to let
the user choose the sex when creating the player: this way he can delete is male
player and create a female one.


9. SERVER CONTROL

The server can be controlled in two ways:
    
    - In game control: server admins or GMs sending particular commands or a
      trough a GUI (the way it is used in Ultima Online).
    
    - A graphical interface which connects to the server trough an open port.
    
The prefferred way is the first one.

 
10. OPERATING SYSTEM

We have two choices about this: the former is to follow as for the client the
cross-compatibility philosophy. This means that the server will compile on every
windows, unix, macos based system. The latter is to choose the best performance
system (probably linux) and implement a unique os server.
Just remember that the current game server run on linux.


11. ADVANCED DATA TRANSMISSION

Other ways to reduce bandwidth can be considered such as:
    
    - Using bitstreams instead of bytestreams: this way if you need to send a
      boolean values only 1 bit will be required instead of 1 byte (compression
      8:1), item types (4 different types) can be represented sending 2 bits
      instead of 1 byte (compression 8:2), player coordinates can be represented
      using 20 bits instead of 4 bytes (compression 24:20)
      
    - Compressing data following packet id could help reducing bandwidth usage
      as well. RLE compression or more advanced statistical techniques can be
      used. Compression can be disabled in very slow systems (If using
      compression is declared to the server when the client connects to map
      server.