OVMS3-idf/docs/en/api-guides/mesh.rst
Darian Leung 27ffeacf07 docs/rewrite_mesh_docs
The following commit rewrites the ESP-MESH docs. This commit includes
new diagrams and a structure approach to the mesh docs. Some mesh api references
have also been modified.
2018-09-03 13:03:32 +08:00

1013 lines
No EOL
50 KiB
ReStructuredText

ESP-MESH
========
This guide provides information regarding the ESP-MESH protocol. Please see the
:doc:`MESH API Reference<../api-reference/mesh/esp_mesh>` for more information
about API usage.
.. ------------------------------- Overview -----------------------------------
Overview
--------
ESP-MESH is a networking protocol built atop the Wi-Fi protocol. ESP-MESH allows
numerous devices (henceforth referred to as nodes) spread over a large physical
area (both indoors and outdoors) to be interconnected under a single WLAN (Wireless
Local-Area Network). ESP-MESH is self-organizing and self-healing meaning the network
can be built and maintained autonomously.
The ESP-MESH guide is split into the following sections:
1. :ref:`mesh-introduction`
2. :ref:`mesh-concepts`
3. :ref:`mesh-building-a-network`
4. :ref:`mesh-managing-a-network`
5. :ref:`mesh-data-transmission`
6. :ref:`mesh-network-performance`
7. :ref:`mesh-further-notes`
.. ----------------------------- Introduction ---------------------------------
.. _mesh-introduction:
Introduction
------------
.. figure:: ../../_static/mesh-traditional-network-architecture.png
:align: center
:alt: Diagram of Traditional Network Architectures
:figclass: align-center
Traditional Wi-Fi Network Architectures
A traditional infrastructure Wi-Fi network is a point-to-multipoint network where a single
central node known as the access point (AP) is directly connected to all other
nodes known as stations. The AP is responsible for arbitrating and forwarding
transmissions between the stations. Some APs also relay transmissions to/from an
external IP network via a router. Traditional infrastructure Wi-Fi networks suffer the
disadvantage of limited coverage area due to the requirement that every station
must be in range to directly connect with the AP. Furthermore, traditional Wi-Fi
networks are susceptible to overloading as the maximum number of stations permitted
in the network is limited by the capacity of the AP.
.. figure:: ../../_static/mesh-esp-mesh-network-architecture.png
:align: center
:alt: Diagram of ESP-MESH Network Architecture
:figclass: align-center
ESP-MESH Network Architecture
ESP-MESH differs from traditional infrastructure Wi-Fi networks in that nodes are not required
to connect to a central node. Instead, nodes are permitted to connect with
neighboring nodes. Nodes are mutually responsible for relaying each others
transmissions. This allows an ESP-MESH network to have much greater coverage area
as nodes can still achieve interconnectivity without needing to be in range of
the central node. Likewise, ESP-MESH is also less susceptible to overloading as
the number of nodes permitted on the network is no longer limited by a single
central node.
.. -------------------------- ESP-MESH Concepts -------------------------------
.. _mesh-concepts:
ESP-MESH Concepts
-----------------
Terminology
^^^^^^^^^^^
+--------------------------+----------------------------------------------------------------+
| Term | Description |
+==========================+================================================================+
| Node | Any device that **is** or **can be** part of an ESP-MESH |
| | network |
+--------------------------+----------------------------------------------------------------+
| Root Node | The top node in the network |
+--------------------------+----------------------------------------------------------------+
| Child Node | A node X is a child node when it is connected to another node |
| | Y where the connection makes node X more distant from the root |
| | node than node Y (in terms of number of connections). |
+--------------------------+----------------------------------------------------------------+
| Parent Node | The converse notion of a child node |
+--------------------------+----------------------------------------------------------------+
| Sub-Child Node | Any node reachable by repeated proceeding from parent to child |
+--------------------------+----------------------------------------------------------------+
| Sibling Nodes | Nodes that share the same parent node |
+--------------------------+----------------------------------------------------------------+
| Connection | A traditional Wi-Fi association between an AP and a station. |
| | A node in ESP-MESH will use its station interface to associate |
| | with the softAP interface of another node, thus forming a |
| | connection. The connection process includes the authentication |
| | and association processes in Wi-Fi. |
+--------------------------+----------------------------------------------------------------+
| Upstream Connection | The connection from a node to its parent node |
+--------------------------+----------------------------------------------------------------+
| Downstream Connection | The connection from a node to one of its child nodes |
+--------------------------+----------------------------------------------------------------+
| Wireless Hop | The portion of the path between source and destination nodes |
| | that corresponds to a single wireless connection. A data |
| | packet that traverses a single connection is known as |
| | **single-hop** whereas traversing multiple connections is |
| | known as **multi-hop**. |
+--------------------------+----------------------------------------------------------------+
| Subnetwork | A subnetwork is subdivision of an ESP-MESH network which |
| | consists of a node and all of its descendant nodes. Therefore |
| | the subnetwork of the root node consists of all nodes in an |
| | ESP-MESH network. |
+--------------------------+----------------------------------------------------------------+
| MAC Address | Media Access Control Address used to uniquely identify each |
| | node or router within an ESP-MESH network. |
+--------------------------+----------------------------------------------------------------+
| DS | Distribution System (External IP Network) |
+--------------------------+----------------------------------------------------------------+
Tree Topology
^^^^^^^^^^^^^
ESP-MESH is built atop the infrastructure Wi-Fi protocol and can be thought of
as a networking protocol that combines many individual Wi-Fi networks into a single
WLAN. In Wi-Fi, stations are limited to a single connection with an AP (upstream
connection) at any time, whilst an AP can be simultaneously connected to multiple
stations (downstream connections). However ESP-MESH allows nodes to simultaneously
act as a station and an AP. Therefore a node in ESP-MESH can have **multiple downstream
connections using its softAP interface**, whilst simultaneously having **a single
upstream connection using its station interface**. This naturally results in a
tree network topology with a parent-child hierarchy consisting of multiple layers.
.. figure:: ../../_static/mesh-tree-topology.png
:align: center
:alt: Diagram of ESP-MESH Tree Topology
:figclass: align-center
ESP-MESH Tree Topology
ESP-MESH is a multiple hop (multi-hop) network meaning nodes can transmit packets
to other nodes in the network through one or more wireless hops. Therefore, nodes
in ESP-MESH not only transmit their own packets, but simultaneously serve as relays
for other nodes. Provided that a path exists between any two nodes on the physical
layer (via one or more wireless hops), any pair of nodes within an ESP-MESH network
can communicate.
.. note::
The size (total number of nodes) in an ESP-MESH network is dependent on the
maximum number of layers permitted in the network, and the maximum number of
downstream connections each node can have. Both of these variables can be
configured to limit the size of the network.
Node Types
^^^^^^^^^^
.. figure:: ../../_static/mesh-node-types.png
:align: center
:alt: Diagram of ESP-MESH Node Types
:figclass: align-center
ESP-MESH Node Types
**Root Node:** The root node is the top node in the network and serves as the only
interface between the ESP-MESH network and an external IP network. The root node
is connected to a conventional Wi-Fi router and relays packets to/from the external
IP network to nodes within the ESP-MESH network. **There can only be one root node
within an ESP-MESH network** and the root node's upstream connection may only be
with the router. Referring to the diagram above, node A is the root node of the
network.
**Leaf Nodes:** A leaf node is a node that is not permitted to have any child nodes
(no downstream connections). Therefore a leaf node can only transmit or receive
its own packets, but cannot forward the packets of other nodes. If a node is situated
on the network's maximum permitted layer, it will be assigned as a leaf node. This
prevents the node from forming any downstream connections thus ensuring the network
does not add an extra layer. Some nodes without a softAP interface (station only)
will also be assigned as leaf nodes due to the requirement of a softAP interface
for any downstream connections. Referring to the diagram above, nodes L/M/N are
situated on the networks maximum permitted layer hence have been assigned as leaf nodes .
**Intermediate Parent Nodes:** Connected nodes that are neither the root node or
a leaf node are intermediate parent nodes. An intermediate parent node must have
a single upstream connection (a single parent node), but can have zero to multiple
downstream connections (zero to multiple child nodes). Therefore an intermediate
parent node can transmit and receive packets, but also forward packets sent from its
upstream and downstream connections. Referring to the diagram above, nodes
B to J are intermediate parent nodes. **Intermediate parent nodes without downstream
connections such as nodes E/F/G/I/J are not equivalent to leaf nodes** as they
are still permitted to form downstream connections in the future.
**Idle Nodes:** Nodes that have yet to join the network are assigned as idle nodes.
Idle nodes will attempt to form an upstream connection with an intermediate parent
node or attempt to become the root node under the correct circumstances (see
`Automatic Root Node Selection`_). Referring to the diagram above, nodes K and O
are idle nodes.
Beacon Frames & RSSI Thresholding
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Every node in ESP-MESH that is able to form downstream connections (i.e. has a
softAP interface) will periodically transmit Wi-Fi beacon frames. A node uses
beacon frames to allow other nodes to detect its presence and know of its status.
Idle nodes will listen for beacon frames to generate a list of potential parent nodes,
one of which the idle node will form an upstream connection with. ESP-MESH uses
the Vendor Information Element to store metadata such as:
- Node Type (Root, Intermediate Parent, Leaf, Idle)
- Current layer of Node
- Maximum number of layers permitted in the network
- Current number of child nodes
- Maximum number of downstream connections to accept
The signal strength of a potential upstream connection is represented by RSSI
(Received Signal Strength Indication) of the beacon frames of the potential parent
node. To prevent nodes from forming a weak upstream connection, ESP-MESH implements
an RSSI threshold mechanism for beacon frames. If a node detects a beacon frame
with an RSSI below a preconfigured threshold, the transmitting node will be
disregarded when forming an upstream connection.
.. figure:: ../../_static/mesh-beacon-frame-rssi.png
:align: center
:alt: Diagram of the Effects of RSSI Thresholding
:figclass: align-center
Effects of RSSI Thresholding
**Panel A** of the illustration above demonstrates how the RSSI threshold affects
the number of parent node candidates an idle node has.
**Panel B** of the illustration above demonstrates how an RF shielding object can
lower the RSSI of a potential parent node. Due to the RF shielding object, the
area in which the RSSI of node X is above the threshold is significantly reduced.
This causes the idle node to disregard node X even though node X is physically
adjacent. The idle node will instead form an upstream connection with the physically
distant node Y due to a stronger RSSI.
.. note::
Nodes technically still receive all beacon frames on the MAC layer. The RSSI
threshold is an ESP-MESH feature that simply filters out all received beacon
frames that are below the preconfigured threshold.
Preferred Parent Node
^^^^^^^^^^^^^^^^^^^^^
When an idle node has multiple parent nodes candidates (potential parent nodes),
the idle node will form an upstream connection with the **preferred parent node**.
The preferred parent node is determined based on the following criteria:
- Which layer the parent node candidate is situated on
- The number of downstream connections (child nodes) the parent node candidate currently has
The selection of the preferred parent node will always prioritize the parent node
candidate on the shallowest layer of the network (including the root node). This
helps minimize the total number of layers in an ESP-MESH network when upstream
connections are formed. For example, given a second layer node and a third layer
node, the second layer node will always be preferred.
If there are multiple parent node candidates within the same layer, the parent
node candidate with the least child nodes will be preferred. This criteria has
the effect of balancing the number of downstream connections amongst nodes of
the same layer.
.. figure:: ../../_static/mesh-preferred-parent-node.png
:align: center
:alt: Diagram of Preferred Parent Node Selection
:figclass: align-center
Preferred Parent Node Selection
**Panel A** of the illustration above demonstrates an example of how the idle
node G selects a preferred parent node given the five parent node candidates
B/C/D/E/F. Nodes on the shallowest layer are preferred, hence nodes B/C are
prioritized since they are second layer nodes whereas nodes D/E/F are on the
third layer. Node C is selected as the preferred parent node due it having fewer
downstream connections (fewer child nodes) compared to node B.
**Panel B** of the illustration above demonstrates the case where the root node
is within range of the idle node G. In other words, the root node's beacon frames
are above the RSSI threshold when received by node G. The root node is always the
shallowest node in an ESP-MESH network hence is always the preferred parent node
given multiple parent node candidates.
.. note::
Users may also define their own algorithm for selecting a preferred parent
node, or force a node to only connect with a specific parent node (see the
:example:`Mesh Manual Networking Example<mesh/manual_networking>`).
Routing Tables
^^^^^^^^^^^^^^
Each node within an ESP-MESH network will maintain its individual routing table
used to correctly route ESP-MESH packets (see `ESP-MESH Packet`_) to the correct
destination node. The routing table of a particular node will **consist of the
MAC addresses of all nodes within the particular node's subnetwork** (including
the MAC address of the particular node itself). Each routing table is internally
partitioned into multiple subtables with each subtable corresponding to the
subnetwork of each child node.
.. figure:: ../../_static/mesh-routing-tables-example.png
:align: center
:alt: Diagram of ESP-MESH Routing Tables Example
:figclass: align-center
ESP-MESH Routing Tables Example
Using the diagram above as an example, the routing table of node B would consist
of the MAC addresses of nodes B to I (i.e. equivalent to the subnetwork of node
B). Node B's routing table is internally partitioned into two subtables containing
of nodes C to F and nodes G to I (i.e. equivalent to the subnetworks of nodes C
and G respectively).
**ESP-MESH utilizes routing tables to determine whether an ESP-MESH packet should
be forwarded upstream or downstream based on the following rules.**
**1.** If the packet's destination MAC address is within the current node's routing
table and is not the current node, select the subtable that contains the destination
MAC address and forward the data packet downstream to the child node corresponding
to the subtable.
**2.** If the destination MAC address is not within the current node's routing table,
forward the data packet upstream to the current node's parent node. Doing so repeatedly
will result in the packet arriving at the root node where the routing table should
contain all nodes within the network.
.. note::
Users can call :cpp:func:`esp_mesh_get_routing_table` to obtain a node's routing
table, or :cpp:func:`esp_mesh_get_routing_table_size` to obtain the size of a
node's routing table.
:cpp:func:`esp_mesh_get_subnet_nodes_list` can be used to obtain the corresponding
subtable of a specific child node. Likewise, :cpp:func:`esp_mesh_get_subnet_nodes_num`
can be used to obtain the size of the subtable.
.. ------------------------ Building a Mesh Network ---------------------------
.. _mesh-building-a-network:
Building a Network
------------------
General Process
^^^^^^^^^^^^^^^
.. warning::
Before the ESP-MESH network building process can begin, certain parts of the
configuration must be uniform across each node in the network (see
:cpp:type:`mesh_cfg_t`). Each node must be configured with **the same Mesh
Network ID, router configuration, and softAP configuration**.
An ESP-MESH network building process involves selecting a root node, then forming
downstream connections layer by layer until all nodes have joined the network. The exact
layout of the network can be dependent on factors such as root node selection,
parent node selection, and asynchronous power-on reset. However, the ESP-MESH network
building process can be generalized into the following steps:
.. figure:: ../../_static/mesh-network-building.png
:align: center
:alt: Diagram of ESP-MESH Network Building Process
:figclass: align-center
ESP-MESH Network Building Process
1. Root Node Selection
""""""""""""""""""""""
The root node can be designated during configuration (see section on
`User Designated Root Node`_), or dynamically elected based on the signal strength
between each node and the router (see `Automatic Root Node Selection`_). Once selected,
the root node will connect with the router and begin allowing downstream connections
to form. Referring to the figure above, node A is selected to be the root node
hence node A forms an upstream connection with the router.
2. Second Layer Formation
"""""""""""""""""""""""""
Once the root node has connected to the router, idle nodes in range of the root
node will begin connecting with the root node thereby forming the second layer
of the network. Once connected, the second layer nodes become intermediate parent
nodes (assuming maximum permitted layers > 2) hence the next layer to form. Referring
to the figure above, nodes B to D are in range of the root node. Therefore nodes
B to D form upstream connections with the root node and become intermediate parent
nodes.
3. Formation of remaining layers
""""""""""""""""""""""""""""""""
The remaining idle nodes will connect with intermediate parent nodes within range
thereby forming a new layer in the network. Once connected, the idles nodes become
intermediate parent node or leaf nodes depending on the networks maximum permitted
layers. This step is repeated until there are no more idle nodes within the network
or until the maximum permitted layer of the network has been reached. Referring to
the figure above, nodes E/F/G connect with nodes B/C/D respectively and become
intermediate parent nodes themselves.
4. Limiting Tree Depth
""""""""""""""""""""""
To prevent the network from exceeding the maximum permitted number of layers, nodes
on the maximum layer will automatically become leaf nodes once connected. This
prevents any other idle node from connecting with the leaf node thereby prevent
a new layer form forming. However if an idle node has no other potential parent
node, it will remain idle indefinitely. Referring to the figure above, the network's
maximum permitted layers is set to four. Therefore when node H connects, it becomes
a leaf node to prevent any downstream connections from forming.
Automatic Root Node Selection
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
The automatic selection of a root node involves an election process amongst
all idle nodes based on their signal strengths with the router. Each idle node
will transmit their MAC addresses and router RSSI values via Wi-Fi beacon frames.
**The MAC address is used to uniquely identify each node in the network** whilst
the **router RSSI** is used to indicate a node's signal strength with reference to
the router.
Each node will then simultaneously scan for the beacon frames from other idle nodes.
If a node detects a beacon frame with a stronger router RSSI, the node will begin
transmitting the contents of that beacon frame (i.e. voting for the node with
the stronger router RSSI). The process of transmission and scanning will repeat
for a preconfigured minimum number of iterations (10 iterations by default) and result
in the beacon frame with the strongest router RSSI being propagated throughout
the network.
After all iterations, each node will individually check for its **vote percentage**
(``number of votes/number of nodes participating in election``) to determine if it should become the
root node. **If a node has a vote percentage larger than a preconfigured threshold
(90% by default), the node will become a root node**.
The following diagram demonstrates how an ESP-MESH network is built when the root
node is automatically selected.
.. figure:: ../../_static/mesh-root-node-election-example.png
:align: center
:alt: Diagram of Root Node Election Process Example
:figclass: align-center
Root Node Election Example
**1.** On power-on reset, each node begins transmitting beacon frames consisting
of their own MAC addresses and their router RSSIs.
**2.** Over multiple iterations of transmission and scanning, the beacon frame
with the strongest router RSSI is propagated throughout the network. Node C has
the strongest router RSSI (-10db) hence its beacon frame is propagated throughout the
network. All nodes participating in the election vote for node C thus giving node
C a vote percentage of 100%. Therefore node C becomes a root node and connects with
the router.
**3.** Once Node C has connected with the router, nodes A/B/D/E connect
with node C as it is the preferred parent node (i.e. the shallowest node). Nodes
A/B/D/E form the second layer of the network.
**4.** Node F and G connect with nodes D and E respectively and the network building
process is complete.
.. note::
The minimum number of iterations for the election process can be configured
using :cpp:func:`esp_mesh_set_attempts`. Users should adjust the number
of iterations based on the number of nodes within the network (i.e. the larger
the network the larger number of scan iterations required).
.. warning::
**Vote percentage threshold** can also be configured using
:cpp:func:`esp_mesh_set_vote_percentage`. Setting a low vote percentage
threshold **can result in two or more nodes becoming root nodes** within the
same ESP-MESH network leading to the building of multiple networks. If
such is the case, ESP-MESH has internal mechanisms to autonomously resolve
the **root node conflict**. The networks of the multiple root nodes will be
combined into a single network with a single root node. However, root node
conflicts where two or more root nodes have the same router SSID but different
router BSSID are not handled.
User Designated Root Node
^^^^^^^^^^^^^^^^^^^^^^^^^
The root node can also be designated by user which will entail the designated root node
to directly connect with the router and forgo the election process. When a root
node is designated, all other nodes within the network must also forgo the election
process to prevent the occurrence of a root node conflict. The following diagram demonstrates
how an ESP-MESH network is built when the root node is designated by the user.
.. figure:: ../../_static/mesh-root-node-designated-example.png
:align: center
:alt: Diagram of Root Node Designation Process Example
:figclass: align-center
Root Node Designation Example (Root Node = A, Max Layers = 4)
**1.** Node A is designated the root node by the user therefore directly connects
with the router. All other nodes forgo the election process.
**2.** Nodes C/D connect with node A as their preferred parent node. Both
nodes form the second layer of the network.
**3.** Likewise, nodes B/E connect with node C, and node F connects with
node D. Nodes B/E/F form the third layer of the network.
**4.** Node G connects with node E, forming the fourth layer of the network.
However the maximum permitted number of layers in this network is configured as
four, therefore node G becomes a leaf node to prevent any new layers from forming.
.. note::
When designating a root node, the root node should call :cpp:func:`esp_mesh_set_parent`
in order to directly connect with the router. Likewise, all other nodes should
call :cpp:func:`esp_mesh_fix_root` to forgo the election process.
Parent Node Selection
^^^^^^^^^^^^^^^^^^^^^
By default, ESP-MESH is self organizing meaning that each node will autonomously
select which potential parent node to form an upstream connection with. The autonomously
selected parent node is known as the preferred parent node. The criteria used for
selecting the preferred parent node is designed to reduce the number of layers in
the ESP-MESH network and to balance the number of downstream connections between
potential parent nodes (see section on `Preferred Parent Node`_).
However ESP-MESH also allows users to disable self-organizing behavior which will
allow users to define their own criteria for parent node selection, or to configure
nodes to have designated parent nodes (see the
:example:`Mesh Manual Networking Example<mesh/manual_networking>`).
Asynchronous Power-on Reset
^^^^^^^^^^^^^^^^^^^^^^^^^^^
ESP-MESH network building can be affected by the order in which nodes power-on.
If certain nodes within the network power-on asynchronously (i.e. separated by
several minutes), **the final structure of the network could differ from the ideal
case where all nodes are powered on synchronously**. Nodes that are delayed in
powering on will adhere to the following rules:
**Rule 1:** If a root node already exists in the network, the delayed node will
not attempt to elect a new root node, even if it has a stronger RSSI with the router.
The delayed node will instead join the network like any other idle node by connecting
with a preferred parent node. If the delayed node is the designated root node,
all other nodes in the network will remain idle until the delayed node powers-on.
**Rule 2:** If a delayed node forms an upstream connection and becomes an intermediate
parent node, it may also become the new preferred parent of other nodes (i.e. being
a shallower node). This will cause the other nodes to switch their upstream connections
to connect with the delayed node (see `Parent Node Switching`_).
**Rule 3:** If an idle node has a designated parent node which is delayed in powering-on,
the idle node will not attempt to form any upstream connections in the absence of
its designated parent node. The idle node will remain idle indefinitely until its
designated parent node powers-on.
The following example demonstrates the effects of asynchronous power-on with regards
to network building.
.. figure:: ../../_static/mesh-asynchronous-power-on-example.png
:align: center
:alt: Diagram of Asynchronous Power On Example
:figclass: align-center
Network Building with Asynchronous Power On Example
**1.** Nodes A/C/D/F/G/H are powered-on synchronously and begin the root
node election process by broadcasting their MAC addresses and router RSSIs. Node
A is elected as the root node as it has the strongest RSSI.
**2.** Once node A becomes the root node, the remaining nodes begin forming upstream
connections layer by layer with their preferred parent nodes. The result is a network
with five layers.
**3.** Node B/E are delayed in powering-on but neither attempt to become the root
node even though they have stronger router RSSIs (-20db and -10db) compared to
node A. Instead both delayed nodes form upstream connections with their preferred
parent nodes A and C respectively. Both Nodes B/E become intermediate parent nodes
after connecting.
**4.** Nodes D/G switch their upstream connections as node B is the new preferred
parent node due to it being on a shallower layer (second layer node). Due to the
switch, the resultant network has three layers instead of the original five layers.
**Synchronous Power-On:** Had all nodes powered-on synchronously, node E would
have become the root node as it has the strongest router RSSI (-10db). This
would result in a significantly different network layout compared to the network
formed under the conditions of asynchronous power-on. **However the synchronous
power-on network layout can still be reached if the user manually switches the
root node** (see :cpp:func:`esp_mesh_waive_root`).
.. note::
Differences in parent node selection caused by asynchronous power-on are
autonomously corrected for to some extent in ESP-MESH (see `Parent Node Switching`_)
Loop-back Avoidance, Detection, and Handling
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
A loop-back is the situation where a particular node forms an upstream connection
with one of its descendant nodes (a node within the particular node's subnetwork).
This results in a circular connection path thereby breaking the tree topology.
ESP-MESH prevents loop-back during parent selection by excluding nodes already
present in the selecting node's routing table (see `Routing Tables`_) thus prevents
a particular node from attempting to connect to any node within its subnetwork.
In the event that a loop-back occurs, ESP-MESH utilizes a path verification
mechanism and energy transfer mechanism to detect the loop-back occurrence. The
parent node of the upstream connection that caused the loop-back will then inform
the child node of the loop-back and initiate a disconnection.
.. -------------------------- Network Management ------------------------------
.. _mesh-managing-a-network:
Managing a Network
------------------
**ESP-MESH is a self healing network meaning it can detect and correct for failures
in network routing**. Failures occur when a parent node with one or more child
nodes breaks down, or when the connection between a parent node and its child nodes
becomes unstable. Child nodes in ESP-MESH will autonomously select a new parent
node and form an upstream connection with it to maintain network interconnectivity.
ESP-MESH can handle both Root Node Failures and Intermediate Parent Node Failures.
Root Node Failure
^^^^^^^^^^^^^^^^^
If the root node breaks down, the nodes connected with it (second layer nodes)
will promptly detect the failure of the root node. The second layer nodes
will initially attempt to reconnect with the root node. However after multiple failed
attempts, the second layer nodes will initialize a new round of root node election.
**The second layer node with the strongest router RSSI will be elected as the new
root node** whilst the remaining second layer nodes will form an upstream connection
with the new root node (or a neighboring parent node if not in range).
If the root node and multiple downstream layers simultaneously break down (e.g.
root node, second layer, and third layer), the shallowest layer that is still
functioning will initialize the root node election. The following example illustrates
an example of self healing from a root node break down.
.. figure:: ../../_static/mesh-root-node-failure.png
:align: center
:alt: Diagram of Self Healing From Root Node Failure
:figclass: align-center
Self Healing From Root Node Failure
**1.** Node C is the root node of the network. Nodes A/B/D/E are second layer
nodes connected to node C.
**2.** Node C breaks down. After multiple failed attempts to reconnect, the second
layer nodes begin the election process by broadcasting their router RSSIs. Node
B has the strongest router RSSI.
**3.** Node B is elected as the root node and begins accepting downstream
connections. The remaining second layer nodes A/D/E form upstream connections with
node B thus the network is healed and can continue operating normally.
.. note::
If a designated root node breaks down, the remaining nodes **will not autonomously
attempt to elect a new root node** as an election process will never be attempted
whilst a designated root node is used.
Intermediate Parent Node Failure
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
If an intermediate parent node breaks down, the disconnected child nodes will initially
attempt to reconnect with the parent node. After multiple failed attempts to reconnect,
each child node will begin to scan for potential parent nodes (see
`Beacon Frames & RSSI Thresholding`_).
If other potential parent nodes are available, each child node will individually
select a new preferred parent node (see `Preferred Parent Node`_) and form an upstream
connection with it. If there are no other potential parent nodes for a particular
child node, it will remain idle indefinitely.
The following diagram illustrates an example of self healing from an Intermediate
Parent Node break down.
.. figure:: ../../_static/mesh-parent-node-failure.png
:align: center
:alt: Diagram of Self Healing From Intermediate Parent Node Failure
:figclass: align-center
Self Healing From Intermediate Parent Node Failure
**1.** The following branch of the network consists of nodes A to G.
**2.** Node C breaks down. Nodes F/G detect the break down and attempt to
reconnect with node C. After multiple failed attempts to reconnect, nodes F/G begin
to select a new preferred parent node.
**3.** Node G is out of range from any other parent node hence remains idle for
the time being. Node F is in range of nodes B/E, however node B is selected as
it is the shallower node. Node F becomes an intermediate parent node after
connecting with Node B thus node G can connect with node F. The network is healed,
however the network routing as been affected and an extra layer has been added.
.. note::
If a child node has a designated parent node that breaks down, the child node
will make no attempt to connect with a new parent node. The child node will
remain idle indefinitely.
Root Node Switching
^^^^^^^^^^^^^^^^^^^
ESP-MESH does not automatically switch the root node unless the root node breaks down. Even
if the root node's router RSSI degrades to the point of disconnection, the root node
will remain unchanged. Root node switching is the act of explicitly starting
a new election such that a node with a stronger router RSSI will be elected as the
new root node. This can be a useful method of adapting to degrading root node performance.
To trigger a root node switch, the current root node must explicitly call :cpp:func:`esp_mesh_waive_root`
to trigger a new election. The current root node will signal all nodes within
the network to begin transmitting and scanning for beacon frames (see `Automatic
Root Node Selection`_) **whilst remaining connected to the network (i.e. not idle)**.
If another node receives more votes than the current root node, a root node switch
will be initiated. **The root node will remain unchanged otherwise**.
A newly elected root node sends a **switch request** to the current root node
which in turn will respond with an acknowledgment signifying both nodes are ready to
switch. Once the acknowledgment is received, the newly elected root node will
disconnect from its parent and promptly form an upstream connection with the router
thereby becoming the new root node of the network. The previous root node will
disconnect from the router **whilst maintaining all of its downstream connections**
and enter the idle state. The previous root node will then begin scanning for
potential parent nodes and selecting a preferred parent.
The following diagram illustrates an example of a root node switch.
.. figure:: ../../_static/mesh-root-node-switch-example.png
:align: center
:alt: Diagram of Root Node Switch Example
:figclass: align-center
Root Node Switch Example
**1.** Node C is the current root node but has degraded signal strength with the
router (-85db). The node C triggers a new election and all nodes begin transmitting
and scanning for beacon frames **whilst still being connected**.
**2.** After multiple rounds of transmission and scanning, node B is elected as
the new root node. Node B sends node C a **switch request** and node C responds
with an acknowledgment.
**3.** Node B disconnects from its parent and connects with the router becoming
the networks new root node. Node C disconnects from the router, enters the idle
state, and begins scanning for and selecting a new preferred parent node. **Node
C maintains all its downstream connections throughout this process**.
**4.** Node C selects node B as its preferred parent node, forms an upstream
connection, and becomes a second layer node. The network layout is similar after
the switch as node C still maintains the same subnetwork. However each node in
node C's subnetwork has been placed one layer deeper as a result of the switch.
`Parent Node Switching`_ may adjust the network layout afterwards if any nodes have
a new preferred parent node as a result of the root node switch.
.. note::
Root node switching must require an election hence is only supported when using
a self-organized ESP-MESH network. In other words, root node switching cannot
occur if a designated root node is used.
Parent Node Switching
^^^^^^^^^^^^^^^^^^^^^
Parent Node Switching entails a child node switching its upstream connection to
another parent node of a shallower layer. **Parent Node Switching occurs autonomously**
meaning that a child node will change its upstream connection automatically if a
potential parent node of a shallower layer becomes available (i.e. due to a
`Asynchronous Power-on Reset`_).
All potential parent nodes periodically transmit beacon frames (see `Beacon Frames
& RSSI Thresholding`_) allowing for a child node to scan for the availability of
a shallower parent node. Due to parent node switching, a self-organized ESP-MESH
network can dynamically adjust its network layout to ensure each connection has a good
RSSI and that the number of layers in the network is minimized.
.. --------------------------- Data Transmission ------------------------------
.. _mesh-data-transmission:
Data Transmission
-----------------
ESP-MESH Packet
^^^^^^^^^^^^^^^
ESP-MESH network data transmissions use ESP-MESH packets. ESP-MESH packets
are **entirely contained within the frame body of a Wi-Fi data frame**. A multi-hop
data transmission in an ESP-MESH network will involve a single ESP-MESH packet
being carried over each wireless hop by a different Wi-Fi data frame.
The following diagram shows the structure of an ESP-MESH packet and its relation
with a Wi-Fi data frame.
.. figure:: ../../_static/mesh-packet.png
:align: center
:alt: Diagram of ESP-MESH Packet
:figclass: align-center
ESP-MESH Packet
**The header** of an ESP-MESH packet contains the MAC addresses of the source and
destination nodes. The options field contains information pertaining to the special
types of ESP-MESH packets such as a group transmission or a packet originating
from the external IP network (see :c:macro:`MESH_OPT_SEND_GROUP` and
:c:macro:`MESH_OPT_RECV_DS_ADDR`).
**The payload** of an ESP-MESH packet contains the actual application data. This
data can be raw binary data, or encoded under an application layer protocol such
as HTTP, MQTT, and JSON (see :cpp:type:`mesh_proto_t`).
.. note::
When sending an ESP-MESH packet to the external IP network, the destination
address field of the header will contain the IP address and port of the target server
rather than the MAC address of a node (see :cpp:type:`mesh_addr_t`). Furthermore
the root node will handle the formation of the outgoing TCP/IP packet.
Group Control & Multicasting
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Multicasting is a feature that allows a single ESP-MESH packet to be transmitted
simultaneously to multiple nodes within the network. Multicasting in ESP-MESH can
be achieved by either specifying a list of target nodes, or specifying a preconfigured
group of nodes. Both methods of multicasting are called via :cpp:func:`esp_mesh_send`.
To multicast by specifying a list of target nodes, users must first set the ESP-MESH
packet's destination address to the **Multicast-Group Address** (``01:00:5E:xx:xx:xx``).
This signifies that the ESP-MESH packet is a multicast packet with a group of addresses,
and that the address should be obtained from the header options. Users must then
list the MAC addresses of the target nodes as options (see :cpp:type:`mesh_opt_t`
and :c:macro:`MESH_OPT_SEND_GROUP`). This method of multicasting requires no prior
setup but can incur a large amount of overhead data as each target node's MAC address
must be listed in the options field of the header.
Multicasting by group allows a ESP-MESH packet to be transmitted to a preconfigured
group of nodes. Each grouping is identified by a unique ID, and a node can be placed
into a group via :cpp:func:`esp_mesh_set_group_id`. Multicasting to a group involves
setting the destination address of the ESP-MESH packet to the target group ID.
Furthermore, the :c:macro:`MESH_DATA_GROUP` flag must set. Using groups to multicast
incurs less overhead, but requires nodes to previously added into groups.
.. note::
During a multicast, all nodes within the network still receive the ESP-MESH
packet on the MAC layer. However, nodes not included in the MAC address list
or the target group will simply filter out the packet.
Broadcasting
^^^^^^^^^^^^
Broadcasting is a feature that allows a single ESP-MESH packet to be transmitted
simultaneously to all nodes within the network. Each node essentially forwards
a broadcast packet to all of its upstream and downstream connections such that
the packet propagates throughout the network as quickly as possible. However,
ESP-MESH utilizes the following methods to avoid wasting bandwidth during a broadcast.
**1.** When an intermediate parent node receives a broadcast packet from its parent,
it will forward the packet to each of its child nodes whilst storing a copy of the
packet for itself.
**2.** When an intermediate parent node is the source node of the broadcast, it
will transmit the broadcast packet upstream to is parent node and downstream to
each of its child nodes.
**3.** When an intermediate parent node receives a broadcast packet from one of its
child nodes, it will forward the packet to its parent node and each of its remaining
child nodes whilst storing a copy of the packet for itself.
**4.** When a leaf node is the source node of a broadcast, it will directly
transmit the packet to its parent node.
**5.** When the root node is the source node of a broadcast, the root node will transmit
the packet to all of its child nodes.
**6.** When the root node receives a broadcast packet from one of its child nodes, it
will forward the packet to each of its remaining child nodes whilst storing a copy
of the packet for itself.
**7.** When a node receives a broadcast packet with a source address matching its
own MAC address, the node will discard the broadcast packet.
**8.** When an intermediate parent node receives a broadcast packet from its parent
node which was originally transmitted from one of its child nodes, it will discard
the broadcast packet
Upstream Flow Control
^^^^^^^^^^^^^^^^^^^^^
ESP-MESH relies on parent nodes to control the upstream data flow of their immediate
child nodes. To prevent a parent node's message buffer from overflowing due to an overload
of upstream transmissions, a parent node will allocate a quota for upstream transmissions
known as a **receiving window** for each of its child nodes. **Each child node must
apply for a receiving window before it is permitted to transmit upstream**. The size
of a receiving window can be dynamically adjusted. An upstream transmission from
a child node to the parent node consists of the following steps:
**1.** Before each transmission, the child node sends a window request to its parent
node. The window request consists of a sequence number which corresponds to the child
node's data packet that is pending transmission.
**2.** The parent node receives the window request and compares the sequence number
with the sequence number of the previous packet sent by the child node. The comparison
is used to calculate the size of the receiving window which is transmitted back
to the child node.
**3.** The child node transmits the data packet in accordance with the window size
specified by the parent node. If the child node depletes its receiving window, it
must obtain another receiving windows by sending a request before it is permitted
to continue transmitting.
.. note::
ESP-MESH does not support any downstream flow control.
.. warning::
Due to `Parent Node Switching`_, packet loss may occur during upstream
transmissions.
Due to the fact that the root node acts as the sole interface to an external IP
network, it is critical that downstream nodes are aware of the root node's connection
status with the external IP network. Failing to do so can lead to nodes attempting
to pass data upstream to the root node whilst it is disconnected from the IP network.
This results in unnecessary transmissions and packet loss. ESP-MESH address this
issue by providing a mechanism to stabilize the throughput of outgoing data based
on the connection status between the root node and the external IP network. The root
node can broadcast its external IP network connection status to all other nodes
by calling :cpp:func:`esp_mesh_post_toDS_state`.
Bi-Directional Data Stream
^^^^^^^^^^^^^^^^^^^^^^^^^^
The following diagram illustrates the various network layers involved in an ESP-MESH
Bidirectional Data Stream.
.. figure:: ../../_static/mesh-bidirectional-data-stream.png
:align: center
:alt: Diagram of ESP-MESH Bidirectional Data Stream
:figclass: align-center
ESP-MESH Bidirectional Data Stream
Due to the use of `Routing Tables`_, **ESP-MESH is able to handle pack forwarding
entirely on the mesh layer**. A TCP/IP layer is only required on the root
node when it transmits/receives a packet to/from an external IP network.
.. ------------------------------ Performance ---------------------------------
.. _mesh-network-performance:
Performance
-----------
The performance of an ESP-MESH network can be evaluated based on multiple metrics
such as the following:
**Network Building Time:** The amount of time taken to build an ESP-MESH network from
scratch.
**Healing Time:** The amount of time taken for the network to detect a node break
down and carry out appropriate actions to heal the network (such as generating a
new root node or forming new connections).
**Per-hop latency:** The latency of data transmission over one wireless hop. In
other words, the time taken to transmit a data packet from a parent node to a
child node or vice versa.
**Network Node Capacity:** The total number of nodes the ESP-MESH network can simultaneously
support. This number is determined by the maximum number of downstream connections
a node can accept and the maximum number of layers permissible in the network.
The following table lists the common performance figures of an ESP-MESH network.
However users should note that performance numbers can vary greatly between
installations based on network configuration and operating environment.
+-------------------------+------------------------------------+
| Function | Description |
+=========================+====================================+
|Networking Building Time | < 60 seconds |
+-------------------------+------------------------------------+
|Healing time | Root Node Break Down: < 10 seconds |
| | |
| | Child Node Break Down: < 5 seconds |
+-------------------------+------------------------------------+
|Per-hop latency | 10 to 30 milliseconds |
+-------------------------+------------------------------------+
.. note::
The following test conditions were used to generate the performance figures
above.
- Number of test devices: **100**
- Maximum Downstream Connections to Accept: **6**
- Maximum Permissible Layers: **6**
.. note::
Throughput depends on packet error rate and hop count.
.. note::
The throughput of root node's access to the external IP network is directly
affected by the number of nodes in the ESP-MESH network and the bandwidth of
the router.
.. ----------------------------- Further Notes --------------------------------
.. _mesh-further-notes:
Further Notes
-------------
- Data transmission uses Wi-Fi WPA2-PSK encryption
- Mesh networking IE uses AES encryption
Router and internet icon made by `Smashicons <https://smashicons.com>`_ from `www.flaticon.com <https://smashicons.com>`_