Reviewer: Bernard Aboba
Review result: Ready with Nits
This document has been reviewed as part of the transport area review team's
ongoing effort to review key IETF documents. These comments were written
primarily for the transport area directors, but are copied to the document's
authors and WG to allow them to address any issues raised and also to the IETF
discussion list for information.
When done at the time of IETF Last Call, the authors should consider this
review as part of the last-call comments they receive. Please always CC
tsv-art@xxxxxxxx if you reply to or forward this review.
Document: draft-ietf-mboned-multicast-telemetry
Reviewer: Bernard Aboba
Result: Ready with Nits
This document explains the deficiencies in existing OAM techniques and lays out
proposals to address them. However, it doesn't provide much detail on some of
the reasoning. Also in places, the wording could use improvement.
Comments
----
1. Introduction
Multicast has many use cases. For example, it can be used by
residential broadband customers across operator networks, private
MPLS customers, and internal customers within corporate intranet.
[BA] s/intranet/intranets/
Multicast provides real time interactive online meetings or podcasts,
[BA] Use of multicast for conferencing is rare nowadays. Do you really
want to include this?
IPTV, and financial markets real-time data, which all have a reliance
on UDP's unreliable transport. End-to-end QOS, therefore, should be
a critical component of multicast deployment in order to provide a
good end user experience. In multicast real-time media streaming,
loss of a single packet containing a reference frame can result in
the inability of thousands of receivers to decode a whole sequence of
packets called Group-of-Picture, introducing black picture for
periods of a few seconds.
Unexpected long delay in propagation of a
packet in such real-time media streaming may equally result in the
packet not being received and create the same results. Multicast
packet drops and delay can therefore severely affect the application
performance and user experience.
[BA] Suggest:
"In multicast real-time media streaming, if a single packet is lost
within a keyframe and cannot be recovered using forward
error correction, this can result in many receivers being unable
to decode subsequent frames within the Group of Pictures (GoP), resulting
in video freezes or black pictures until another keyframe is
delivered.
Unexpectedly long delays in delivery of packets can result in
timeouts within similar results. Multicast packet loss and
delays can therefore affect application performance and the
user experience."
It is important to monitor the performance of the multicast traffic.
New on-path telemetry techniques such as In-situ OAM (IOAM)
[RFC9197], IOAM Direct Export (DEX) [RFC9326] IOAM Marking-based
Postcard (PBT-M) [I-D.song-ippm-postcard-based-telemetry], and Hybrid
Two-Step (HTS) [I-D.ietf-ippm-hybrid-two-step] are useful and
complementary to the existing active OAM performance monitoring
methods (e.g., ICMP ping [RFC0792]),
provide promising means to
directly monitor the network experience of multicast traffic.
However, multicast traffic has some unique characteristics which pose
some challenges on applying such techniques in an efficient way.
Suggest:
"providing a way to monitor multicast performance. However, multicast
has unique characteristics that make the efficient application
of these techniques challenging."
2. Requirements for Multicast Traffic Telemetry
Multicast traffic is forwarded through a multicast tree. With PIM
and P2MP, the forwarding tree is established and maintained by the
multicast routing protocol. With BIER, no state is created in the
network to establish a forwarding tree; instead, a bier header
provides the necessary information for each packet to know the egress
points. Multicast packets are only replicated at each tree branch
fork node for efficiency.
There are several requirements for multicast traffic telemetry, a few
of which are:
* Reconstruct and visualize the multicast tree through data plane
monitoring.
* Gather the multicast packet delay and jitter performance on each
path.
* Find the multicast packet drop location and reason.
* Gather the VPN state and tunnel information in case of P2MP
multicast.
In order to meet these requirements, we need the ability to directly
monitor the multicast traffic and derive data from the multicast
packets. The conventional OAM mechanisms, such as multicast ping
[RFC6450] and trace [RFC8487], are not sufficient to meet these
requirements.
{BA] Can you provide more detail on why existing mechanisms are not sufficient?
When conventional mechanisms are combined with RTCP, it seems like the first
three requirements are covered.
3. Issues of Existing Techniques
On-path Telemetry techniques that directly retrieve data from
multicast traffic's live network experience are ideal for addressing
the aforementioned requirements. The representative techniques
include In-situ OAM (IOAM) Trace option [RFC9197], IOAM Direct Export
(DEX) option [RFC9326], and PBT-M
[I-D.song-ippm-postcard-based-telemetry]. However, unlike unicast,
multicast poses some unique challenges to applying these techniques.
Multicast packets are replicated at each branch fork node in the
corresponding multicast tree. Therefore, there are multiple copies
of the original multicast packet in the network.
If the IOAM trace option is used for on-path data collection, the
partial trace data will also be replicated into the packet copy for
each branch. The end result is that, at the multicast tree leaves,
each copy of the multicast packet has a complete trace. Most of the
data (except data from the last leaf branch) appear in multiple
copies while only one copy is sufficient. Data redundancy introduces
unnecessary header overhead, wastes network bandwidth, and
complicates the data processing. The larger the multicast tree, or
the longer the multicast path, the more severe the redundancy problem
becomes.
The postcard-based solutions (e.g., IOAM DEX), can be used to
eliminate such data redundancy, because each node on the tree only
sends a postcard covering local data. However, they cannot track and
correlate the tree branches properly due to the lack of branching
information, so they can bring confusion about the multicast tree
topology. For example, in a multicast tree, Node A has two branches,
one to Node B and the other to node C; further, Node B leads to Node
D and Node C leads to Node E. When applying postcard-based methods,
one cannot tell whether or not Node D(E) is the next hop of Node B(C)
from the received postcards alone, unless one correlates the
exporting nodes with knowledge about the tree collected by other
means (e.g., mtrace). Such correlation is undesirable because it
introduces extra work and complexity.
The fundamental reason for this problem is that there is not an
identifier (either implicit or explicit) to correlate the data on
each branch.
[BA] Can't the IP address be used as an identifier? Does the proposed
solution address this issue in a simpler way? Note that "extra work"
(e.g. new software) that is implemented outside network devices has
an advantage over new protocols which can conceivably impact network
device footprint and reliability (due to bugs). So the where and how
matters with respect to "extra work and complexity".
--
last-call mailing list -- last-call@xxxxxxxx
To unsubscribe send an email to last-call-leave@xxxxxxxx
