FT Baseline: What to expect and look for

Fault Tolerance:
===================

Two servers will be running on two computers.

We have no distinction between "primary" and "backup" servers.  The
processing for a game may be distributed across multiple servers.

Servers lock the database when they read and unlock after they write.
This ensures determinism for the servers.  Because of this, all state
is stored in the database and therefore no checkpointing is used.

When a client contacts the naming service, it retrieves the entire
list of servers that support the GameServer object.  When a fault
occurs (a server is ctrl-c'ed), the client chooses a new server and
attempts to connect to it.  If none of the servers can be connected
to, then the client reconnects to the naming service in order to
download a new list of servers.  If this list is null, the game cannot
continue until a new server registers with the naming service.

A fault may not be caused until clients have finished their exchange
with the naming service.


Recovery:
===================

"Girltalk" does not seem to run the server application well.
"Monopoly" does not seem to run the client application well.

On a successful recovery run, you can expect clients to display a
message indicating which server they have reconnected to and the name
of the fault that triggered the reconnect.  The game will then
continue.


Exception Handling:
===================

Currently, exceptions are handled only for Game::Update().  The reason
we decided to implement this first is because it is called most frequently.

Handling of exceptions for naming-service related exceptions has not been implemented.

Most non-CORBA related exceptions have not been caught.


How to run the fault tolerant demo:
====================================

1) set library path:
   export LD_LIBRARY_PATH=/afs/ece/class/ece749/public_html/teams/team7/orbacus/lib

2) create a symbolic link onto the directory you extracted the .tar file on.
   ln -s /afs/ece/class/ece749/public_html/teams/team7/orbacus orbacus

3) make the source code. (Type "make" in /src/)

4) run the nameserver (Execute /src/common/runname.sh)

5) run the servers on multiple computers (cwd to /src/server/, execute
   runserver.sh $1, where $1 is sequentially increasing from 1 and
   identifies the server to the naming service.  Servers must be started
   in order, and a server must display "Binding Gameserver $1 To Naming 
   Service" before another server is started)

6) run the clients on muliple computers (cwd to
   /src/client/ascii_client, execute runclient.sh $1, where $1 indicates
   which server to connect to.  $1 must be the id of a valid server)

7) ctrl-c server 0, and witness the clients reconnect to server 1.

8) rerun server 0, and ctrl-c server 1.  Watch the clients reconnect to server 0