6lo Working Group G. Li
Internet-Draft D. Lou
Intended status: Informational L. Iannone
Expires: 27 April 2023 Huawei
24 October 2022
Reliability Considerations of Path-Aware Semantic Addressing
draft-li-6lo-pasa-reliability-00
Abstract
Path-Aware Semantic Address (PASA
[I-D.li-6lo-path-aware-semantic-addressing]), proposes to
algorithmically assign short addresses to nodes in a 6lo environment
so to achieve stateless forwarding, hence, avoiding using a routing
protocol. PASA is more suitable in case of stable and static
wireline connectivity, in order to avoid renumbering due to topology
changes. Even in such kind of scenarios, reliability remains an
issue. This memo tackles specifically reliability in PASA
deployments, analyzing possible broad solution categories to solve
the issue.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on 27 April 2023.
Copyright Notice
Copyright (c) 2022 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents (https://trustee.ietf.org/
license-info) in effect on the date of publication of this document.
Li, et al. Expires 27 April 2023 [Page 1]
Internet-Draft PASA Reliability October 2022
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document. Code Components
extracted from this document must include Simplified BSD License text
as described in Section 4.e of the Trust Legal Provisions and are
provided without warranty as described in the Simplified BSD License.
Table of Contents
1. Introduction and Problem Statement . . . . . . . . . . . . . 2
2. Requirements Notation . . . . . . . . . . . . . . . . . . . . 3
3. Potential Solution Approaches . . . . . . . . . . . . . . . . 3
3.1. Multi-Address Approach . . . . . . . . . . . . . . . . . 4
3.1.1. Topology building . . . . . . . . . . . . . . . . . . 4
3.1.2. Link Failures . . . . . . . . . . . . . . . . . . . . 8
3.1.3. Node Failures . . . . . . . . . . . . . . . . . . . . 11
3.1.4. Nodes Forwarding Procedure . . . . . . . . . . . . . 13
3.2. Single-Address Approach . . . . . . . . . . . . . . . . . 15
3.2.1. Link Failure . . . . . . . . . . . . . . . . . . . . 17
3.2.2. Node Failure . . . . . . . . . . . . . . . . . . . . 19
3.2.3. Node Forwarding Procedure . . . . . . . . . . . . . . 19
4. Links/Nodes Failure Detection and Recovery . . . . . . . . . 21
5. Robustness . . . . . . . . . . . . . . . . . . . . . . . . . 22
6. Security Considerations . . . . . . . . . . . . . . . . . . . 22
7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 22
8. References . . . . . . . . . . . . . . . . . . . . . . . . . 22
8.1. Normative References . . . . . . . . . . . . . . . . . . 22
8.2. Informative References . . . . . . . . . . . . . . . . . 23
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 24
1. Introduction and Problem Statement
The common characteristic of various topological addressing schemes
([I-D.daniel-6lowpan-hilow-hierarchical-routing],
[I-D.li-6lo-path-aware-semantic-addressing], [KIM07]) is the
possibility of nodes to forward packets without the need of routing
tables, and hence, without the need of routing protocols. In such
context the addresses are build in such a way that a node is capable
of forwarding a packet to the next hop by comparing the destination
address with its own address. It is not required to build routing
table on which to execute look-up algorithms, only neighbor awareness
is sufficient. However, this kind of stateless forwarding typically
works in a single topology with static paths, where high reliability
is hard to reach. Once a link (or a node) fails, the traffic will
not be routable and packets will be dropped, even in the presence of
alternative physical paths. Indeed, in order to use these
alternative paths renumbering is necessary to (re)build an
alternative logical topology. Such a solution, while looking as a
simple operation, may be not enough and complicate in practice, since
Li, et al. Expires 27 April 2023 [Page 2]
Internet-Draft PASA Reliability October 2022
it implies to put the system offline during the renumbering process.
What is desirable is to have some mechanism that with little extra
effort may quickly enable the usage of alternative paths, without the
need to put the system offline, hence providing the desired
reliability. The present memo analyzes two possible approaches to
guarantee reliability in PASA domains.
2. Requirements Notation
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in BCP
14 [RFC2119] and [RFC8174] when, and only when, they appear in all
capitals, as shown here.
3. Potential Solution Approaches
In order to improve the reliability of the system, the pre-requisite
is to have redundant links. This means that nodes are connected
likely in a meshed fashion, where some of the links are actively
used, and others not. In a normal situation, in the context of PASA,
the actively used links form a tree. This is the same concept of
spanning trees used in layer 2 technologies (e.g. [IEEE802.1W]).
When a problem is detected, various possibilities arise in order to
logically guarantee connectivity by starting using previously unused
links. In the specific case of PASA
[I-D.li-6lo-path-aware-semantic-addressing], all nodes, except the
root, have at least one secondary parent, which is used only if a
problem is detected when communicating with the primary one. In this
way, when the link toward the primary parent is broken, an
alternative link toward the secondary parent can be used. In such
context two different approaches can be identified:
* Multi-Address: using multiple addresses per node, one for each
alternative parent (logically creating multiple topologies).
* Single-Address: using one single address per node, even if an
alternative parent is present.
Both approaches, with pros and cons, are described and analyzed
hereafter.
Li, et al. Expires 27 April 2023 [Page 3]
Internet-Draft PASA Reliability October 2022
3.1. Multi-Address Approach
In the multi-address case multiple logical topologies are built using
different addresses and different links. In the following it is
assumed the two logical topologies are built on top of the physical
connectivity, however, the principles can be easily extended to more
than two topologies.
3.1.1. Topology building
In the multi-address case two root nodes are used. Each root node is
the root of a different tree covering all the nodes. Nodes have the
same role in both topologies. Meaning that a node will have the same
"Leaf" or "Forwarder" role in both topologies (see
[I-D.li-6lo-path-aware-semantic-addressing] for role definition).
The Allocation Function (AF) used to assign addresses in the two
parallel topologies might differ, however, attention should be given
to guarantee that addresses in the two topologies are different.
This can be easily achieved by using two different addresses for the
root nodes. Indeed such addresses will be the prefix of the whole
tree, which also mean that the address of the root nodes can be used
to actually identify the different topologies. For both topologies,
the address allocation works in the exact same way as described in
[I-D.li-6lo-path-aware-semantic-addressing], the only additional
action to be taken is that a node cannot choose the same parent node
in both topologies. This can be easily achieved by imposing that the
two parent must not have the same "node-id".
Let's make a simple example with the topology depicted in Figure 1,
where there are two root nodes, named "R-1" and "R-2" and a set of
few nodes with different roles, where "L" stands for leaf nodes and
"F" stands for forwarder nodes. Physical links are not depicted in
the figure but, as already mentioned, the assumption is that each
node is connected at least to two potential parents.
Li, et al. Expires 27 April 2023 [Page 4]
Internet-Draft PASA Reliability October 2022
+---+ +---+
|R-1| |R-2|
+---+ +---+
+----+ +----+ +----+ +----+
|F-11| |L-11| |F-12| |L-12|
+----+ +----+ +----+ +----+
+----+ +----+ +----+
|F-21| |L-21| |F-22|
+----+ +----+ +----+
+----+ +----+ +----+
|L-31| |L-32| |L-33|
+----+ +----+ +----+
Figure 1: Simple Topology example.
Let's now assume that R-1 has the address 1, which is used to
allocate the address to other nodes. After applying the allocation
function presented in [I-D.li-6lo-path-aware-semantic-addressing], a
possible outcome is the one presented in Figure 2, where the links
selected to form the logical topology are shown, as well as the
assigned addresses.
+---+ +---+
| 1|----------+ |R-2|
+---+-----+ \ +---+
/ \ \ \
/ \ \ \
+---+ +---+ +----+ +----+
| 10| | 11| | 110| | 111|
+---+ +---+ +----+ +----+
/ \ \
/ \ +------+
/ \ \
+----+ +----+ +-----+
| 100| | 101| | 1010|
+----+ +----+ +-----+
/ \ \
/ \ +-----------+
/ \ \
+----+ +-----+ +------+
|1001| |10011| |100111|
+----+ +-----+ +------+
Li, et al. Expires 27 April 2023 [Page 5]
Internet-Draft PASA Reliability October 2022
Figure 2: Possible PASA assignment and logical topology using R-1
as root.
In a similar way, assuming root R-2 has the address 01, and again
applying the allocation function presented in
[I-D.li-6lo-path-aware-semantic-addressing], a possible outcome is
the one presented in Figure 3, where the links selected to for the
logical topology are shown, as well as the assigned addresses.
+---+ +---+
|R-1| +-----------| 01|
+---+ / +------+---+
/ / / \
/ / / \
+---+ +---+ +----+ +----+
|010| |011| |0110| |0111|
+---+ +---+ +----+ +----+
/ / |
+----------+ / |
/ +---+ |
/ / |
+-----+ +-----+ +------+
|01100| |01101| |011010|
+-----+ +-----+ +------+
/ / |
+----------+ / |
/ +---+ |
/ / |
+-------+ +--------+ +---------+
|0110101| |01101011| |011010111|
+-------+ +--------+ +---------+
Figure 3: Possible PASA assignment and logical topology using R-2
as root.
When everything is working without problem, one of the logical
topologies can be used as primary topology, while using the second
one only in case of link/node failures. A simple selection can be
done for example with the rule:
* Interpreting root nodes' addresses as integers, choose the tree
with the smallest value.
Another approach could be to try to use some load balancing approach,
where sockets open on the various nodes are bound to one of the
available addresses based on some algorithm. The algorithm can be as
simple as a random choice, however it has to be considered, that
random local choices can uniformly distribute connections on
Li, et al. Expires 27 April 2023 [Page 6]
Internet-Draft PASA Reliability October 2022
different addresses, but it does not mean that the traffic is
uniformly distributed on the network as a whole [SINGH20]. Such kind
of optimization algorithm are out of the scope of this document. In
the following let's assume that a primary/secondary approach is used,
where the topology in Figure 2 is the primary one.
In [I-D.li-6lo-path-aware-semantic-addressing], the root node has
always address 1. Such simple approach allows to easily switch
between PASA and IPv6 addresses. Indeed, because all PASA addresses
start with a 1, to obtain an IPv6 address it is sufficient to prepend
a sufficient number of zeros to the PASA address, so align to the /64
suffix length, then prepend it with the whole /64 prefix of the
network. Vice-versa, to obtain the corresponding PASA address from
an IPv6 address, the /64 prefix is removed as well as all leading
zeros in the remaining suffix. This is not anymore sufficient.
Taking the example in Figure 3, the root node has to be aware that it
does not have to remove the last leading zero. In order to maintain
the simplicity of the design of PASA, the addresses of root nodes are
assigned as follows:
* Each root has an address where the least significant bit is set to
1 and all the others to zero.
* Each root has a different address length that has to be known by
the root.
* An address length of 1, means no leading zeros.
* An address length of n, means n-1 leading zeros followed by 1.
Coming back to our example, root R-1, has an address length equal to
1, hence its address is "1", as depicted in Figure 2, while R-2 has
an address length equal to 2, hence its address is "01", as depicted
in Figure 3. Only the root nodes have to be aware of the root
address length, because they are the only one performing address
translation (from PASA to IPv6 and vice-versa). Leafs and Forwarders
do not need to be aware of the root address length, since it is
implicit in the prefix used in the tree for the addresses. For
instance, the leafs at the bottom of Figure 3 can easily understand
that the root address length is 2 because their address starts with
01. The only requirement imposed by this solution on the nodes is to
allow addresses that start with zeros
([I-D.li-6lo-path-aware-semantic-addressing] only specifies addresses
starting with 1).
Li, et al. Expires 27 April 2023 [Page 7]
Internet-Draft PASA Reliability October 2022
3.1.2. Link Failures
In case of link failure there are three actions that need to be
performed in order to ensure connectivity.
1. The parent node, with respect to the link in object, has to
inform all the nodes between itself and the root that a certain
sub-tree is not reachable anymore through it, by using the
primary topology. This can be achieved by sending an ICMPv6
message toward the root, where each node the message traverses
stores the sub-tree unreachability status, so that packets
destined to that sub-tree are actually re-directed toward the
root. After this procedure, when a node sees a packet that is
destined to the node in the unreachable sub-tree, it sends it up
to the root.
2. The child node, with respect to the link in object, has to inform
the root that its sub-tree can still be reached, if traffic is
sent through the secondary topology using the secondary address
of the node that is the root of the sub-tree. This can be
achieved by sending an ICMPv6 message toward the root of the
primary tree, but actually using the secondary tree, hence using
the secondary address as a source of the message. With this
operation the root of the primary tree is now aware that to reach
a certain sub-tree, traffic has to be sent through the secondary
tree to a specific address (the secondary address of the child on
the broken link). In order actually ship a packet destined to an
address in the primary tree through the secondary tree, two
options are possible: encapsulation or routing.
* Encapsulation is pretty simple. Whenever there is a packet
destined to the sub-tree with a redirect entry on the primary
root, the root encapsulates (tunnels) the packet to the
secondary address of the child node of the broken link and
sends it to the secondary root. The packet will be forwarded
according to the stateless PASA procedure until it reaches the
intended node. There, it is decapsulated and the original
packet is routed in the sub-tree until its final destination.
In the other direction, all packets coming from the sub-tree
can be encapsulated toward the primary root, hence being
forwarded on the secondary tree, circumventing the broken
link.
* Routing relies on some forwarding entries stored on the nodes
along the path on the secondary tree. Basically, when the
ICMPv6 message, sent by the child node on the broken link, is
forwarded on the secondary tree, each node along the path
stores the fact that they are part of a forwarding path toward
Li, et al. Expires 27 April 2023 [Page 8]
Internet-Draft PASA Reliability October 2022
the sub-tree specified in the ICMPv6 message itself. In this
way, no additional encapsulation is necessary, since the
packet can be forwarded from the primary root to the secondary
root, who in turn will forward it to the child from which it
received the ICMPv6 message, and so on until the message
reaches the sub-tree where it is forwarded using the normal
PASA stateless forwarding. In the opposite direction, for
packets coming from the sub-tree, nodes along the alternate
path on the secondary tree will simply forward the packets to
the secondary root, who will forward them to the primary root.
The first solution (encapsulation) may increase the likelihood
to have MTU issues. Indeed, an additional encapsulation will
increase the packet size. The second solution does not create
MTU issues, but needs to store state in nodes along the
alternative paths. While the number of entries is certainly
limited, because it is the number of sub-trees unreachable
through the primary tree and using the node as part of the
alternative path. This may be an issue on devices with strong
memory constraints. Yet, if the state grows big it is the
symptom of massive failures in the network, which may be a far
bigger/urgent problem. In both cases the root nodes have to
keep some state: the redirection rules for all unreachable
sub-trees. This is not a problem since root gateways are
usually more powerful than the other nodes and do not run on
batteries. However, if the number of entries grows large,
this is again a symptom of massive failures.
3. Optionally, for optimization purposes, the child node, with
respect to the link in object, may inform all the nodes of its
sub-tree that they should start use the secondary tree (i.e. the
secondary address). This can be achieved by sending a specific
ICMPv6 message to all of its children, who will do the same
recursively. In this way communications will take advantage from
the stateless forwarding. However, communication using the
primary address, with the mechanism describe in the previous
points must still be supported, for ongoing communications that
would otherwise break and for any communication initiated from
the Internet toward and address in the primary tree. For
instance because only primary addresses are shared publicly (via
DNS or other means).
All of the above-mentioned ICMPv6 messages are forwarded using PASA
stateless forwarding procedure.
Using the example previously introduced, let's assume that the link
between F-11 and F-21 breaks (cf. Figure 1). This means that in the
primary topology (see Figure 2) the link between nodes 10 and 100 is
Li, et al. Expires 27 April 2023 [Page 9]
Internet-Draft PASA Reliability October 2022
broken. According the procedure presented above, the following
action are taken:
1. 10 sends an ICMPv6 message to the root. Root will register that
100-sub-tree is not reachable through 10 but has to be
redirected.
2. 100 sends an ICMPv6 message to 1 (root of primary tree) using
01100 as source address (see Figure 3). This message will be
forwarded first to 01, the root of the secondary tree, and then
to 1. Let's assume encapsulation is used, now root 1 has an
entry stating:
* For 100-sub-tree encapsulate to 01100 and forward to 01
3. 100 will send an ICMPv6 message to its children suggesting to use
the secondary addresses.
At this point connection is guaranteed. Let's assume the in the
primary tree (see Figure 2) nodes 11 and 1001 where communicating to
each other. Packets will flow in the following way:
* From 11 to 1001:
1. Packet is transmitted from 11 to 1 (on the primary tree).
2. Because of the redirect entry, 1 encapsulates packet toward
100 and transmits it to 01 (root secondary).
3. 01 will use PASA stateless forwarding to transmit the packet
to 0110 (on the secondary tree).
4. 0110 will use PASA stateless forwarding to transmit the packet
to 01100 (on the secondary tree).
5. 01100 will decapsulate, note the destination is on the primary
tree, use the PASA stateless forwarding to transmit the packet
to 1001 (on the primary tree).
* From 1001 to 11:
1. Packet is transmitted from 1001 to 100 (on the primary tree).
2. Because 100 knows the upstream link is broken it encapsulates
the packet with source 01100 and destination 1 (root primary
tree) then transmits the packet to 0110 (on the secondary
tree).
Li, et al. Expires 27 April 2023 [Page 10]
Internet-Draft PASA Reliability October 2022
3. 0110 will use PASA stateless forwarding to transmit the packet
to 01 (on the secondary tree).
4. 01 will see that packet is destined to the primary root and
transmits it to 1.
5. 1 will decapsulate, note the destination is on the primary
tree, use the PASA stateless forwarding to transmit the packet
to 11 (on the primary tree).
In case of communication toward/from the public Internet the
procedure is similar. For outgoing packets the primary root will
expand the PASA header to a full IPv6 header and forward it upstream.
For incoming packets, the root will first reduce the IPv6 header to
an PASA header then forward it as described above. PASA header
expansion and IPv6 header reduction are operations described in
[I-D.li-6lo-path-aware-semantic-addressing].
3.1.3. Node Failures
In case that an entire node fails, several links will not be usable
anymore. Nevertheless, the procedure described in the previous
section can be still applied, what may change is who is performing
the action. More specifically:
1. The parent of the failed node, has to inform all the nodes
between itself the root that a certain sub-tree is not reachable
anymore through it. This is the exact same procedure like in
Section 3.1.2.
2. All of the children of the failed node, have to independently
inform the root that its sub-tree can still be reached if traffic
is sent through the secondary topology, using the secondary
address of the node that is the root of the sub-tree. This is
the exact same procedure like in Section 3.1.2, just done by all
children.
3. All of the children of the node, optionally, for optimization
purposes, may inform all the nodes of their sub-trees that they
should start use the secondary tree (i.e. the secondary address).
This is the exact same procedure like in Section 3.1.2, just done
by all children.
Using again the example previously introduced, let's assume that node
F-21 fails (cf. Figure 1). This means that in the primary topology
(see Figure 2) the links between nodes 10 and 100 is unusable, as
well as the links between 100 and its three children, namely 1001,
Li, et al. Expires 27 April 2023 [Page 11]
Internet-Draft PASA Reliability October 2022
10011, and 100111. According the procedure presented above, the
following action are taken:
1. 10 sends an ICMPv6 message to the root. Root will register that
100-sub-tree is not reachable through 10 but has to be
redirected.
2. The three children of 100 will perform the following:
* 1001 sends an ICMPv6 message to 1 (root of primary tree) using
01100 as source address (see Figure 3). This message will be
forwarded first to 01, the root of the secondary tree, and
then to 1. Let's assume encapsulation is used, now root 1 has
an entries stating:
- For 1001-sub-tree encapsulate to 0110101 and forward to 01
* 10011 sends an ICMPv6 message to 1 (root of primary tree)
using 01100 as source address (see Figure 3). This message
will be forwarded first to 01, the root of the secondary tree,
and then to 1. Let's assume encapsulation is used, now root 1
has an entry stating:
- For 10011-sub-tree encapsulate to 01101011 and forward to
01
* 100111 sends an ICMPv6 message to 1 (root of primary tree)
using 01100 as source address (see Figure 3). This message
will be forwarded first to 01, the root of the secondary tree,
and then to 1. Let's assume encapsulation is used, now root 1
has an entry stating:
- For 100111-sub-tree encapsulate to 011010111 and forward to
01
3. The children of 100, will send an ICMPv6 message to their
children (if any) suggesting to use the secondary addresses.
At this point connection is guaranteed. Let's assume, like in the
example for the link failure, that in the primary tree (see Figure 2)
nodes 11 and 1001 where communicating to each other. Packets will
flow in the following path:
* From 11 to 1001:
1. Packet is transmitted from 11 to 1 (on the primary tree).
Li, et al. Expires 27 April 2023 [Page 12]
Internet-Draft PASA Reliability October 2022
2. Because of the redirect entry, 1 encapsulates packet toward
100 and transmits it to 01 (root secondary).
3. 01 will use PASA stateless forwarding to transmit the packet
to 0110101 (on the secondary tree).
4. 0110101 will decapsulate, note the destination is its own
primary address, the packet will be decapsulate once more and
delivered to the upper layer.
* From 1001 to 11:
1. Because 1001 knows the upstream link is broken it encapsulates
the packet with source 0110101 and destination 1 (root primary
tree) then, using PASA stateless forwarding, it will transmit
the packet to 01 (on the secondary tree).
2. 01 will see that packet is destined to the primary root and
transmits it to 1.
3. 1 will decapsulate, note the destination is on the primary
tree, use the PASA stateless forwarding to transmit the packet
to 11 (on the primary tree).
In case of communication toward/from the public Internet the
procedure is the same as described in Section 3.1.2.
3.1.4. Nodes Forwarding Procedure
Nodes, other than leafs, have to forward packets according to the
procedures described in the previous sections. Nevertheless,
compared to the original specification the modifications are very
limited. Hereafter, the forwarding procedure for both forwarder and
root nodes is provided. The mention "PASA Native Forwarding" is used
where the original procedure described in
[I-D.li-6lo-path-aware-semantic-addressing] is employed.
3.1.4.1. Forwarder Nodes
As describe in Figure 4, in the context of multiple topologies, when
a a forwarder node receives a packet, it needs first to verify if
there is any rule that redirects the packet. If it is not the case,
it needs to check if there is an encapsulation rule, if it is the
case then the packets needs to be encapsulated accordingly. Then
normal PASA forwarding is applied.
Li, et al. Expires 27 April 2023 [Page 13]
Internet-Draft PASA Reliability October 2022
+-----------------+
| Packet Received |
+-----------------+
|
V
+---------------+ +-----------------+
/ Is there a \ Yes | Forward |
| redirect rule |------->| according |---+
\ that applies? / | to rule | |
+--------------+ +-----------------+ |
| No |
| |
V |
+---------------+ +-----------------+ |
/ Is there an \ Yes | Encapsulate | |
| encap. rule |------->| according | |
\ that applies? / | to rule | |
+--------------+ +-----------------+ |
| No | |
|<--------------------------+ |
V |
+-----------------+ |
| PASA | |
|Native Forwarding| |
+-----------------+ |
| <--------------------------------------+
V
+------------+
| END |
+------------+
Figure 4: Forwarder node forwarding procedure in case of multiple
topologies.
3.1.4.2. Root Nodes
In the case of a root node, and in the context of multiple
topologies, the PASA native forwarding is always applied for outward
packets. Only in case of inward packets, the node has to check
whether there is an encapsulation rule through an alternative
topology to bypass a failed link/node. Figure 5 show this simple
case.
Li, et al. Expires 27 April 2023 [Page 14]
Internet-Draft PASA Reliability October 2022
+-----------------+
| Packet Received |
+-----------------+
|
V
+---------------+ +-----------------+
/ Is there an \ Yes | Encapsulate |
| encap. rule |------->| according |
\ that applies? / | to rule |
+--------------+ +-----------------+
| No |
| |
V V
+-----------------+ +-----------------+
| PASA | | Forward to |
|Native Forwarding| | Alternative Root|
+-----------------+ +-----------------+
| |
| <--------------------------+
V
+------------+
| END |
+------------+
Figure 5: Root node forwarding procedure in case of multiple
topologies.
3.2. Single-Address Approach
In this approach, starting from the root node, we can assign a single
address to each node in the PASA network based on the address
allocation function described in
[I-D.li-6lo-path-aware-semantic-addressing]. All nodes with assigned
addresses will send a message to the root to register themselves so
that the root has an overview of the nodes and the topology in the
PASA network. The nodes with the links used to assign the addresses
form the primary tree topology. By default, the node forwards the
packet via the primary tree by using the native PASA forwarding
method defined in [I-D.li-6lo-path-aware-semantic-addressing].
The root will have a backup with the same address 1, and Virtual
Router Redundancy Protocol (VRRP [RFC5798]) could be used to
implement same address root redundancy. In order to provide
reliability inside the topology, each node will have at least one
alternative parent for redundancy. This alternative uplink is stored
added to the already existing Neighbor Discovery table. For a
forwarder node, once an alternative downlink is established, because
it is an alternative parent, it has to record this downlink into the
Li, et al. Expires 27 April 2023 [Page 15]
Internet-Draft PASA Reliability October 2022
ND table as well. For leaf nodes, there will be only alternative
uplink entries. All the alternative links will be reported to the
root using ICMPv6 messages. Therefore for forwarder nodes, there
will be alternative uplink(s) and alternative downlink(s) stored in
the ND table, and leafs nodes will have a ND table only with
alternative uplink(s). An example of ND table which includes the
alternative parent(s)/children is shown in Figure 6). In particular,
it shows what should be the ND table content for node 100 in the
topology shown in Figure 7.
+-------------+-------+
| Destination | Flags |
+-------------+-------+
| 100 | I | I = Current Node
+-------------+-------+
| 10 | PP | PP = Primary Parent
+-------------+-------+
| 1000 | PFC | PFC = Primary Forwarder Child
+-------------+-------+
| 10010 | PFC |
+-------------+-------+
| 1001 | PLC | PLC = Primary Leaf Child
+-------------+-------+
| 10011 | PLC |
+-------------+-------+
| 110 | AP | AP = Alternative Parent
+-------------+-------+
| 10100 | AFC | AFC = Alternative Forwarder Child whose
+-------------+-------+ alternative parent is the current node
| 10101 | ALC | ALC = Alternative Leaf Child whose
+-------------+-------+ alternative parent is the current node
Figure 6: Example of a ND Table of a forwarder node with address
of '100'.
There can be more than one forwarder and leaf children, "Primary"
here means that they belong to the primary topology, to differentiate
them from the backup alternative role. The first entry of Figure 6
shows the address of the node itself '100'. This node's parent on
the primary tree is '10' that is recorded in second entry and marked
accordingly a Primary Parent (PP). There are two Primary Forwarder
Children (PFC), namely '1000' and '10010, followed by two Primary
Leaf Children (PLC), namely '1001' and '10011'. Then one alternative
parent (AP) follows, namely '110'. Finally, for the sake of clarity,
two alternative children have been added to complete the table (not
depicted in Figure 7), an Alternative Forwarder Child (AFC) with
address 10100, and an Alternative Leaf Child (ALC) with address
10101.
Li, et al. Expires 27 April 2023 [Page 16]
Internet-Draft PASA Reliability October 2022
As there is only one primary tree, in general, the packet forwarding
will follow the normal PASA forwarding method if there is no link or
node failure. Even when there are failures on the alternative links,
the normal PASA forwarding method is not impacted. However, if there
is a link failure on the primary tree, the forwarding behavior will
change as described in the following.
3.2.1. Link Failure
Upon a link failure an ICMPv6 message will be generated to report the
event to the root. The root will then compute a new forwarding path
based on the current state and encapsulate (tunnel) the packet to
nodes where broken links could be avoided.
1 '1'
/-\ /-\
| |\----------\...| |.......
\-/\ \--\......\...\-/. .
/ | \....\- \...\. . .
/ .|.... \...... \_._\-----.-----\ .
/.... | .... \ . \__ . \ .
10 /.. 11| . \ 110 \__ .111 \ . 1110
/-\ . /-\ /-\ \ /-\ \- /-\
| |. | | .| | | | | |
/\-/\.... \-/ .. .\-\ \-/ \-/
/ | \ \ ... ..... .. . \
/ | \ \.. ..... .. . \
/ | \.. \ .. .. ... \
Failure X | .... ...\ . . . .. \
/ ... ..\ .. .\ . .. \
/ ... | ... \. \ . ..\
100 /-\..101/-\. /-\1010 \/-\ 1011 .\/-\ 1101
| | | | .| | | | | |
\-/ \ \-/ . . \-/ \-/ \-/
/ | \ \ . . . .
/ | \ \ .. . . .
/ | \..\ . .. .
/ | .\ . \ . .
/ | . . \.. .
/ ... . \ .. \ .
/ ... | . \. \ .
/-\.. /-\.. /-\ \/-\
| | | | | | | |
\-/ \-/ \-/ \-/
1000 1001 10010 10011
Figure 7: An example of link failure in single address topology.
Li, et al. Expires 27 April 2023 [Page 17]
Internet-Draft PASA Reliability October 2022
In order to give an example to what happens to packets flowing
downlink, let's assume a packet initiated from node 1101 destined to
node 1001, and that the link between node 10 and node 100 is broken.
When the link fails, upon detection of the failure, node 10 will send
an ICMPv6 message to the root, to make it aware of the failure. The
packet forwarding will happen as follows:
1. Packet is transmitted from node 1101 to the root 1, using PASA
native forwarding.
2. Root 1 is aware that the path to destinations in the 100 sub-tree
are not reachable through normal PASA forwarding because of the
link failure, hence computes an alternative path. In this
example: 1 -> 110 -> 100 -> 1001. Since normal PASA forwarding
does not allow to go first through node 110 and then node 100,
the root 1 will encapsulate the addresses of node 110 and node
100 in an extension header so to perform segmented routing
[I-D.geng-spring-sr-redundancy-protection].
3. Once packet reaches 100, the segment routing extension is
dropped, and the packet is sent to its destination 1001 using
PASA native forwarding.
In the unlikely case that the root is not yet aware of the link
failure during the packet transmission, the packet forwarding will
happen as follows:
1. Packet is transmitted from node 1101 to the root 1, using PASA
native forwarding.
2. Packet is transmitted from root 1 to node 10, following the
normal PASA forwarding method.
3. Node 10, which is aware about the link failure, redirects the
packet back to the root with SRv6 encapsulation.
4. Root 1, which should in the meantime have received an ICMPv6
message notifying the failure of the link, receives the
encapsulated packet and, after decapsulation, it operates like in
the previous example. Since it is now aware that the path to
destinations in the 100 sub-tree are not reachable through normal
PASA forwarding because of the link failure, hence computes an
alternative path. In this example: 1 -> 110 -> 100 -> 1001.
Since normal PASA forwarding does not allow to go first through
node 110 and then node 100, the root 1 will encapsulate the
addresses of node 110 and node 100 in an extension header so to
perform segmented routing
[I-D.geng-spring-sr-redundancy-protection].
Li, et al. Expires 27 April 2023 [Page 18]
Internet-Draft PASA Reliability October 2022
5. Once packet reaches 100, segment routing extension dropped, and
packet is sent to its destination 1001 using PASA native
forwarding.
Let's now look at what happens to packets flowing in the opposite
direction, when packets are sent from 1001 to 1101, with the same
link failed, namely the link between 100 and 10. Upon link failure
detection by 100, the node will send an ICMPv6 message through an
alternative parent, toward the root, to report the link failure, The
packet will be handled as follows:
1. Packet is transmitted from node 1001 to node 100 using PASA
native forwarding.
2. Because of the failed link, node 100 sends the packet to an
alternative parent node.
3. PASA native forwarding is used then. If the alternative parent
is in the same sub-tree like the destination, the packet is
forwarded down ward to the right child, otherwise it is sent
upward to the its own parent. This goes on recursively until the
packet reaches the root in the worst case, where it is then sent
downward to the correct forwarder child, until it reaches the
destination. In our example the path would be: 100 -> 110 ->
1101.
3.2.2. Node Failure
As for the multiple-address case, a node failure can be seen as
multiple link failures, basically all links the node connects to. In
this case the parent of the failed node and its children will simply
apply the same procedure described in the previous section.
3.2.3. Node Forwarding Procedure
3.2.3.1. Forwarder Node
Li, et al. Expires 27 April 2023 [Page 19]
Internet-Draft PASA Reliability October 2022
+----------------+
| Received Packet|
+-------+--------+
|
V
+------------------------+
| Perform PASA Forwarding |
+-----------+------------+
|
V
+---------------------+
/ \
| Outgoing Link working?|---------------------------------+
\ / Yes |
+---------------------+ |
| |
| No |
V |
+---------------------+ V
/ \ Down +-------------------+ +-----+
| Down/Up Link Failure? |----->| Redirect to Root |--->| END |
\ / +-------------------+ +-----+
+--------------------+ ^
| |
| Up |
V |
+---------------------+ |
| Send the Packet to | |
| the Alternative |----------------------------------+
| Parent |
+---------------------+
Figure 8: Forwarding Procedure of Forwarder Node
As describe in Figure 8, in the context of single-address approach,
when a forwarder node receives a packet, it performs the normal PASA
native forwarding (after decapsulation, if needed). If case of link
failure, the forwarder will take different actions depending on
downlink or uplink failure, as depicted in the Section 3.2.1.
3.2.3.2. Root Node
In the case of a root node, and in the context of single-address
approach, the PASA native forwarding is always applied, for outward
packets. Only in case of inward packets, the node has to check
whether there is a redirection needed. If it is the case, it will
compute the path and define the segment routing header in order to
forward the packet to avoid the broken link(s).
Li, et al. Expires 27 April 2023 [Page 20]
Internet-Draft PASA Reliability October 2022
+----------------+
| Received Packet|
+-------+--------+
|
V
+---------------------+
/ Is the a \ No
| redirect rule due to |-----------+
\ broken links / |
+---------------------+ |
|Yes |
V |
+---------------------+ |
| Encapsulate to | |
| alternative path | |
+---------------------+ |
| |
V |
+------------------------+ |
| PASA Native Forwarding |<---------+
+------------------------+
|
V
+-------+
| END |
+-------+
Figure 9: Forwarding Procedure on root node.
4. Links/Nodes Failure Detection and Recovery
Previous sections describe actions and possible solution to failures
events, but never discussed how failures are detected. This memo
assumes that depending on the specific technology in use and the
level of desired reliability, the most suitable failure detection
mechanism is used to trigger the above-described actions. It is
considered not desirable to define one single failure detection
technique to be used in the context of PASA, neither to define new
ones.
The link failure could be detected leveraging layer 2 feedback, like
for instance the lack of acknowledgement upon packet transmission.
It can also be detected using existing network layer solution, like
for instance using Bidirectional Forwarding Detection (BFD [RFC7130])
or IPv6 specific mechanisms [RFC5534].
Another aspect of the general failure management is to recover from
failures, going back to the original state. In the context of PASA
there are a couple of possible approaches that can be used. Use
Li, et al. Expires 27 April 2023 [Page 21]
Internet-Draft PASA Reliability October 2022
native addresses lifetime. Addresses can be assigned associated with
a lifetime. When such lifetime expires, node have to undergo the
same initial procedure address allocation. This is also a good
moment to check whether a certain link or node is back to normal
functioning. If it is not the case, the algorithmic procedure will
anyway create topologies that do not take into account failed links/
nodes. A faster approach could be based, like in the case of failure
detection, on periodic checks that may leverage on layer 2 features
or on some neighbor discovery messages. The former method being more
effective, the latter introducing communication overhead.
5. Robustness
Real robustness provided by the different approaches depends from the
specific topology. The single-address solution may introduce more
state. Indeed, the root has the overview of the PASA network. It
knows all nodes' addresses, the alternative links and the broken
links. It is able to compute a usable path towards a destination.
This comes with the benefit of potentially being able to find a
higher number of alternative path, hence, in the end providing a
stronger protection against multiple failures. The forwarder node
and the leaf node are rather dummy and use PASA stateless forwarding.
They only are aware of link state toward their direct neighbors, and
take action accordingly. The multi-address approach leverages more
on the stateless forwarding of PASA. The root is in general unaware
of nodes' addresses, and the network topology. In case of failure, a
redirection rule is set on the root, hence there amount of state is
proportional to the number of failures. This means less state, but
may be less robust to multiple failures. Differently from the single
address solution, a small amount state is also required on forwarder
nodes, because if a link fails a redirect rule has to be used.
The above mentioned pros and cons need to be pondered when choosing a
reliability solution to be deployed in an PASA domain.
6. Security Considerations
TBD
7. IANA Considerations
TBD.
8. References
8.1. Normative References
Li, et al. Expires 27 April 2023 [Page 22]
Internet-Draft PASA Reliability October 2022
[I-D.li-6lo-path-aware-semantic-addressing]
"*** BROKEN REFERENCE ***".
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/info/rfc2119>.
[RFC5534] Arkko, J. and I. van Beijnum, "Failure Detection and
Locator Pair Exploration Protocol for IPv6 Multihoming",
RFC 5534, DOI 10.17487/RFC5534, June 2009,
<https://www.rfc-editor.org/info/rfc5534>.
[RFC5798] Nadas, S., Ed., "Virtual Router Redundancy Protocol (VRRP)
Version 3 for IPv4 and IPv6", RFC 5798,
DOI 10.17487/RFC5798, March 2010,
<https://www.rfc-editor.org/info/rfc5798>.
[RFC7130] Bhatia, M., Ed., Chen, M., Ed., Boutros, S., Ed.,
Binderberger, M., Ed., and J. Haas, Ed., "Bidirectional
Forwarding Detection (BFD) on Link Aggregation Group (LAG)
Interfaces", RFC 7130, DOI 10.17487/RFC7130, February
2014, <https://www.rfc-editor.org/info/rfc7130>.
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
May 2017, <https://www.rfc-editor.org/info/rfc8174>.
8.2. Informative References
[I-D.daniel-6lowpan-hilow-hierarchical-routing]
Park, S. D., "Hierarchical Routing over 6LoWPAN (HiLow)",
Work in Progress, Internet-Draft, draft-daniel-6lowpan-
hilow-hierarchical-routing-01, 18 June 2007,
<https://www.ietf.org/archive/id/draft-daniel-6lowpan-
hilow-hierarchical-routing-01.txt>.
[I-D.geng-spring-sr-redundancy-protection]
Geng, X., Chen, M., Yang, F., Camarillo, P., and G. S.
Mishra, "SRv6 for Redundancy Protection", Work in
Progress, Internet-Draft, draft-geng-spring-sr-redundancy-
protection-05, 2 August 2021,
<https://www.ietf.org/archive/id/draft-geng-spring-sr-
redundancy-protection-05.txt>.
[IEEE802.1W]
"IEEE Std 802.1w-2001, IEEE Std for Local and metropolitan
are networks - Common specifications Part 3; Media Access
Li, et al. Expires 27 April 2023 [Page 23]
Internet-Draft PASA Reliability October 2022
Control (MAC) Bridges - Amendment 2; Rapid
Reconfiguration", n.d.,
<https://standards.ieee.org/ieee/802.1w/1046/>.
[KIM07] Kim, Y.-S., Lee, E. J., Kim, B. S., and H. S. Kim,
"Extended Tree-Based Routing Algorithm in IPv6-enabled
Wireless Sensor Networks", IEEE 2007 International
Conference on Convergence Information Technology (ICCIT
2007), pp. 1269-1274, 2007.
[SINGH20] Singh, S. K. and J. Prakash, "Energy Efficiency and Load
Balancing in MANET: A Survey", IEEE 2020 6th International
Conference on Advanced Computing and Communication Systems
(ICACCS), 2020, pp. 832-837, 2020.
Authors' Addresses
Guangpeng Li
Huawei Technologies
Beiqing Road, Haidian District
Beijing
100095
China
Email: liguangpeng@huawei.com
David Lou
Huawei Technologies
Riesstrasse 25
80992 Munich
Germany
Email: zhe.lou@huawei.com
Luigi Iannone
Huawei Technologies France S.A.S.U.
18, Quai du Point du Jour
92100 Boulogne-Billancourt
France
Email: luigi.iannone@huawei.com
Li, et al. Expires 27 April 2023 [Page 24]