One of the main reasons for using Erlang instead of other functional languages is Erlang's ability to handle concurrency and distributed programming. By concurrency we mean programs which can handle several threads of execution at the same time. For example, modern operating systems would allow you to use a word processor, a spreadsheet, a mail client and a print job all running at the same time. Of course each processor (CPU) in the system is probably only handling one thread (or job) at a time, but it swaps between the jobs a such a rate that it gives the illusion of running them all at the same time. It is easy to create parallel threads of execution in an Erlang program and it is easy to allow these threads to communicate with each other. In Erlang we call each thread of execution a process.

(Aside: the term "process" is usually used when the threads of execution share no data with each other and the term "thread" when they share data in some way. Threads of execution in Erlang share no data, that's why we call them processes).

The Erlang BIF spawn is used to create a new process: spawn(Module, Exported_Function, List of Arguments). Consider the following module:

5> c(tut14).
{ok,tut14}
6> tut14:say_something(hello, 3).
hello
hello
hello
done

We can see that function say_something writes its first argument the number of times specified by second argument. Now look at the function start. It starts two Erlang processes, one which writes "hello" three times and one which writes "goodbye" three times. Both of these processes use the function say_something. Note that a function used in this way by spawn to start a process must be exported from the module (i.e. in the -export at the start of the module).

9> tut14:start().
hello
goodbye
<0.63.0>
hello
goodbye
hello
goodbye

Notice that it didn't write "hello" three times and then "goodbye" three times, but the first process wrote a "hello", the second a "goodbye", the first another "hello" and so forth. But where did the <0.63.0> come from? The return value of a function is of course the return value of the last "thing" in the function. The last thing in the function start is:

spawn returns a process identifier, or pid, which uniquely identifies the process. So <0.63.0> is the pid of the spawn function call above. We will see how to use pids in the next example.

Note as well that we have used ~p instead of ~w in io:format. To quote the manual: "~p Writes the data with standard syntax in the same way as ~w, but breaks terms whose printed representation is longer than one line into many lines and indents each line sensibly. It also tries to detect lists of printable characters and to output these as strings".

In the following example we create two processes which send messages to each other a number of times.

1> c(tut15).
{ok,tut15}
2> tut15: start().
<0.36.0>
Pong received ping
Ping received pong
Pong received ping
Ping received pong
Pong received ping
Ping received pong
ping finished
Pong finished

The function start first creates a process, let's call it "pong":

This process executes tut15:pong(). Pong_PID is the process identity of the "pong" process. The function start now creates another process "ping".

This process executes:

<0.36.0> is the return value from the start function.

The process "pong" now does:

The receive construct is used to allow processes to wait for messages from other processes. It has the format:

Note: no ";" before the end.

Messages between Erlang processes are simply valid Erlang terms. I.e. they can be lists, tuples, integers, atoms, pids etc.

Each process has its own input queue for messages it receives. New messages received are put at the end of the queue. When a process executes a receive, the first message in the queue is matched against the first pattern in the receive, if this matches, the message is removed from the queue and the actions corresponding to the the pattern are executed.

However, if the first pattern does not match, the second pattern is tested, if this matches the message is removed from the queue and the actions corresponding to the second pattern are executed. If the second pattern does not match the third is tried and so on until there are no more pattern to test. If there are no more patterns to test, the first message is kept in the queue and we try the second message instead. If this matches any pattern, the appropriate actions are executed and the second message is removed from the queue (keeping the first message and any other messages in the queue). If the second message does not match we try the third message and so on until we reach the end of the queue. If we reach the end of the queue, the process blocks (stops execution) and waits until a new message is received and this procedure is repeated.

Of course the Erlang implementation is "clever" and minimizes the number of times each message is tested against the patterns in each receive.

Now back to the ping pong example.

"Pong" is waiting for messages. If the atom finished is received, "pong" writes "Pong finished" to the output and as it has nothing more to do, terminates. If it receives a message with the format:

it writes "Pong received ping" to the output and sends the atom pong to the process "ping":

Note how the operator "!" is used to send messages. The syntax of "!" is:

I.e. Message (any Erlang term) is sent to the process with identity Pid.

After sending the message pong to the process "ping", "pong" calls the pong function again, which causes it to get back to the receive again and wait for another message. Now let's look at the process "ping". Recall that it was started by executing:

Looking at the function ping/2 we see that the second clause of ping/2 is executed since the value of the first argument is 3 (not 0) (first clause head is ping(0,Pong_PID), second clause head is ping(N,Pong_PID), so N becomes 3).

The second clause sends a message to "pong":

self() returns the pid of the process which executes self(), in this case the pid of "ping". (Recall the code for "pong", this will land up in the variable Ping_PID in the receive previously explained.)

"Ping" now waits for a reply from "pong":

and writes "Ping received pong" when this reply arrives, after which "ping" calls the ping function again.

N-1 causes the first argument to be decremented until it becomes 0. When this occurs, the first clause of ping/2 will be executed:

The atom finished is sent to "pong" (causing it to terminate as described above) and "ping finished" is written to the output. "Ping" then itself terminates as it has nothing left to do.

In the above example, we first created "pong" so as to be able to give the identity of "pong" when we started "ping". I.e. in some way "ping" must be able to know the identity of "pong" in order to be able to send a message to it. Sometimes processes which need to know each others identities are started completely independently of each other. Erlang thus provides a mechanism for processes to be given names so that these names can be used as identities instead of pids. This is done by using the register BIF:

We will now re-write the ping pong example using this and giving the name pong to the "pong" process:

2> c(tut16).
{ok, tut16}
3> tut16:start().
<0.38.0>
Pong received ping
Ping received pong
Pong received ping
Ping received pong
Pong received ping
Ping received pong
ping finished
Pong finished

In the start/0 function,

both spawns the "pong" process and gives it the name pong. In the "ping" process we can now send messages to pong by:

so that ping/2 now becomes ping/1 as we don't have to use the argument Pong_PID.

Now let's re-write the ping pong program with "ping" and "pong" on different computers. Before we do this, there are a few things we need to set up to get this to work. The distributed Erlang implementation provides a basic security mechanism to prevent unauthorized access to an Erlang system on another computer. Erlang systems which talk to each other must have the same magic cookie. The easiest way to achieve this is by having a file called .erlang.cookie in your home directory on all machines which on which you are going to run Erlang systems communicating with each other (on Windows systems the home directory is the directory where pointed to by the $HOME environment variable - you may need to set this. On Linux or Unix you can safely ignore this and simply create a file called .erlang.cookie in the directory you get to after executing the command cd without any argument). The .erlang.cookie file should contain one line with the same atom. For example, on Linux or Unix in the OS shell:

$ cd
$ cat > .erlang.cookie
this_is_very_secret
$ chmod 400 .erlang.cookie

The chmod above make the .erlang.cookie file accessible only by the owner of the file. This is a requirement.

When you start an Erlang system which is going to talk to other Erlang systems, you must give it a name, e.g.:

$ erl -sname my_name

We will see more details of this later. If you want to experiment with distributed Erlang, but you only have one computer to work on, you can start two separate Erlang systems on the same computer but give them different names. Each Erlang system running on a computer is called an Erlang node.

(Note: erl -sname assumes that all nodes are in the same IP domain and we can use only the first component of the IP address, if we want to use nodes in different domains we use -name instead, but then all IP address must be given in full.)

Here is the ping pong example modified to run on two separate nodes:

Let us assume we have two computers called gollum and kosken. We will start a node on kosken called ping and then a node on gollum called pong.

On kosken (on a Linux/Unix system):

kosken> erl -sname ping
Erlang (BEAM) emulator version 5.2.3.7 [hipe] [threads:0]

Eshell V5.2.3.7  (abort with ^G)
(ping@kosken)1>

On gollum:

gollum> erl -sname pong
Erlang (BEAM) emulator version 5.2.3.7 [hipe] [threads:0]

Eshell V5.2.3.7  (abort with ^G)
(pong@gollum)1>

Now we start the "pong" process on gollum:

(pong@gollum)1> tut17:start_pong().
true

and start the "ping" process on kosken (from the code above you will see that a parameter of the start_ping function is the node name of the Erlang system where "pong" is running):

(ping@kosken)1> tut17:start_ping(pong@gollum).
<0.37.0>
Ping received pong
Ping received pong 
Ping received pong
ping finished

Here we see that the ping pong program has run, on the "pong" side we see:

(pong@gollum)2>
Pong received ping                 
Pong received ping                 
Pong received ping                 
Pong finished                      
(pong@gollum)2>

Looking at the tut17 code we see that the pong function itself is unchanged, the lines:

work in the same way irrespective of on which node the "ping" process is executing. Thus Erlang pids contain information about where the process executes so if you know the pid of a process, the "!" operator can be used to send it a message if the process is on the same node or on a different node.

A difference is how we send messages to a registered process on another node:

We use a tuple {registered_name,node_name} instead of just the registered_name.

In the previous example, we started "ping" and "pong" from the shells of two separate Erlang nodes. spawn can also be used to start processes in other nodes. The next example is the ping pong program, yet again, but this time we will start "ping" in another node:

Assuming an Erlang system called ping (but not the "ping" process) has already been started on kosken, then on gollum we do:

(pong@gollum)1> tut18:start(ping@kosken).
<3934.39.0>
Pong received ping
Ping received pong
Pong received ping
Ping received pong
Pong received ping
Ping received pong
Pong finished
ping finished

Notice we get all the output on gollum. This is because the io system finds out where the process is spawned from and sends all output there.

Now for a larger example. We will make an extremely simple "messenger". The messenger is a program which allows users to log in on different nodes and send simple messages to each other.

Before we start, let's note the following:

We will set up the messenger by allowing "clients" to connect to a central server and say who and where they are. I.e. a user won't need to know the name of the Erlang node where another user is located to send a message.

File messenger.erl:

To use this program you need to:

In the following example of use of this program I have started nodes on four different computers, but if you don't have that many machines available on your network you could start up several nodes on the same machine.

We start up four Erlang nodes: messenger@super, c1@bilbo, c2@kosken, c3@gollum.

First we start up a the server at messenger@super:

(messenger@super)1> messenger:start_server().
true

Now Peter logs on at c1@bilbo:

(c1@bilbo)1> messenger:logon(peter).
true
logged_on

James logs on at c2@kosken:

(c2@kosken)1> messenger:logon(james).
true
logged_on

and Fred logs on at c3@gollum:

(c3@gollum)1> messenger:logon(fred).
true
logged_on

Now Peter sends Fred a message:

(c1@bilbo)2> messenger:message(fred, "hello").
ok
sent

And Fred receives the message and sends a message to Peter and logs off:

Message from peter: "hello"
(c3@gollum)2> messenger:message(peter, "go away, I'm busy").
ok
sent
(c3@gollum)3> messenger:logoff().
logoff

James now tries to send a message to Fred:

(c2@kosken)2> messenger:message(fred, "peter doesn't like you").
ok
receiver_not_found

But this fails as Fred has already logged off.

First let's look at some of the new concepts we have introduced.

There are two versions of the server_transfer function: one with four arguments (server_transfer/4) and one with five (server_transfer/5). These are regarded by Erlang as two separate functions.

Note how we write the server function so that it calls itself, via server(User_List), and thus creates a loop. The Erlang compiler is "clever" and optimizes the code so that this really is a sort of loop and not a proper function call. But this only works if there is no code after the call, otherwise the compiler will expect the call to return and make a proper function call. This would result in the process getting bigger and bigger for every loop.

We use functions from the lists module. This is a very useful module and a study of the manual page is recommended (erl -man lists). lists:keymember(Key,Position,Lists) looks through a list of tuples and looks at Position in each tuple to see if it is the same as Key. The first element is position 1. If it finds a tuple where the element at Position is the same as Key, it returns true, otherwise false.

3> lists:keymember(a, 2, [{x,y,z},{b,b,b},{b,a,c},{q,r,s}]).
true
4> lists:keymember(p, 2, [{x,y,z},{b,b,b},{b,a,c},{q,r,s}]).
false

lists:keydelete works in the same way but deletes the first tuple found (if any) and returns the remaining list:

5> lists:keydelete(a, 2, [{x,y,z},{b,b,b},{b,a,c},{q,r,s}]).
[{x,y,z},{b,b,b},{q,r,s}]

lists:keysearch is like lists:keymember, but it returns {value,Tuple_Found} or the atom false.

There are a lot more very useful functions in the lists module.

An Erlang process will (conceptually) run until it does a receive and there is no message which it wants to receive in the message queue. I say "conceptually" because the Erlang system shares the CPU time between the active processes in the system.

A process terminates when there is nothing more for it to do, i.e. the last function it calls simply returns and doesn't call another function. Another way for a process to terminate is for it to call exit/1. The argument to exit/1 has a special meaning which we will look at later. In this example we will do exit(normal) which has the same effect as a process running out of functions to call.

The BIF whereis(RegisteredName) checks if a registered process of name RegisteredName exists and return the pid of the process if it does exist or the atom undefined if it does not.

You should by now be able to understand most of the code above so I'll just go through one case: a message is sent from one user to another.

The first user "sends" the message in the example above by:

After testing that the client process exists:

and a message is sent to mess_client:

The client sends the message to the server by:

and waits for a reply from the server.

The server receives this message and calls:

which checks that the pid From is in the User_List:

If keysearch returns the atom false, some sort of error has occurred and the server sends back the message:

which is received by the client which in turn does exit(normal) and terminates. If keysearch returns {value,{From,Name}} we know that the user is logged on and is his name (peter) is in variable Name. We now call:

Note that as this is server_transfer/5 it is not the same as the previous function server_transfer/4. We do another keysearch on User_List to find the pid of the client corresponding to fred:

This time we use argument 2 which is the second element in the tuple. If this returns the atom false we know that fred is not logged on and we send the message:

which is received by the client, if keysearch returns:

we send the message:

to fred's client and the message:

to peter's client.

Fred's client receives the message and prints it:

and peter's client receives the message in the await_result function.