The IPv6 Flow Label within a RPL domain
Cisco Systems
Village d'Entreprises Green Side400, Avenue de RoumanilleBatiment T3Biot - Sophia Antipolis06410FRANCE+33 4 97 23 26 34pthubert@cisco.com
Routing
6MAN
This document present how the Flow Label can be used inside a RPL domain
as a replacement to the RPL option and provides rules for the root to set
and reset the Flow Label when forwarding between the inside of RPL
domain and the larger Internet, in both direction. This new operation
saves 44 bits in each frame, and an eventual IP-in-IP encapsulation within
the RPL domain that is required for all packets that reach outside of the
RPL domain.
The emergence of radio technology enabled a large variety of new types of devices
to be interconnected, at a very low marginal cost compared to wire, at any range from
Near Field to interplanetary distances, and in circumstances where wiring would be less
than practical, for instance rotating devices.
In particular, IEEE802.14.5 that is
chartered to specify PHY and MAC layers for radio Lowpower Lossy
Networks (LLNs), defined the
TimeSlotted Channel Hopping (TSCH) mode of operation as part of
the IEEE802.15.4e MAC specification in order to address Time Sensitive
applications.
The
6TISCH architecture specifies the operation IPv6 over TSCH
wireless networks attached and synchronized by backbone routers.
In that model, route Computation may be achieved in a centralized
fashion by a Path Computation Element (PCE), in a distributed fashion
using the Routing Protocol for Low Power and
Lossy Networks (RPL), or in a mixed mode.
The Backbone Routers may typically serve as roots for the RPL domain.
6TiSCH was created to simplify the adoption of IETF technology by other
Standard Defining Organizations (SDOs), in particular in the Industrial
Automation space, which already relies on variations of IEEE802.15.4e
TSCH for Wireless Sensor Networking.
ISA100.11a is an example of such industrial WSN standard, using
IEEE802.15.4e over the classical IEEE802.14.5 PHY. In that case, after
security is applied, roughly 80 octets are available per frame for
IP and Payload. In order to 1) avoid fragmentation and 2) conserve
energy, the SDO will scrutinize any bit in the frame and reject any
waste.
The challenge to obtain the adoption of IPv6 in the original standard
was really to save any possible bit in the frames, including the UDP
checksum which was an interesting discussion on its own. This work was
actually one of the roots for the 6LoWPAN
Header Compression work, which goes down to the individual bits
to save space in the frames for actual data, and allowed ISA100.11a to
adopt IPv6.
The design of Lowpower Lossy Networks is generally focussed on saving
energy, which is the most constrained resource of all. The other
constraints, such as the memory capacity and the duty cycling of the LLN
devices, derive from that primary concern. Energy is typically available
from batteries that are expected to last for years, or scavenged from the
environment in very limited quantities. Any protocol that is intended for
use in LLNs must be designed with the primary concern of saving energy as
a strict requirement.
The Routing Protocol for Low Power and Lossy Networks (RPL)
specification defines a generic Distance Vector protocol that is indeed
designed for very low energy consumption and adapted to a variety of LLNs.
RPL forms Destination Oriented Directed Acyclic Graphs (DODAGs) which root
often acts as the Border Router to connect the RPL domain to the Internet.
The root is responsible to select the RPL Instance that is used to forward
a packet coming from the Internet into the RPL domain and set the related
RPL information in the packets.
A classical RPL implementation will use the RPL
Option for Carrying RPL Information in Data-Plane Datagrams to tag
a packet with the Instance ID and other information that RPL requires for
its operation within the RPL domain.
In particular, the Rank, which is the scalar metric computed by an specialized Objective Function
such as , is modified at each hop and allows to validate that the packet
progresses in the expected direction each upwards or downwards in along the DODAG.
With the RPL option is encoded as 6 Octets;
it must be placed in a Hop-by-Hop header that represents 2 additional
octets for a total of 8. In order to limit its range to the inside the RPL domain,
the Hop-by-Hop header must be added to (or removed from) packets
that cross the border of the RPL domain. For reasons such as the capability
to send ICMP errors back to the source, this operation involves an extra
IP-in-IP encapsulation inside the RPL domain for all the packets which path is
not contained within the RPL domain.
The 8-octets overhead is detrimental to the LLN operation, in particular
with regards to bandwidth and battery constraints. The extra encapsulation
may cause a containing frame to grow above maximum frame size, leading to
Layer 2 or 6LoWPAN fragmentation,
which in turn cause even more energy spending and issues discussed in the
LLN Fragment Forwarding
and Recovery.
Considering that, in the classical IEEE802.14.5 PHY that is used
by ISA100.11a, roughly 80 octets are available per frame after security is
applied, and any additional transmitted bit weights in the energy
consumption and drains the batteries.
Regrettably, does not provide an efficient
compression for the RPL option so the cost in current implementations can
not be alleviated in any fashion. So even for packets that are confined
within the RPL domain and do not need the IP-in-IP encapsulation, the use
of the flow label instead of the RPL option would be a valuable saving.
In Industrial Automation and Control Systems (IACS) ,
a packet loss is usually acceptable but jitter and latency must be
strictly controlled as they can play a critical role in the interpretation
of the measured information.
Sensory systems are often distributed, and the control information can
in fact be originated from multiple sources and aggregated.
In such cases, related packets from multiple
sources should not be load-balanced along their path in the Internet.
In a typical LLN application, the bulk of the traffic consists of small
chunks of data (in the order few bytes to a few tens of bytes) at a time.
4Hz is a typical loop frequency in Process Control, though it can be a
lot slower than that in, say, environmental monitoring. The granularity
of traffic from a single source is too small to make a lot of sense in
load balancing application.
As a result, it can be a requirement for related measurements from multiple sources
to be treated as a single flow following a same path over the Internet so
as to experience similar jitter and latency. The traditional tuple of source,
destination and ports might then not be the proper indication to isolate
a consistent flow. On the other hand, the flow integrity can be preserved
in a simple manner if the setting of the Flow Label in the IPv6
header of packets outgoing a RPL domain, is centralized to the root of the
RPL DODAG structure, as opposed to distributed across the actual sources.
Considering that the goal for setting the Flow Label as prescribed in
the IPv6 Flow Label Specification is to
improve load balancing in the core of the Internet, it is unlikely that
LLN devices will consume energy to generate and then transmit a Flow Label
to serve outside interests and the Flow Label is generally left to zero
so as to be elided in the 6LoWPAN compression. So
in a general manner the interests of the core are better served if the RPL
roots systematically rewrite the flow label rather than if they never do.
For packets coming into the RPL domain from the Internet, the value for
setting the Flow Label as prescribed in
is consumed once the packet has traversed the core and reaches the LLN.
Then again, there is little value but a high cost for the LLN in spending
20 bits to transport a Flow Label from the Internet
over the constrained network to a destination node that has no use of it.
All the packets from all the nodes in a same DODAG that are leaving a RPL
domain towards the Internet will transit via a same RPL root. The RPL root
segregates the Internet and the RPL domain, which enables the capability
to reuse the Flow Label within the RPL domain.
On the other hand, the operation of resetting or reusing the IPv6 Flow
Label at the root of a RPL domain is a deviation from
the IPv6 Flow Label Specification , in that
it is neither the source nor the first hop router that sets the final Flow
Label for use outside the RPL domain.
Additionally, using the Flow Label to transport the information that is
classically present in the RPL option implies that the Flow Label is
modified at each hop inside the RPL domain, which again
is a limited deviation from ,
which explicitly requires that the flow label cannot be modified once set.
But if we consider the whole RPL domain as a large virtual host from the
standpoint of the rest of the Internet, the interests that lead to
, and in
particular load balancing in the core of the Internet, are probably better
served if the root guarantees that the Flow Label is set in a compliant
fashion than if we rely on each individual sensor that may
not use it at all, or use it slightly differently such as done in ISA100.11a.
Additionally, LLN flows can be compound flows aggregating information from multiple sources.
The root is an ideal place to rewrite the Flow Label to a same value for a same flow across multiple
sources, ensuring compliance with the rules defined by for use outside
of the RPL domain and in particular in the core of the Internet. It can be noted that provides an efficient header compression for packets
that do have the Flow Label set in the IPv6 header. It results that the overhead for transporting the RPL information
can be down from 64 to 20 bits, alleviating at the same time the need for IP-in-IP encapsulation.
This optimization cannot be ignored, and can make the difference for the adoption of RPL and 6TiSCH
by external standard bodies.
This document specifies how the Flow Label can be reused within the RPL domain as a
replacement to the RPL option. The use of the Flow Label within a RPL domain is an instance of
the stateful scenarios as discussed in where the states include the Rank
of a node and the RPLInstanceID that identifies the routing topology.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL",
"SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY",
and "OPTIONAL" in this document are to be interpreted as
described in .The Terminology used in this document is consistent with and
incorporates that described in `Terminology in Low power And Lossy
Networks'
and .This specification applies to a RPL domain that
forms a stub LLN and is connected to the Internet by and only by its RPL
root(s), which act(s) as Border Router(s) for the LLN. With RPL, a root
is the bottleneck for all the traffic between the Internet and the
Destination-Oriented Directed Acyclic Graph (DODAG) that it serves.
In that context, the specification entitles a RPL root to rewrite the
IPv6 Flow Label of all packets entering or
leaving the RPL domain in both directions, from and towards the Internet,
regardless of its original setting. This may seem contradictory with
the IPv6 Flow Label Specification which
stipulates that once it is set, the Flow Label is left unchanged; but the
RFC also indicates a violation to the rule can be accepted for compelling
reasons, and that security is a case justifying such a violation.
This specification suggests that energy-saving is another compelling
reason for a violation to the aforementioned rule.
For the compelling reason of saving energy, this specification allows
that regardless of its original setting, a root of a RPL domain MAY reset
the Flow Label of IPv6 packets
entering the RPL domain to zero for an optimal Header Compression by
6LoWPAN . The specification also allows that
the root and LLN routers MAY reuse the Flow Label inside the LLN for LLN
purposes, such as to carry the RPL Information as detailed hereafter.
This specification also allows that regardless of its original setting, a
a root of a RPL domain MAY set low Label of IPv6 packets that exits the
RPL domain MAY be set by the RPL, in a manner that SHOULD conform the
prescriptions in ,
and that a source in the RPL domain MAY NOT expect that its setting of the
Flow Label be preserved end-to-end. From there,
the capability by RPL routers inside the LLN to alter a non-zero Flow
Label between the source and the root is another minor deviation to
that is also acceptable since it is transparent
to the core of the Internet.
section 11.2 specifies the fields that are
to be placed into the packets for the purpose of Instance Identification,
as well as Loop Avoidance and Detection. Those fields include an 'O', and 'R'
and an 'F' bits, the 8-bit RPLInstanceID, and the 16-bit SenderRank.
SenderRank is the result of the DAGRank operation on the rank of the sender,
where the DAGRank operation is defined in section 3.5.1 as:
DAGRank(rank) = floor(rank/MinHopRankIncrease)If MinHopRankIncrease is set to a multiple of 256, it appears that
the most significant 8 bits of the SenderRank will be all zeroes and
could be omitted. In that case, the Flow Label MAY be used as a
replacement to the RPL option. To achieve this, the
SenderRank is expressed with 8 least significant bits, and the information
carried within the Flow Label in a packet is constructed follows: The first (leftmost) bit of the Flow Label is reserved and should be set to zero.
section 3 intentionally does not consider flow label values in
which any of the bits have semantic significance. However, the present specification assigns
semantics to various bits in the flow label, destroying within the edge network that is the
RPL domain the property of belonging to a statistically uniform distribution that is desirable
in the rest of the Internet.
It can be noted that the rationale for the statistically uniform distribution does not
necessarily bring a lot of value within the RPL domain. In a specific use case where it would,
that value must be compared with that of the battery savings in order to decide which technique
the deployment will use to transport the RPL information.
When routing a packet towards the RPL domain, the root applies a policy to determine whether
the Flow Label is to be used to carry the RPL information. If so, the root MUST reset the Flow Label and
then it MUST set all the fields in the Flow Label as prescribed by using the
format specified in . In particular, the root selects the Instance that will
be used to forward the packet within the RPL domain.
When routing a packet outside the RPL domain, the root applies a policy to determine whether
the Flow Label was used to carry the RPL information. If so, the root MUST reset the Flow Label.
The root SHOULD recompute a Flow Label following the rules prescribed by .
In particular, the root MAY ignore the source address but it SHOULD use the RPLInstanceID for the computation.
Depending on the policy in place, the source of a packet will decide whether to use this specification
to transport the RPL information in the IPv6 packets. If it does, the source in the LLN SHOULD set the
Flow Label to zero and MUST NOT expect that the flow label will be conserved end-to-end".
Because the flow label is not protected by IPSec, it is expected that
Layer-2 security is deployed in the LLN where is specification is applied.
This is the actual best practice in LLNs, which serves in particular to
avoid forwarding of untrusted packets over the constrained network.
If the link layer is secured adequately, using the Flow Label as opposed
to the RPL option does not create an opening for a new threat compared to
.
No IANA action is required for this specification.
The author wishes to thank Brian Carpenter for his in-depth review and constructive approach to the problem resolution.IEEE std. 802.15.4, Part. 15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area NetworksIEEE standard for Information TechnologyISA100, Wireless Systems for AutomationISA