The port number is only one of several attributes used to identify the
connection.

Port numbers are defined by TCP (and UDP). IP does not know about port
numbers, it only knows about IP addresses.

IP passes incoming packets to TCP (or UDP), where the recipient is
determined based on this 4-tuple:

 { local address, local port, remote address, remote port }

Any two connections will differ in at least one of the four fields, so
they are distinct and there is no ambiguity. TCP must examine all four
fields in order to dispatch incoming packets to the right destination.

A "connection" for a typical ServerSocket might look like this:

 { *, 80, *, * }  

where * represent any address or port number.

When a client connects, the listening socket is cloned and the
wildcard fields are bound, resulting in a connection that looks like
this:

 { Server_IP, 80, HostA_IP, Client1_Port }  

A second client connecting from the same host would result in this
connection:

 { Server_IP, 80, HostA_IP, Client2_Port }  

A third client connecting, this time from a different host:

 { Server_IP, 80, HostB_IP, Client3_Port }

/gordon

This really doesn't have anything to do with threads although it is common
(but not required) for each connection to be serviced by its own thread. The
full connection between one machine and another is uniquely identified by
the client IP, client port, server IP and server port.  Multiple connections
to the same service (server IP & port) are disambiguated by the fact that
the come from different clients  (client IP-port pairs)  Each connection
end-point receives its own stream from its corresponding client.

Cheers,

Not exactly. IP is not centered around hosts or nodes, but interfaces.
The host ID part of an IP address does not address that particular host
as a whole, but an IP interface at that host. A single machine can have
multiple interfaces, each with an own IP address.

It doesn't have to be multiple processes. A single process can listen to
multiple ports. The common usage for ports is to differentiate between
services - services which can, but don't need to, be provided by
different processes.

No, it is the so called "half association", aka "socket", aka "transport
address" which determines the ultimate destination.

A "half association" consists of

    - protocol designation (e.g. TCP or UDP)
    - IP address (to identify the host's interface)
    - port

The port alone is not sufficient. E.g. TCP port 1234 and UDP port 1234
are different ports. And TCP port 9876 on the host's IP address
192.168.1.1 is different from TCP port 9876 on the host's IP address
192.168.1.2.

/Thomas

There are two sides to every connection, so there are two IP addresses
and TWO port numbers.  The connection is identified by the tuple
including both pairs of address:port endpoints.

Thus, (10.0.0.1:12345,10.0.0.2:80) and (10.0.0.1:12346,10.0.0.2:80) are
two separate connections between the same two machines with the same
server port, but because the client port is different, the connections
are distinct.

it is, but many clients can share it.
See http://mindprod.com/jgloss/port.html

Please see http://mindprod.com/jgloss/tcpip.html

It will should clear up your various misconceptions.