Approaches on Supporting IOAM in IPv6FutureweiUSAhaoyu.song@futurewei.comHuawei TechnologiesChinalizhenbin@huawei.comHuawei TechnologiesChinapengshuping@huawei.comFutureweiUSAjames.n.guichard@futurewei.com
OPS
IPPMIt has been proposed to encapsulate IOAM tracing option data fields in IPv6 HbH options header.
However, due to size of the trace data and the extension header location in the IPv6 packets,
the proposal creates practical challenges for implementation, especially when other extension headers,
such as routing header, also exist and require in-network processing.
We propose several alternative approaches to address this challenge, including
separating the IOAM trace data to a different extension header, using the postcard-based telemetry (e.g., IOAM DEX) instead,
and applying the segment IOAM trace export scheme, based on the network scenario and application requirements.
We discuss the pros and cons of each approach and hope to foster standard convergence and industry adoption. 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 RFC 2119.In-situ OAM (IOAM) defines two tracing options, pre-allocated tracing option and incremental tracing option,
which record hop-by-hop data along a packet's forwarding path.
The draft proposes the method to encapsulate IOAM trace option data fields
in IPv6. Because the tracing options requires per hop processing, such options can only be encapsulated in IPv6 Hop-by-Hop (HbH)
options header. The draft further describes some deployment approaches. mandates that the HbH options header, if exists, must be the first extension header following the IPv6 header.
However, the IOAM trace data can be large, which easily amount to tens to hundreds of bytes, making accessing other headers after it difficult.
There are practical limitations on how far the hardware can reach into a packet in forwarding hardware. The IOAM tracing option cannot
be applied if it makes other extension headers inaccessible. Even if the other headers can be
reached, the deeper they are, the higher the cost to access and process them, and the lower the forwarding performance.
A potentially more detrimental issue is that the incremental tracing option will expand the HbH header at each hop and push back all other headers,
which keeps shifting the locations of the other extension headers, further complicating the hardware implementation and impeding the forwarding.The issue becomes severe when SRv6 and IOAM coexist. The Segment Routing Extension Header (SRH) is encapsulated in a routing header which is after the HbH options header.
SRH itself can be large. It requires read and write operations at each SRv6 node. If it is deeply embedded in a packet
and its location keeps shifting, either it is beyond the reach of hardware or the forwarding performance degrades.We can avoid the problem by simply not using both at the same time, but apparently this is not ideal, because IOAM is an important OAM tool and it is
even more wanted when SRv6 brings more operational complexity into IPv6 networks.Our second recourse is to limit the IOAM to SRv6 nodes only. That is, consider SRv6 as an overlay tunnel over IPv6 and apply the IOAM pipe mode as discussed in
, which only collects data at each SRv6 nodes. To realize this, describes
an approach that encapsulates the IOAM option data fields in an SRH TLV. describes another approach to enable
postcard-based telemetry for SRv6 without needing IOAM option encapsulation.
In either case, the SRH is close to the packet front and its location is fixed.
While these approaches are useful for use cases that only need to monitor the segment performance, it fails to cover all the IPv6 nodes in a network.So the proposition of this draft is, suppose we need to apply IOAM on all nodes in an SRv6 network, how we can amend the approach in
or use alternative approaches to circumvent the aforementioned issues. In this draft, we
propose three such approaches:
(1) separating the IOAM trace data to a different extension header, (2) using the postcard-based telemetry (e.g., IOAM DEX) instead,
and (3) applying the segment IOAM trace export scheme.
We discuss the pros and cons of each approach and hope to foster standard convergence and industry adoption. An IOAM trace type data fields contain two parts: instruction and trace data. Although by convention the trace data part immediately follow
the instruction part, there is not fundamental reason why these two parts must stick together. This observation provides us an optimization
opportunity to amend the original proposal in . We separate the IOAM trace type data fields into the instruction part and the trace data part. We encapsulate only the instruction part in the
HbH options header, and encapsulate the trace data part in another metadata extension header after all the IPv6 extension headers
and before upper layer protocol headers.
This arrangement allows high performance hardware implementation. When using the incremental data trace, even if the data trace size increases at
each node, all IPv6 extension headers remain intact and new data is inserted at a fixed location.Figure 1 shows the HbH option format for IOAM trace type instruction. The field specification is identical to that in and
. Figure 2 shows the TLV option format for IOAM trace type data. The IOAM trace type data format is compliant with .We have basically two methods to encapsulate the IOAM trace data. First, we can define a new IPv6 extension header which is dedicated
to metadata. Once standardized, this extension header can also be used to host potential metadata from other applications such as NSH for SFC
. Second, this option can be carried as a TLV option in another existing extension header such as the destination header.
The only requirement is that this extension header should be the last one in the extension header chain.
The first method is cleaner but it requires extra standard effort; the second method is simpler but it needs to overcome the access constraints exerted
by .If the overhead of the IOAM trace type data fields is under control, we may still manage to
encapsulate both instruction and data in HbH options header as in . To this end, we introduce
two sub-approaches. proposes an enhancement to IOAM trace type which can configure the allowable overhead of the IOAM
trace type data fields. Once the trace data size is up to the limit at a network node (i.e., a segment or a fixed number of network nodes are
traversed), the trace data will be stripped and exported
so room is made to accommodate new trace data from nodes in the next segment of the forwarding path.This approach requires some moderate updates to the IOAM trace type data fields, as described in .
Figure 3 shows the format of the HbH Option Header containing Segment IOAM trace type data fields. A flag bit (#23) in
the Flags field is used to indicate the current header is a segment
IOAM header. In this context, the last octet in the IOAM header is
partitioned into two 4-bit nibbles. The first nibble (SSize) is used
to save the segment size and the second nibble (RHop) is used to save
the remaining hops. This limits the maximum segment size to 15.
At the beginning of each segment, the
segment size (SSize) and the remaining hops (RHop) are initialized: RHop is set to equal to SSize. At each
hop, if RHop is not zero, the node data is added to the node data list
and then RHop is decremented by 1. If RHop
is equal to 0 when receiving the packet, the node needs to remove (in
incremental trace option) or clear (in pre-allocated trace option)
the IOAM node data list and reset RHop to SSize.In this case, if we use the IOAM pre-allocated trace type, the size and location of each IPv6 extension header is fixed and predictable,
and the hardware capability and performance can be guaranteed.Whenever a packet with the IOAM option reaches a SRv6 node which needs to access the SRH, we can configure the node to export immediately the
IOAM trace data accumulated so far. In this case, basically at each SRv6 node, the HbH header size is fixed and the header contains an IOAM option with only
the instruction part. After the SRH processing, this node can add local IOAM trace data in the HbH option header before forwarding the packet.The incremental trace type can be used in this approach. In an extreme case when every node is also an SRv6 node, this approach regresses to a
per-hop postcard-based telemetry approach as described in . In this case, the HbH option
for IOAM can even be avoided altogether if we can find a way to simply mark the packet for postcard export.As an embodiment of the PBT-I approach introduced in , IOAM Direct Export (DEX) Option Type
discussed in can be used to replace the IOAM trace type. IOAM DEX only needs to encapsulate
a fix-size instruction header in the HbH option header. Figure 4 shows the HbH option format for IOAM DEX type fields. The field specification is identical to that in and
. The following table compares the existing approach and the four other alternative approaches proposed in this draft. TBD.