spider/techdoc/protocol.pod

# -*- perl -*-
=head1 NAME

Aranea Orthogonal Communications Protocol 

$Revision$

=head1 SYNOPSIS

 <Origin>,<Group>,<TimeSeq>,<Hop>[,<From>]|<Tag>,<Data>...

=head1 ABSTRACT

For many years DX Clusters have used a protocol which was designed 
for a non-looped tree ofL</Node>s. This environment has probably never, reliably, 
been achieved in practice; certainly not recently.

There have always been loops, sometimes bringing the network to its 
knees. In modern usage, both in order to get some resilience and also
to expedite information flow, we use internet based, deliberately
looped networks with filtering. Whilst this works, after a fashion, there 
are all sorts of problems that the current PC protocol can never
address.

This document 
describes a complete replacement for the PC protocol. It allows a
fully looped network, is inherently extensible and should be simple
to implement (especially in perl).

All implementations shall use b<only> this protocol
for inter-node communications. 

=head1 DESCRIPTION

This protocol is
designed to be an extensible basis for any type of one -> many
"instant" line-based communications tasks.

The protocol is designed to be flood routed in a meshed network in
as efficient a manner as possible. The reason we have chosen this
mechanism is that most L</Messages> need to be broadcast to allL</Node>s.

Experience has shown thatL</Node>s will appear and (more infrequently) 
disappear without much (or any) notice. 
Therefore, the constantly changing and uncoordinated
nature of the network doesn't lend itself to fixed routing policies. Therefore,
whilst metrics and routing tables (more like routing hint tables) will be 
built up over time, an aggressive aging algorithm will also be employed to prevent
a lot of stale routing information being retained.

Having said that: where routes have
been learned through past traffic, and this data is recent, then direct routing should be used.
Those L</Messages> that could be routed (likely to be mainly single line one to 
one "talk" L</Messages>) that, anyway,
happen sufficiently infrequently that, should they need to be flood routed
(because no route has been learned yet), it is a small cost overall.

=head1 Messages

A message is a single line of UTF8 encoded and HTTP escaped text 
terminated in the standard internet manner with a <CR><LF>. 

Each message consists of a L</Routing Section> and a L</Command Section>. 
The two sections are separated with the '|' character. 
It follows that these
characters (as well as non-printable characters, <CR>, <LF> and
a small number of other reserved characters)
can only be sent escaped. This is described further in the 
L</Command Section> and L</Fields>.

Most of this document is concerned with the L</Routing Section>, however
some L</Standard Commands> which all implementations should issue and
must accept are described.

=head1 Applications

In the past messaging applications such as DX Cluster software have maintained
a fairly strict division between L</Node>s and L</User>s. This protocol attempts
to get away from that by deliberately blurring (or, in some cases, removing) 
any distinction between the two. 

Applications that use this protocol are essentially all peers and therefore the
only real difference between L</Node>s and L</User>s is that a node has one or more 
listeners running that will,
potentially, allow incoming connections from other L</Node>s, L</Endpoint>s or L</User>s. These 
routable entities are called L</Terminal>s.  

Any application that is a sink and/or source of data; is capable of obeying
the protocol message construction rules and understands how to deduplicate incoming messages
correctly can operate as a routeable entity or L</Terminal> in this protocol. It is called an L</Endpoint>.

An L</Endpoint> is called a L</Node> if it accepts connections from L</Endpoint>s and is 
prepared to route messages on their behalf to other L</Node>s or L</Endpoint>s. In addition it
may provide some other, usually simpler, interface (eg simple telnet access) for direct user access. Acting
in the protocol, on their behalf. 

The concept of an L</Endpoint> has been invented because modern clients are 
capable of being more intelligent than simple
character based connections such as telnet or ax25. They also wish to be able to
distinguish between the various classes of message, such as: DX spots, 
announces, talk, logging info etc. It is a pain to have to do it, as now,
by trying to make sense of the (slightly different for each piece of node 
software) human readable "user" version of the output. Far better to pass on
regular, specified, easily computer decodable versions of the message,
(i.e. in this protocol) and leave
the human presentation to the application.

It also helps to modularise the various interfaces that may be implemented such 
as the  legacy, character based connections of existing PC protocol based nodes. 
They should be treated
as local clients, in fact as L</User>s, B<not> as peers in this protocol. It is likely that, in order
to do this, some extra L</Tag>s will need to be defined at application level. 

=head1 Definitions

In this document we use a number of terms that need to be defined.

=head2 Terminal

A L</Terminal> is a routable entity, in other words: a callsign or service that can be routed 
to, that lives at one or a few L</Node>s.  

=head2 User

A L</User> is a connection to a L</Node> (that allows such connections)
that does not occur in protocol. All L</User>s shall be identified with a name
of up to 12 characters in the set [-0-9A-Z_]. All messages have to be routed via the 
L</Node> to which this L</User> is connected. 

=head2 Endpoint

An L</Endpoint> is a connection to a L</Node> that uses the protocol. From a routing point of
view, it is indistiguishable from a L</Node>. The L</Endpoint> is responsible for creating and decoding
well formed protocol messages. An L</Endpoint> does not route beyond the immediate L</Node>(s) to
which it is connected. It may also be a L</Service> connected to a L</Node> which provides some 
addressable service (such as a database) that can be queried.

=head2 Node

A L</Node> is connected to other L</Node>s. It is responsible for routing messages in protocol
from other L</Node>s or L</Endpoint>s, whether directly connected or not. Optionally, a L</Node>
may provide other interfaces, such as direct L</User> connections or legacy PC protocol speaking
DX Clusters. 

=head2 Channel

A L</Channel> is a L</Group> address that is not a L</Terminal>. It is (unless qualified by a L</Terminal>)
broadcast on all a L</Node>s interfaces unless preventing by some filtering or other local policy on 
that L</Node>.

=head2 Service

A L</Service> is application that either plugs into or connects as an L</Endpoint> to a L</Node>. It is an
application that, in effect, is a database. In other words: queries are sent to the L</Service> and it sends
back a reply.

=head1 Routing Section

The application that implements this protocol is essentially a line
oriented message router. One line equals one message. Each line is
effectively a datagram. 

It is assumed thatL</Node>s are connected to
each other using a "reliable" streaming protocol such as TCP/IP or
AX25. Having said that: in context, L</Messages> in this protocol could be 
multi/broadcast, either "as is" or wrapped in some other framing
protocol. 

Although the physical transport between L</Node>s is reliable, the actual message
is unreliable, because this is an unreliable, best effort, "please route my packets
through your node" protocol. There is no guarantee that a message
will get to the other side of a mesh of L</Node>s. There may be a
discontinuity either caused by outage or deliberate filtering. 

However, as it is envisaged that most L</Messages> will be flood routed or,
in the case of directed L</Messages> (those that have a L</Group> containing a L</Terminal> of some kind)
down some/most/all interfaces showing a route for that
direction, it is unlikely that L</Messages> will be lost in practice.

Assuming that there is a path between all the L</Node>s in a network, then it is guaranteed
that a message will be delivered everywhere, eventually. It is possible (indeed likely) that 
copies of  a message
will arrive at L</Node>s more than once. L</Node>s are responsible for deduplicating those messages
using the information in the L</Routing Section>.

=head2 Field Description

All fields in the L</Routing Section> are compulsory except the L</From> field. If it is missing
so is the separating comma.  

The L</Hop> field is incremented on receipt of a message on a node.

Fields are separated by the comma ',' character with the last field 
required followed by the vertical bar '|' character.

The characters allowed in the routing section are restricted. Any 
invalid characters in any field will cause the whole message to be
silently dropped.

More detailed descriptions of the fields follow:

=over

=item B<Origin>

This is a compulsory field. It is the name of the originating node.
The field can contain up to 12 characters in the set [-A-Z0-9_/] in
any order. Higher layers may restrict this further.

The field must not be changed by any other node.

=item B<Group>

This is the Group (or Channel) to be used for this data. It is compulsory.  

It is a string of up to 12 characters 
in the set [-A-Z0-9_/] in any order. 

Optionally, for extra routing to
a particular L</Terminal> connected at a specific L</Node>, or even a
particular L</Terminal> in a L</Group>, 
it may have another 12 character 
string in the same set, concatenated with the first string. The two strings are separated by a ':'
character. For example:

  DX                        # the DX group
  GB7DJK                    # the node GB7DJK
  G1TLH                     # the user or endpoint G1TLH
  GB7DJK:G1TLH              # the user G1TLH at GB7DJK
  DX:G1TLH                  # the user G1TLH in the DX group

This field can contain either a L</Terminal> or some other string which is interpreted
as broadcastable group address. Any message that has a L</Group> that is not recognised as a L</Terminal> must
be broadcast.

This means that messages to callsigns, for whom no specific routing information is available,
will be found by means of a broadcast. Hopefully this will cause some kind of activity o.b.o
that callsign will allow routing tables to be gathered that narrow down the scope of any future
message to that callsign through the network. 

Remember that not all L</Node>s may pass every L</Group> field, depending on local policy.

=item B<TimeSeq>

This is a compulsory field. It is a 10 hexadecimal digit string which
consists of a day no (1-31), 
a flag to indicate NTP syncronisation in use,
seconds within that day (0-86399) [total of 6 hex digits] 
that are concatenated with a sequence number (0-65535)
[4 hex digits] making the total of 10 hexadecimal digits.

The date portion is constructed as:

  my $date = ((((gmtime)[3] << 1) | $ntpflag) << 18) |  (time % 86400);

The sequence number is simply an unsigned short (or 16 bit) number
starting at 0. 

Each message originated at this node will increment the sequence
number.

=item B<Hop>

This is a compulsory field. It is the number of hops from the 
originating node. It is incremented immediately on receipt and
before determining its value. 

So the originating node sends a message with a L</Hop> of 0, the
neighbouring nodes must increment this field before passing
it on to higher layers for onward processing.

Implementations may have an upper limit to this field and may
silently drop incoming L</Messages> with a L</Hop> count greater than the
limit.

=item B<From>

The L</From> field is optional. When present, it represents a L</Terminal> at
the originating L</Node>. If it is missing then either it is not relevant or it
is assumed to be the L</Origin>. 

=back 

=head2 Routing

It is assumed that nodes will be connected in a looped network with
more than one route available (in many cases) to another node.

In anycase, most traffic is not directed, but broadcast to all users
on all nodes.

Each message is uniquely identified by the (L</Origin>,L</TimeSeq>) 
tuple. The basic system will learn which interfaces can see what nodes
by looking at the tuple and merging that with the L</Hop> count. 
Each interface remembers the latest L</TimeSeq> with the lowest L</Hop>
for each L</Origin> that arrives on that interface. It also remembers
the number of L</Messages> for that L</Origin> that has been received on
that interface.

Any message for onward broadcast is duplicated and sent out on all
interfaces that it did not come in on. 

Any message that is directed to a particular node will be sent out on
the "best" interface based on routing information gathered so far. If there
is more than one possible route then, depending on network or local
policy, the message may be duplicated and sent on other interfaces
as well.

=head2 DeDuplication

On receipt of a message, its unique tuple (L</Origin>,L</TimeSeq>) is
checked against a hash table. If it exists: the message is silently
dropped. If it does not exist in the hash table then the tuple is
added.

The hash table is periodically cleaned, removing tuples that 
have expired. The length of time a tuple remains in the hash table
is implementation dependant but could easily be several days, if
required.

This mechanism only ensures that a message broadcast around the network
travels the least distance and through the fewest nodes possible. It
is up to higher layers to make sure that data carried is not, itself,
duplicated!

=head2 Examples

 # on link startup from GB7BAA (both sides hello)
 GB7TLH,ROUTE,3D02350001,0|HELLO,Aranea,1.2,24.123
 GB7BAA,ROUTE,3D02355421,1|HELLO,Aranea,1.1,23.245

 # on user startup to GB7TLH
 GB7TLH,ROUTE,3D042506F2,0,G1TLH|HELLO,PClient,1.3

 # on user disconnection
 GB7TLH,ROUTE,3D9534F32D,0,G1TLH|BYE

 # a talk (actually 'text') message to a user (some distance away
 # from the origin node)
 GB7TLH,G8TIC,3D03450019,3,G1TLH|T,Hiya Mike whats happening?

 # a talk/chat/text message to a Group
 GB7TLH,VHF,0413525F23,2,G1TLH|T,2m is opening on MS

 # a ping to find the whereabouts and distance of a user from a node
 # the hex number on the end is the ping ID
 GB7TLH,G7BRN,1512346543,0,G1TLH|PING,9F4D 

 # this effectively asks whether the user is on-line on a particular node
 GB7TLH,GB7BAA:G7BRN,1512346543,0,G1TLH|PING,35DE

 # A possible reply, same ID as ping followed by the no of hops on the 
 # ping that was received thus telling you how far away it is.
 GB7BAA,G1TLH,1512450534,3,G7BRN|PONG,35DE,3 


=head1 Command Section

The L</Command Section> of the message contains the actual data being
passed. It is called the Command Section because all commands
are identified with a L</Tag> each of which is implemented by 
the software using this protocol. Each </Tag> (usually) is followed by one
or more L</Fields>. 

=head2 Tag

The L</Tag> consists of string of uppercase letters and digits, starting
with a leading, uppercase, letter. Tags should be as short as is meaningful.

Valid tags would be:

 DX
 PC23
 ANN

Invalid tags include:

 1AAA
 dx
 Ann

The L</Tag> is separated from its data L</Fields> by a comma ','. 

=head2 Fields

All fields
in any subsequent data shall be separated by a comma ','.
All fields shall
be HTTP encoded such that reserved characters (comma ',', 
vertical bar '|',
percent '%', 
equals '=' 
and non printable characters less than 127 (or %7F in hex)
[including newline and carraige return] are translated to
their two hex digit equivalent preceeded by the percent '%' character.

For example:

 "%0D%0A" is "<carriage return><linefeed>".
 "hello%2C there" is "hello, there"

This is not standard CSV, fields are not quoted (delimited with either
' or ").

All national characters above 127 are UTF8 encoded in the
standard perl 5.8.x way. It follows that all (perl) programs that
are written according to this specification must say:

 use UTF8;

A message (or line) is terminated with <carriage return><linefeed>
0x0d 0x0a. Incoming L</Messages> must be accepted even when terminated
with just <linefeed>.

Care must be taken to make sure that fields have any reserved characters
encoded. In particular: it is perfectly permissible to have <linefeed>
characters in a field - so long as they are escaped.

Fields come in two styles: either simple fields (just containing
data) or B<key>=B<value> pairs. Each pair must be separated from
the next by a comma ','. The B<key> must consist of the set of
characters [a-z0-9_] (ie lowercase letters, digits and underscore),
with a leading letter. The B<value> must be HTTP encoded as
specified above and can otherwise contain any character.

There is no maximum size specified for a message. It is up to each
implimentation to enforce one (if only for their own protection).

=head2 Standard Commands

There are a number of L</Standard Commands> which must be accepted by 
all implementations.

=over

=item B<HELLO>

 HELLO,<software name>,<version>,<build>,<comments>

Command sent on connection to another node. Both sides send their information
to the other. All the possible arguments are optional, although some of the
arguments should be sent in order to help diagnose problems. This command is
broadcast.

=item B<BYE> 

 BYE,<comments>

Command sent to all connections when the software is shutting down. This is sent
by the node just before shutdown occurs. This is really only used to help the
network prune its routing tables. It isn't a requirement. The <comment> field
is optional. 

=item B<DISC>

 DISC,<node name>,<comments>

Command sent when a node has disconnected from this node. This message is sent when
an interface shuts down. It need not be sent if a L<BYE> from an interface for
that node has just been received. This command should be broadcast.

The <node name> is mandatory and is the name of the interface that has just 
disconnected.

=item B<PING>

 PING,<user>,<ping id>

Command to send a ping to a node or user. This command is used both by the software
and users to determine a) whether a node or user exists and b) how good the path is
between them. 

The <ping id> is a unique string which is usually the hexadecimal equivalent of an 
integer that is incremented every time it is used. But it can be anything that
will identify this ping using the tuple (L<Origin>,<ping id>) as unique.

=item B<PONG>

 PONG,<ping id>,<user>,<no of hops on ping>

Command to reply to a ping. This is sent as a reply to an incoming ping command.
The <ping id> is the one supplied and the <no of hops on ping> is the number of
hops it took for the ping to arrive.

=item B<T>

 T,<text>

All implementations must be able to send "text" (encoded as specified in 
L</Fields>). There would be little point in doing all this otherwise!

=back

=head1 AUTHOR

Dirk Koopman, G1TLH, E<lt>djk@tobit.co.ukE<gt>

=head1 COPYRIGHT AND LICENSE

Copyright 2004-2005 by Dirk Koopman, G1TLH

This library is free software; you can redistribute it and/or modify
it under the same terms as Perl itself.

$Revision$

=cut