| Internet-Draft | fragment forwarding | November 2019 |
| Watteyne, et al. | Expires 30 May 2020 | [Page] |
This document introduces the capability to forward 6LoWPAN fragments. This method reduces the latency and increases end-to-end reliability in route-over forwarding. It is the companion to using virtual reassembly buffers which is a pure implementation technique.¶
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 30 May 2020.¶
Copyright (c) 2019 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. 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.¶
The original 6LoWPAN fragmentation is defined in [6LoWPAN] and it is implicitly defined for use over a single IP hop through possibly multiple Layer-2 (mesh-under) hops in a meshed 6LoWPAN Network. Although [6LoWPAN-HC] updates [6LoWPAN], it does not redefine 6LoWPAN fragmentation.¶
This means that over a Layer-3 (route-over) network, an IP packet is expected to be reassembled at every hop at the 6LoWPAN sublayer, pushed to Layer-3 to be routed, and then fragmented again if the next hop is another similar 6LoWPAN link. This draft introduces an alternate approach called 6LoWPAN Fragment Forwarding (FF) whereby an intermediate node forwards a fragment as soon as it is received if the next hop is a similar 6LoWPAN link. The routing decision is made on the first fragment, which has all the IPv6 routing information. The first fragment is forwarded immediately and a state is stored to enable forwarding the next fragments along the same path.¶
Done right, 6LoWPAN Fragment Forwarding techniques lead to more streamlined operations, less buffer bloat and lower latency. It may be wasteful if some fragments are missing after the first one since the first fragment will still continue till the 6LoWPAN endpoint that will attempt to perform the reassembly, and may be misused to the point that performances fall behind that of per-hop recomposition. This specification provides a generic overview of FF, discusses advantages and caveats, and introduces a particular 6LoWPAN Fragment Forwarding technique called Virtual Reassembly Buffer that can be used while conserving the message formats defined in [6LoWPAN].¶
We use Figure 1 to illustrate 6LoWPAN fragmentation. We assume node A forwards a packet to node B, possibly as part of a multi-hop route between IPv6 source and destination nodes which are neither A nor B.¶
+---+ +---+
... ---| A |-------------------->| B |--- ...
+---+ +---+
# (frag. 5)
123456789 123456789
+---------+ +---------+
| # ###| |### # |
+---------+ +---------+
outgoing incoming
fragmentation reassembly
buffer buffer
Node A starts by compacting the IPv6 packet using the header compression mechanism defined in [6LoWPAN-HC]. If the resulting 6LoWPAN packet does not fit into a single Link-Layer frame, node A's 6LoWPAN sublayer cuts it into multiple 6LoWPAN fragments, which it transmits as separate Link-Layer frames to node B. Node B's 6LoWPAN sublayer reassembles these fragments, inflates the compressed header fields back to the original IPv6 header, and hands over the full IPv6 packet to its IPv6 layer.¶
In Figure 1, a packet forwarded by node A to node B is cut into nine fragments, numbered 1 to 9 as follows:¶
The reassembly buffer for 6LoWPAN is indexed in node B by:¶
Because it may be hard for node B to correlate all possible Link-Layer addresses that node A may use (e.g., short vs. long addresses), node A must use the same Link-Layer address to send all the fragments of the same datagram to node B.¶
Conceptually, the reassembly buffer in node B contains:¶
A fragmentation header is added to each fragment; it indicates what portion of the packet that fragment corresponds to. Section 5.3 of [6LoWPAN] defines the format of the header for the first and subsequent fragments. All fragments are tagged with a 16-bit "datagram_tag", used to identify which packet each fragment belongs to. Each datagram can be uniquely identified by the sender Link-Layer addresses of the frame that carries it and the datagram_tag that the sender allocated for this datagram. [6LoWPAN] also mandates that the first fragment is sent first and with a particular format that is different than that of the next fragments. Each fragment but the first one can be identified within its datagram by the datagram-offset.¶
Node B's typical behavior, per [6LoWPAN], is as follows. Upon receiving a fragment from node A with a datagram_tag previously unseen from node A, node B allocates a buffer large enough to hold the entire packet. The length of the packet is indicated in each fragment (the datagram_size field), so node B can allocate the buffer even if the first fragment it receives is not fragment 1. As fragments come in, node B fills the buffer. When all fragments have been received, node B inflates the compressed header fields into an IPv6 header, and hands the resulting IPv6 packet to the IPv6 layer which performs the route lookup. This behavior typically results in per-hop fragmentation and reassembly. That is, the packet is fully reassembled, then (re)fragmented, at every hop.¶
There are at least 2 limits to doing per-hop fragmentation and reassembly. See [ARTICLE] for detailed simulation results on both limits.¶
When reassembling, a node needs to wait for all the fragments to be received before being able to generate the IPv6 packet, and possibly forward it to the next hop. This repeats at every hop.¶
This may result in increased end-to-end latency compared to a case where each fragment is forwarded without per-hop reassembly.¶
Constrained nodes have limited memory. Assuming a reassembly buffer for a 6LoWPAN MTU of 1280 bytes as defined in section 4 of [6LoWPAN], typical nodes only have enough memory for 1-3 reassembly buffers.¶
To illustrate this we use the topology from Figure 2, where nodes A, B, C and D all send packets through node E. We further assume that node E's memory can only hold 3 reassembly buffers.¶
+---+ +---+
... --->| A |------>| B |
+---+ +---+\
\
+---+ +---+
| E |--->| F | ...
+---+ +---+
/
/
+---+ +---+
... --->| C |------>| D |
+---+ +---+
When nodes A, B and C concurrently send fragmented packets, all 3 reassembly buffers in node E are occupied. If, at that moment, node D also sends a fragmented packet, node E has no option but to drop one of the packets, lowering end-to-end reliability.¶
A 6LoWPAN Fragment Forwarding technique makes the routing decision on the first fragment, which is always the one with the IPv6 address of the destination. Upon a first fragment, a forwarding node (e.g. node B in a A->B->C sequence) that does fragment forwarding MUST attempt to create a state and forward the fragment. This is an atomic operation, and if the first fragment cannot be forwarded then the state MUST be removed. When a forwarding node receives a fragment other than a first fragment, it MUST look up state based on the source Link-Layer address and the datagram_tag in the received fragment. If no such state is found, the fragment MUST be dropped; otherwise the fragment MUST be forwarded using the information in the state found. Since the datagram_tag is uniquely associated to the source Link-Layer address of the fragment, the forwarding node MUST assign a new datagram_tag from its own namespace for the next hop and rewrite the fragment header of each fragment with that datagram_tag.¶
Compared to Section 2, the conceptual reassembly buffer in node B now contains, assuming that node B is neither the source nor the final destination:¶
A node that has not received the first fragment cannot forward the next fragments. This means that if node B receives a fragment, node A was in possession of the first fragment at some point. In order to keep the operation simple, it makes sense to be consistent with [6LoWPAN] and enforce that the first fragment is always sent first. When that is done, if node B receives a fragment that is not the first and for which it has no state, then node B treats this as an error and refrain from creating a state or attempting to forward. This also means that node A should perform all its possible retries on the first fragment before it attempts to send the next fragments, and that it should abort the datagram and release its state if it fails to send the first fragment.¶
One benefit of Fragment Forwarding is that the memory that is used to store the packet is now distributed along the path, which limits the buffer bloat effect. Multiple fragments may progress in parallel along the network as long as they do not interfere. An associated caveat is that on a half duplex radio, if node A sends the next fragment at the same time as node B forwards the previous fragment to a node C down the path then node B will miss the next fragment. If node C forwards the previous fragment to a node D at the same time and on the same frequency as node A sends the next fragment to node B, this may result in a hidden terminal problem at B whereby the transmission from C interferes with that from A unbeknownst of node A. It results that consecutive fragments must be reasonably spaced in order to avoid the 2 forms of collision described above. A node that has multiple packets or fragments to send via different next-hop routers may interleave the messages in order to alleviate those effects.¶
Virtual Reassembly Buffer (VRB) is the implementation technique described in [LWIG-VRB] in which a forwarder does not reassemble each packet in its entirety before forwarding it.¶
VRB overcomes the limits listed in Section 3. Nodes do not wait for the last fragment before forwarding, reducing end-to-end latency. Similarly, the memory footprint of VRB is just the VRB table, reducing the packet drop probability significantly.¶
There are, however, limits:¶
The severity and occurrence of these limits depends on the Link-Layer used. Whether these limits are acceptable depends entirely on the requirements the application places on the network.¶
If the limits are present and not acceptable for the application, future specifications may define new protocols to overcome these limits. One example is [FRAG-RECOV] which defines a protocol which allows fragment recovery.¶
Secure joining and the Link-Layer security that it sets up protects against those attacks from network outsiders.¶
"IP Fragmentation Considered Fragile" [FRAG-ILE] discusses security threats that are linked to using IP fragmentation. The 6LoWPAN fragmentation takes place underneath, but some issues described there may still apply to 6lo fragments.¶
No requests to IANA are made by this document.¶
The authors would like to thank Yasuyuki Tanaka, Ines Robles and Dave Thaler for their in-depth review of this document and improvement suggestions. Also many thanks to Georgies Papadopoulos and Dominique Barthel for their own reviews.¶