William/Reticulum
1
0
Fork 0
mirror of https://github.com/markqvist/Reticulum.git synced 2024-11-22 21:50:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

Added overview document

Browse Source
This commit is contained in:
Mark Qvist 2020-01-26 13:51:04 +01:00
parent 91728b100e
commit 03705372f8
1 changed files with 492 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

492
Documentation/Reticulum Overview.md Normal file
View File

@ -0,0 +1,492 @@
# Reticulum Overview
This paper will briefly describe the overall purpose and operating principles of Reticulum, a
networking stack designed for reliable and secure communication over high-latency, low-bandwidth
links. It should give you an overview of how the stack works, and an understanding of how to
develop networked applications using Reticulum.
This document is not an exhaustive source of information on Reticulum, at least not yet. Currently,
the best place to go for such information is the Python reference implementation of Reticulum. Both
the reference implementation and this document may (and will) change rapidly in the current phase
of development, but historical versions will always be available in the Git repositories.
After reading this document, you should be well-equipped to understand how a Reticulum network
operates, what it can achieve, and how you can use it yourself. If you want to help out with the
development, this is also the place to start, since it will also provide a pretty clear overview of the
sentiments and the philosophy behind Reticulum.
## Motivation
The primary motivation for designing and implementing Reticulum has been the current lack of
reliable, functional and secure minimal-infrastructure modes of digital communication. It is my
belief that it is highly desirable to create a cheap and reliable way to set up a wide-range digital
communication network that can securely allow exchange of information between people and
machines, with no central point of authority, control, censorship or barrier to entry.
Almost all of the various networking stacks in wide use today share a common limitation, namely
that they require large amounts of coordination to work. You can’t just plug in a bunch of ethernet
cables to the same switch, or turn on a number of WiFi radios, and expect such a setup to provide a
reliable platform for communication.
The designers of the Internet Protocol had the foresight to create a protocol that powers the modern
Internet, and works brilliantly in world very different from when it was conceived. But networks
using the traditional IP stack needs large amounts of coordination from the people involved, and
without central actors in ultimate control of network segments, it is very easy for a single person to
render the platform unusable for everyone else. These limitations are inherent to the design
principles of IP, and during the design of IP, this was a very reasonable tradeoff indeed.
Reticulum aims to require as little coordination and trust as possible. In fact, the only
“coordination” required is to know how to get connected to a Reticulum network. Since Reticulum
is medium agnostic, this could be whatever is best suited to the situation. In some cases, this might
be 1200 baud packet radio links over VHF frequencies, in other cases it might be a microwave
network using off-the-shelf radios. At the time of release of this document, the recommended setup
is using cheap LoRa radio modules with an open source firmware (see the chapter _Reference System
Setup_ ), connected to a small computer like a Raspberry Pi. As an example, the default reference
setup provides a channel capacity of 5.4 Kbps, and a usable direct node-to-node range of around 15
kilometers (indefinitely extendable by using multiple hops).
## Goals
To be as widely usable and easy to implement as possible, the following goals have been used to
guide the design of Reticulum:
- **Fully useable as open source software stack**
Reticulum must be implemented, and be able to run using only open source software. This is
critical to ensuring availability, security and transparency of the system.
- **Hardware layer agnosticism**
Reticulum shall be fully hardware agnostic, and should be useable over a wide range
physical networking layers, such as data radios, serial lines, modems, handheld transceivers,
wired ethernet, wifi, or anything else that can carry a digital data stream. Hardware made for
dedicated Reticulum use shall be as cheap as possible and use off-the-shelf components, so
it can be easily replicated.
- **Very low bandwidth requirements**
Reticulum should be able to function reliably over links with a data capacity as low as _1,_
_bps_.
- **Encryption by default**
Reticulum must use encryption by default where possible and applicable.
- **Unlicensed use**
Reticulum shall be functional over physical communication mediums that do not require any
form of license to use. Reticulum must be designed in a way, so it is usable over ISM radio
frequency bands, and can provide functional long distance links in such conditions.
- **Supplied software**
Apart from the core networking stack and API, that allows any developer to build
applications with Reticulum, a basic communication suite using Reticulum must be
implemented and released at the same time as Reticulum itself. This shall serve both as a
functional communication suite, and as an example and learning resource to others wishing
to build applications with Reticulum.
- **Ease of use**
The reference implementation of Reticulum is written in Python, to make it very easy to use
and understand. Any programmer with only basic experience should be able to use
Reticulum in their own applications.
- **Low cost**
It shall be as cheap as possible to deploy a communication system based on Reticulum. This
should be achieved by using cheap off-the-shelf hardware that potential users might already
own. The cost of setting up a functioning node should be less than $100 even if all parts
needs to be purchased.
# Introduction & Basic Functionality
Reticulum is a networking stack suited for high-latency, low-bandwidth links. Reticulum is at it’s
core _message oriented_ , but can provide connection oriented sessions. It is suited for both local
point-to-point or point-to-multipoint scenarios where alle nodes are within range of each other, as
well as scenarios where packets need to be transported over multiple hops to reach the recipient.
Reticulum does away with the idea of addresses and ports known from IP, TCP and UDP. Instead
Reticulum uses the singular concept of _destinations_. Any application using Reticulum as it’s
networking stack will need to create one or more destinations to receive data, and know the
destinations it needs to send data to.
Reticulum encrypts all data by default using public-key cryptography. Any message sent to a
destination is encrypted with that destinations public key. Reticulum also offers symmetric key
encryption for group-oriented communications, as well as unencrypted packets for broadcast
purposes, or situations where you need the communication to be in plain text. The multi-hop
transport, coordination, verification and reliability layers are fully autonomous and based on public
key cryptography.
Reticulum can connect to a variety of interfaces such as radio modems, data radios and serial ports,
and offers the possibility to easily tunnel Reticulum traffic over IP links such as the Internet or
private IP networks.
## Destinations
To receive and send data with the Reticulum stack, an application needs to create one or more
destinations. Reticulum uses three different basic destination types, and one special:
- **Single**
The _single_ destination type defines a public-key encrypted destination. Any data sent to this
destination will be encrypted with the destination’s public key, and will only be readable by
the creator of the destination.
- **Group**
The _group_ destination type defines a symmetrically encrypted destination. Data sent to this
destination will be encrypted with a symmetric key, and will be readable by anyone in
possession of the key. The _group_ destination can be used just as well by only two peers, as it
can by many.
- **Plain**
A _plain_ destination type is unencrypted, and suited for traffic that should be broadcast to a
number of users, or should be readable by anyone.
- **Link**
A _link_ is a special destination type, that serves as an abstract channel between two _single_
destinations, directly connected or over multiple hops. The _link_ also offers reliability and
more efficient encryption, and as such is useful even when nodes are directly connected.
## Destination Naming
Destinations are created and named in an easy to understand dotted notation of _aspects_ , and
represented on the network as a hash of this value. The hash is a SHA-256 truncated to 80 bits. The
top level aspect should always be the a unique identifier for the application using the destination.
The next levels of aspects can be defined in any way by the creator of the application. For example,
a destination for a messaging application could be made up of the application name and a username,
and look like this:
```
name: simplemessenger.someuser hash: 2a7ddfab5213f916dea
```
For the _single_ destination, Reticulum will automatically append the associated public key as a
destination aspect before hashing. This is done to ensure only the correct destination is reached,
since anyone can listen to any destination name. Appending the public key ensures that a given
packet is only directed at the destination that holds the corresponding private key to decrypt the
packet. It is important to understand that anyone can use the destination name
_simplemessenger.myusername_ , but each person that does so will still have a different destination
hash, because their public keys will differ. In actual use of _single_ destination naming, it is advisable
not to use any uniquely identifying features in aspect naming, though. In the simple messenger
example, when using _single_ destinations, we would instead use a destination naming scheme such
as _simplemessenger.user_ where appending the public key expands the destination into a uniquely
identifying one.
To recap, the destination types should be used in the following situations:
- **Single**
When private communication between two endpoints is needed. Supports routing.
- **Group**
When private communication between two or more endpoints is needed. More efficient in
data usage than _single_ destinations. Supports routing indirectly, but must first be established
through a _single_ destination.
- **Plain**
When plain-text communication is desirable, for example when broadcasting information.
To communicate with a _single_ destination, you need to know it’s public key. Any method for
obtaining the public key is valid, but Reticulum includes a simple mechanism for making other
nodes aware of your destinations public key, called the _announce_.
Note that this information could be shared and verified in many other ways, and that it is therefore
not required to use the announce functionality, although it is by far the easiest, and should probably
be used if you are not confident in how to verify public keys and signatures manually.
## Public key announcements
An _announce_ will send a special packet over any configured interfaces, containing all needed
information about the destination hash and public key, and can also contain some additional,
application specific data. The entire packet is signed by the sender to ensure authenticity. It is not
required to use the announce functionality, but in many cases it will be the simplest way to share
public keys on the network. As an example, an announce in a simple messenger application might
contain the following information:
- The announcers destination hash
- The announcers public key
- Application specific data, in this case the users nickname and availability status
- A random blob, making each new announce unique
- A signature of the above information, verifying authenticity
With this information, any Reticulum node that receives it will be able to reconstruct an outgoing
destination to securely communicate with that destination. You might have noticed that there is one
piece of information lacking to reconstruct full knowledge of the announced destination, and that is
the aspect names of the destination. These are intentionally left out to save bandwidth, since they
will be implicit in almost all cases. If a destination name is not entirely implicit, information can be
included in the application specific data part that will allow the receiver to infer the naming.
It is important to note that announcements will be forwarded throughout the network according to a
certain pattern. This will be detailed later. Seeing how _single_ destinations are always tied to a
private/public key pair leads us to the next topic.
## Identities
In Reticulum, an _identity_ does not necessarily represent a personal identity, but is an abstraction that
can represent any kind of _verified entity_. This could very well be a person, but it could also be the
control interface of a machine, a program, robot, computer, sensor or something else entirely. In
general, any kind of agent that can act, or be acted upon, or store or manipulate information, can be
represented as an identity.
As we have seen, a _single_ destination will always have an _identity_ tied to it, but not _plain_ or _group_
destinations. Destinations and identities share a multilateral connection. You can create a
destination, and if it is not connected to an identity upon creation, it will just create a new one to use
automatically. This may be desirable in some situations, but often you will probably want to create
the identity first, and then link it to created destinations.
Building upon the simple messenger example, we could use an identity to represent the user of the
application. Destinations created will then be linked to this identity to allow communication to
reach the user. In such a case it is of great importance to store the user’s identity securely and
privately.
## Getting Further
The above functions and principles form the core of Reticulum, and would suffice to create
functional networked applications in local clusters, for example over radio links where all interested
nodes can hear each other. But to be truly useful, we need a way to go further. In the next chapter,
two concepts that allow this will be introduced, _paths_ and _resources_.
# Transport
I have purposefully avoided the term routing until now, and will continue to do so, because the
current methods of routing used in IP based networks are fundamentally incompatible for the link
types that Reticulum was designed to handle. These routing methodologies assume trust at the
physical layer. Since Reticulum is designed to run over open radio spectrum, no such trust exists.
Furthermore, existing routing protocols like BGP or OSPF carry too much overhead to be
practically useable over bandwidth-limited, high-latency links.
To overcome such challenges, Reticulum’s _Transport_ system uses public-key cryptography to
implement the concept of _paths_ that allow discovery of how to get information to a certain
destination, and _resources_ that help alleviate congestion and make reliable communication more
efficient and less bandwidth-hungry.
## Threading a Path
In networks with changing topology and trustless connectivity, nodes need a way to establish
_verified connectivity_ with each other. To do this, the following process is employed:
- First, the node that wishes to establish connectivity will send out a special packet, that
traverses the network and locates the desired destination. Along the way, the nodes that
forward the packet will take note of this _link request_.
- Second, if the destination accepts the _link request_ , it will send back a packet that proves the
authenticity of it’s identity (and the receipt of the link request) to the initiating node. All
nodes that initially forwarded the packet will also be able to verify this proof, and thus
accept the validity of the _link_ throughout the network.
- When the validity of the _link_ has been accepted by forwarding nodes, these nodes will
remember the _link_ , and it can subsequently be used by referring to a hash representing it.
- As a part of the _link request_ , a Diffie-Hellman key exchange takes place, that sets up an
efficient symmetrically encrypted tunnel between the two nodes, using elliptic curve
cryptography. As such, this mode of communication is preferred, even for situations when
nodes can directly communicate, when the amount of data to be exchanged numbers in the
tens of packets.
- When a _link_ has been set up, it automatically provides message receipt functionality, so the
sending node can obtain verified confirmation that the information reached the intended
recipient.
In a moment, we will discuss the specifics of how this methodology is implemented, but let’s first
recap what purposes this serves. We first ensure that the node answering our request is actually the
one we want to communicate with, and not a malicious actor pretending to be so. At the same time
we establish an efficient encrypted channel. The setup of this is relatively cheap in terms of
bandwidth, so it can be used just for a short exchange, and then recreated as needed, which will also
rotate encryption keys (keys can also be rotated over an existing path), but the link can also be kept
alive for longer periods of time, if this is more suitable to the application. The amount of bandwidth
used on keeping a link open is practically negligible. The procedure also inserts the _link id_ , a hash
calculated from the link request packet, into the memory of forwarding nodes, which means that the
communicating nodes can thereafter reach each other simply by referring to this _link id_.
**Step 1, pathfinding**
The pathfinding method builds on the _announce_ functionality discussed earlier. When an announce
is sent out by a node, it will be forwarded by any node receiving it, but according to some specific
rules:
- If this announce has already been received before, ignore it.
- Record into a table which node the announce was received from, and how many times in
total it has been retransmitted to get here.
- If the announce has been retransmitted _m+1_ times, it will not be forwarded. By default, _m_ is
set to 18.
- The announce will be assigned a delay _d_ = _ch_ seconds, where _c_ is a decay constant, by
default 2, and _h_ is the amount of times this packet has already been forwarded.
- The packet will be given a priority _p = 1/d_.
- If at least _d_ seconds has passed since the announce was received, and no other packets with a
priority higher than _p_ are waiting in the queue (see Packet Prioritisation), and the channel is
not utilized by other traffic, the announce will be forwarded.
- If no other nodes are heard retransmitting the announce with a greater hop count than when
it left this node, transmitting it will be retried _r_ times. By default, _r_ is set to 2. Retries follow
same rules as above, with the exception that it must wait for at least _d = ch+1 + t_ seconds, ie.,
the amount of time it would take the next node to retransmit the packet. By default, _t_ is set to
10.
- If a newer announce from the same destination arrives, while an identical one is already in
the queue, the newest announce is discarded. If the newest announce contains different
application specific data, it will replace the old announce, but will use _d_ and _p_ of the old
announce.
Once an announce has reached a node in the network, any other node in direct contact with that
node will be able to reach the destination the announce originated from, simply by sending a packet
addressed to that destination. Any node with knowledge of the announce will be able to direct the
packet towards the destination by looking up the next node with the shortest amount of hops to the
destination. The specifics of this process is detailed in _Path Calculation_.
According to these rules and default constants, an announce will propagate throughout the network
in a predictable way. In an example network utilising the default constants, and with an average link
distance of _Lavg =_ 15 kilometers, an announce will be able to propagate outwards to a radius of 180
kilometers in 34 minutes, and a _maximum announce radius_ of 270 kilometers in approximately 3
days. Methods for overcoming the distance limitation of _m * Lavg_ will be introduced later in this
chapter.
**Step 2, link establishment**
After seeing how the conditions for finding a path through the network are created, we will now
explore how two nodes can establish reliable communications over multiple hops. The _link_ in
Reticulum terminology should not be viewed as a direct node-to-node link on the physical layer, but
as an abstract channel, that can be open for any amount of time, and can span an arbitrary number
of hops, where information will be exchanged between two nodes.
- When a node in the network wants to establish verified connectivity with another node, it
will create a _link request_ packet, and broadcast it.
- The _link request_ packet contains the destination hash _Hd_ , and an asymmetrically encrypted
part containing the following data: The source hash _Hs_ , a symmetric key _Lk_ , a truncated
hash of a random number _Hr_ , and a signature _S_ of the plaintext values of _Hd_ , _Hs_ , _Lk_ and _Hr_.
- The broadcasted packet will be directed through the network according to the rules laid out
previously.
- Any node that forwards the link request will store a _link id_ in it’s _link table_ , along with the
amount of hops the packet had taken when received. The link id is a hash of the entire link
request packet. If the path is not _proven_ within some set amount of time, the entry will be
dropped from the table again.
- When the destination receives the link request packet, it will decide whether to accept the
request. If it is accepted, it will create a special packet called a _proof_. A _proof_ is a simple
construct, consisting of a truncated hash of the message that needs to be proven, and a
signature (made by the destination’s private key) of this hash. This _proof_ effectively verifies
that the intended recipient got the packet, and also serves to verify the discovered path
through the network. Since the _proof_ hash matches the _path id_ in the intermediary nodes’
_path tables_ , the intermediary nodes can forward the proof all the way back to the source.
- When the source receives the _proof_ , it will know unequivocally that a verified path has been
established to the destination, and that information can now be exchanged reliably and
securely.
It’s important to note that this methodology ensures that the source of the request does not need to
reveal any identifying information. Only the intended destination will know “who called”, so to
speak. This is a huge improvement to protocols like IP, where by design, you have to reveal your
own address to communicate with anyone, unless you jump through a lot of hoops to hide it.
Reticulum offers initiator anonymity by design.
When using _links_ , Reticulum will automatically verify anything sent over the link, and also
automates retransmissions if parts of a message was lost along the way. Due to the caching features
of Reticulum, such a retransmission does not need to travel the entire length of an established path.
If a packet is lost on the 8th hop of a 12 hop path, it can be fetched from the last hop that received it
reliably.
## Crossing Continents
When a packet needs to travel farther than local network topology knowledge stretches, a system of
geographical or topological hinting is used to direct the packet towards a network segment with
direct knowledge of the intended destination. This functionality is currently left out of the protocol
for simplicity of testing other parts, but will be activated in a future release. For more information
on when, refer to the roadmap on the website.
## Resourceful Memory
In traditional networks, large amounts of data is rapidly exchanged with very low latency. Links of
several thousand kilometers will often only have round-trip latency in the tens of milliseconds, and
as such, traditional protocols are often designed to not store any transmitted data at intermediary
hops. If a transmission error occurs, the sending node will simply notice the lack of a packet
acknowledgement, and retransmit the packet all the way, until it hears back from the receiver that it
got the intended data.
In bandwidth-limited and high-latency conditions, such behaviour quickly causes congestion on the
network, and communications that span many hops become exceedingly expensive in terms of
bandwidth usage, due to the higher risk of some packets failing.
Reticulum alleviates this in part with it’s _path_ discovery methodology, and in part by implementing
_resource_ caching at all nodes that can support it. Network operation can be made much more
efficient by caching everything for a period of time, and given the availability of cheap memory and
storage, this is a very welcome tradeoff. A gigabyte of memory can store millions of Reticulum
packets, and since everything is encrypted by default, the storing poses very little privacy risk.
In a Reticulum network, any node that is able to do so, should cache as many packets as it’s
memory will allow for. When a packet is received, a timestamp and a hash of the packet is stored
along with the full packet itself, and it will be kept in storage until the allocated cache storage is
full, whereupon the packet that was last accessed in the cache will be deleted. If a packet is accessed
from the cache, it’s timestamp will be updated to the current time, to ensure that packets that are
used stay in the cache, and packets that are not used are dropped from memory.
Some packet types are stored in separate caching tables, that allow easier lookup for other nodes.
For example, an announce is stored in a way, that allows other nodes to request the public key for a
certain destination, and as such the network as a whole operates as a distributed key ledger.
For more details on how the caching works and is used, see the reference implementation source
code.
# Reference System Setup
This section will detail the recommended _Reference System Setup_ for Reticulum. It is important to
note that Reticulum is designed to be usable over more or less any medium that allows you to send
and receive data in a digital form, and satisfies some very low minimum requirements. The
communication channel must support at least half-duplex operation, and provide an average
throughput of around 1000 bits per second, and supports a physical layer MTU of 500 bytes. The
Reticulum software should be able to run on more or less any hardware that can provide a Python
runtime environment.
That being said, the reference setup has been outlined to provide a common platform for anyone
who wants to help in the development of Reticulum, and for everyone who wants to know a
recommended setup to get started. A reference system consists of three parts:
- **A channel access device**
Or _CAD_ , in short, provides access to the physical medium whereupon the communication
takes place, for example a radio with an integrated modem. A setup with a separate modem
connected to a radio would also be termed a “channel access device”.
- **A host device**
Some sort of computing device that can run the necessary software, communicates with the
channel access device, and provides user interaction.
- **A software stack**
The software implementing the Reticulum protocol and applications using it.
The reference setup can be considered a relatively stable platform to develop on, and also to start
building networks on. While details of the implementation might change at the current stage of
development, it is the goal to maintain hardware compatibility for as long as entirely possible, and
the current reference setup has been determined to provide a functional platform for many years
into the future. The current Reference System Setup is as follows:
- **Channel Access Device**
A data radio consisting of a LoRa radio module, and a microcontroller with open source
firmware, that can connect to host devices via USB. It operates in either the 430, 868 or 900
MHz frequency bands. More details on the exact parts and how to get/make one can be
found on the website.
- **Host device**
Any computer device running Linux and Python. A Raspberry Pi with Raspbian is
recommended.
- **Software stack**
The current Reference Implementation Release of Reticulum, running on a Debian based
operating system.
It is very important to note, that the reference channel access device **does not** use the LoRaWAN
standard, but uses a custom MAC layer on top of the plain LoRa modulation! As such, you will
need a plain LoRa radio module connected to an MCU with the correct Reticulum firmware. Full
details on how to get or make such a device is available on the website.
With the current reference setup, it should be possible to get on a Reticulum network for around 70$
even if you have none of the hardware already.
# Protocol Specifics
This chapter will detail protocol specific information that is essential to the implementation of
Reticulum, but non critical in understanding how the protocol works on a general level. It should be
treated more as a reference than as essential reading.
## Node Types
Currently Reticulum defines two node types, the _Station_ and the _Peer_. A node is a _station_ if it fixed
in one place, and if it is intended to be kept online at all times. Otherwise the node is a _peer_. This
distinction is made by the user configuring the node, and is used to determine what nodes on the
network will help forward traffic, and what nodes rely on other nodes for connectivity.
## Packet Prioritisation
_The packet prioritisation algorithms are subject to rapid change at the moment, and for now, they
are not documented here. See the reference implementation for more info on how this functionality
works._
## Path Calculation
_The path calculation algorithms are subject to rapid change at the moment, and for now, they are
not documented here. See the reference implementation for more info on how this functionality
works._
## Binary Packet Format
_The binary packet format is subject to rapid change at the moment, and for now, it is not
documented here. See the reference implementation for the specific details on this topic._