Hi Zhenqiang and Thomas,
Thanks very much for sharing your experience and the discussion in the
mailing list.
IMO, whether to use gRPC or IPFIX is not a technique problem, but
different choice/preference from users.
I agree with Zhenqiang our intention is not to have many options for the
same feature.
But the world is diverse. Like we always have the same IGP extension for
both OSPF and ISIS. At present I cannot see the dominance of one protocol.
I have practice on both gRPC and IPFIX. I believe both approaches can
achieve this case.
Because this is neither large volume nor frequent.
The fast path will process delay per-packet, but the export will only
work on mean/min/max delay.
Best,
Tianran
*From:*li_zhenqi...@hotmail.com [mailto:li_zhenqi...@hotmail.com]
*Sent:* Wednesday, January 11, 2023 4:23 PM
*To:* Benoit Claise <benoit.cla...@huawei.com>;
thomas.g...@swisscom.com; Tianran Zhou <zhoutian...@huawei.com>; ippm
<i...@ietf.org>
*Cc:* opsawg@ietf.org
*Subject:* Re: Re: [ippm] 回复: FW: WG Adoption Call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01
Hello Mr. Claise,
please see my comments beginning with [ZQ].
------------------------------------------------------------------------
Best Regards,
Zhenqiang Li
li_zhenqi...@hotmail.com <mailto:li_zhenqi...@hotmail.com>
*发件人:*Benoit Claise <mailto:benoit.cla...@huawei.com>
*发送时间:* 2023-01-10 05:15
*收件人:*thomas.g...@swisscom.com
<mailto:thomas.g...@swisscom.com>; li_zhenqi...@hotmail.com
<mailto:li_zhenqi...@hotmail.com>;
zhoutianran=40huawei....@dmarc.ietf.org
<mailto:zhoutianran=40huawei....@dmarc.ietf.org>; i...@ietf.org
<mailto:i...@ietf.org>
*抄送:*opsawg@ietf.org <mailto:opsawg@ietf.org>
*主题:* Re: [ippm] 回复: FW: WG Adoption Call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01
Dear all,
I agree with Thomas on a couple of points.
What is important is that, for accounting information, monitored at
high frequency, such as IPFIX flow records metering data plane
traffic, should be exported directly from line cards, without
proxying to the route processor. As such, UDP is the way to go (*).
On top of that, losing an export packet or two is not the end of the
world.
In this case (draft-tgraf-opsawg-ipfix-on-path-telemetry-01), where
we monitor delay over (all packets of) the flow record lifetime,
it's the same principle: we need the UDP-based export. So the IPFIX
protocol.
[ZQ] I don't think it is apropriate to limit the implementation
especailly the internal software and hardware architecture when we
define protocol. Our lab tests show gRPC works well for exporting
the iOAM metrics and gRPC is TCP based. I am not sure whether or not
our vendors implement this by proxying the collected data plane
information to the route processor and then packeting and exporting
all the relevant information to the colloctor. Why do we need to
tell the vendors how to do it? From the specification point, IPFIX
does not require UDP, TCP based SCTP is a valid option.
[ZQ] I agree with you that the packets of iOAM flow should not be
sampled, i.e. all packets of the flow need to be monitored.
Btw, the flow record lifetime is configurable, based on timers... in
case you want to act on the delay (in the flow record) faster... at
the cost of exporting more flow records.
[ZQ] iOAM metrics should be exported in seconds, in minutes is not
acceptable. IPFIX exporting timer is usually in minutes.
Now, if we speak about events, this is a different story, as events
needs to acted upon. So a reliable transport is required here.
Let's keep in mind that we speak about hybrid type I (see
https://www.rfc-editor.org/rfc/rfc7799#section-3.8
<https://www.rfc-editor.org/rfc/rfc7799#section-3.8>), so the issue
of isolating the specific packets to look at at intermediate node is
trivial.
(*) and this is the reason why SCTP, the compulsory export protocol
in RFC 7011 was never a success. See
https://www.claise.be/from-netflow-to-ipfix-via-psamp-13-years-of-standardization-explained-2/
<https://www.claise.be/from-netflow-to-ipfix-via-psamp-13-years-of-standardization-explained-2/>
Regards, Benoit
On 1/9/2023 11:06 AM, thomas.g...@swisscom.com
<mailto:thomas.g...@swisscom.com> wrote:
Dear Zhenqiang,
See response inline.
Best wishes
Thomas
*From:*li_zhenqi...@hotmail.com
<mailto:li_zhenqi...@hotmail.com> <li_zhenqi...@hotmail.com>
<mailto:li_zhenqi...@hotmail.com>
*Sent:* Monday, January 9, 2023 8:43 AM
*To:* Graf Thomas, INI-NET-VNC-HCS <thomas.g...@swisscom.com>
<mailto:thomas.g...@swisscom.com>; Tianran Zhou
<zhoutianran=40huawei....@dmarc.ietf.org>
<mailto:zhoutianran=40huawei....@dmarc.ietf.org>; ippm
<i...@ietf.org> <mailto:i...@ietf.org>
*Cc:* opsawg@ietf.org <mailto:opsawg@ietf.org>
*Subject:* Re: RE: [ippm] 回复: FW: WG Adoption Call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01
DearThomas,
Please see my further discussion beginning with [ZQ].
------------------------------------------------------------------------
Best Regards,
Zhenqiang Li
li_zhenqi...@hotmail.com <mailto:li_zhenqi...@hotmail.com>
*发件人:*thomas.g...@swisscom.com
<mailto:thomas.g...@swisscom.com>
*发送时间:* 2023-01-07 01:23
*收件人:*li_zhenqi...@hotmail.com
<mailto:li_zhenqi...@hotmail.com>;
zhoutianran=40huawei....@dmarc.ietf.org
<mailto:zhoutianran=40huawei....@dmarc.ietf.org>;
i...@ietf.org <mailto:i...@ietf.org>
*抄送:*opsawg@ietf.org <mailto:opsawg@ietf.org>
*主题:* RE: RE: [ippm] 回复: FW: WG Adoption Call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01
Dear Zhenqiang,
Thanks a lot for the feedback. Much appreciated.
I do not disagree that YANG push isn't capable of exporting
control and forwarding plane metrics. However it is not the
best choice in terms of scale. Table 1 of RFC 9232 gives a
good summary. It even makes the distinction between network
processor and route processor export. What it doesn't show
however is that gRPC isn't suitable for network processors.
To date, no vendor implementations are present. From my
research with network and ASIC vendors as being the
co-author of draft-ietf-netconf-distributed
(https://datatracker.ietf.org/doc/html/draft-ietf-netconf-distributed-notif
<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fdoc%2Fhtml%2Fdraft-ietf-netconf-distributed-notif&data=05%7C01%7CThomas.Graf%40swisscom.com%7C502beb5de8ee458ffbcb08daf214f30f%7C364e5b87c1c7420d9beec35d19b557a1%7C0%7C0%7C638088469032680658%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=sA%2FvpsWKZOlBaLZEvBESmZZ8WFMKsw2Az7Wf54Og8KQ%3D&reserved=0>)
I understood that gRPC won't be implementable in its current form. One of the main reason is the
lack of backpressure support on network processors.
I fully agree that export for probing metrics such a TWAMP,
TWAMP light or STAMP makes perfectly sense. With
draft-ietf-ippm-stamp-yang and RFC 8913 we have well defined
YANG modules. Speaking as network operator with approx. 7000
YANG publishers I can confirm that this works well. Either
when both probing and export runs on the route processor or
both on the network processor directly.
[ZQ] For export implementaion of data plane iOAM, we'd
better hear the voice from vendors. Some co-authors of this
doc are from vendors. As I stated in the previous mail,
almost all our vendors have already supported exporting
passport iOAM metrics by gRPC with GPB encoding. We have
finished lab tests with good results and have plan to deploy
this kind of data plane iOAM in our field network this year.
As for the implementation, I don't know the details. There
are several other processors on the line card besides
network processor.
[TG] The network processor is the processor which forwards
the packet and export the forwarding plane metrics. Ideally
directly without proxying over the route processor. We are
in contact with major vendors since over a year and have two
working implementations and working on another technical
feasibility. The results will be presented at IETF 116
hackathon and documented in the next version.
·In my understanding, the aggregation benifit of IPFIX is
not applicable for iOAM metrics, such as delay and packet
loss, because iOAM metrics are different for different flows.
This I don't understand fully. Could you describe more
detailed please.
To my understanding and maybe this helps to establish a
clearer picture. IOAM is applied to existing or replicated
packets and is classified as One-Way Delay Hybrid Type I
according to Section 1 of draft-ietf-ippm-ioam-deployment
(https://datatracker.ietf.org/doc/html/draft-ietf-ippm-ioam-deployment#section-1
<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fdoc%2Fhtml%2Fdraft-ietf-ippm-ioam-deployment%23section-1&data=05%7C01%7CThomas.Graf%40swisscom.com%7C502beb5de8ee458ffbcb08daf214f30f%7C364e5b87c1c7420d9beec35d19b557a1%7C0%7C0%7C638088469032680658%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=94g93GZq4TB7diC2nK5TcXCLBESnS6mUfITwKyq08Q8%3D&reserved=0>).
There are two modes. Active and Passive. In passive, where
draft-tgraf-opsawg-ipfix-on-path-telemetry focuses on, existing packets are being used. Where in
active mode, existing packets are replicated. IPFIX takes the packets and accounts them into
flows. In flows we account bytes and packets over a given time period. With
draft-tgraf-opsawg-ipfix-on-path-telemetry we also account for delay.
[ZQ] I am mainly talking about passive mode. Bytes and
packets can be accumulated for flows. But I can not imagine
how to accumulate delays for flows? For my understanding,
aggregation is different from accumulation. For example,
Bytes and packets accumulated for different flows can be
further aggregated on destination prefix if configured and
if the destinations of the flows belong to one common route
entry in the routing talbe. Since delay metric may be
different for different flows, it is not appropriate to
aggregate this metric based on destionation, AS, community etc.
[TG] As mentioned in my previous email. The delay is
accounted per flow. Section 1
(https://datatracker.ietf.org/doc/html/draft-tgraf-opsawg-ipfix-on-path-telemetry-01#section-1
<https://datatracker.ietf.org/doc/html/draft-tgraf-opsawg-ipfix-on-path-telemetry-01#section-1>)
describes how the delay is being measured. I am quoting:
The delay is measured by calculating the difference between
the timestamp imposed with On-Path Telemetry in the packet
at the IOAM encapsulation node and the timestamp exported in
the IPFIX flow record from the IOAM transit and
decapsulation nodes. Depending on the IE, the lowest,
highest or the sum of measured path delay is being exported.
IOAM-Domain
.........................................
. .
. D1 .
. <------> .
. .
. D2 .
. <--------------------> .
. .
. D3 .
. <-----------------------------------> .
. .
(H1) ------ (R1) ------- (R2) ------- (R3) -------- (R4)
------ (H2)
Host 1 Encapsulation Transit Transit
Decapsulation Host 2
Node Node 1 Node 2 Node
. .
. .
.........................................
Figure 1: IOAM Delay use case. Packets flow from host 1 to
host 2.
On the usecase showed in Figure 1 using IOAM to export the
delay metrics, the node R2 exports the delay D1, the node R3
exports the delay D2 and the decapsulation node R4 exports
the total delay D3 using IPFIX.
[TG] The same as IPFIX aggregates for packets and bytes, the
sum of the delay for all flows is accounted in
PathDelaySumDeltaMicroseconds/PathDelaySumDeltaNanoseconds.
The
PathDelayMeanDeltaMicroseconds/PathDelayMeanDeltaNanoseconds
is calculated as described in Section 7.2
https://datatracker.ietf.org/doc/html/draft-tgraf-opsawg-ipfix-on-path-telemetry-01#section-7.2
<https://datatracker.ietf.org/doc/html/draft-tgraf-opsawg-ipfix-on-path-telemetry-01#section-7.2>.
·Neither sampling, how can you insure that the flow you want
its iOAM metrics will be sampled?
Packets have to be selected in the IOAM encapsulation node
as described in Section 7.4 of
draft-ietf-ippm-ioam-deployment
https://datatracker.ietf.org/doc/html/draft-ietf-ippm-ioam-deployment#section-7.4
<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fdoc%2Fhtml%2Fdraft-ietf-ippm-ioam-deployment%23section-7.4&data=05%7C01%7CThomas.Graf%40swisscom.com%7C502beb5de8ee458ffbcb08daf214f30f%7C364e5b87c1c7420d9beec35d19b557a1%7C0%7C0%7C638088469032680658%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=GgV98jt9iibX3ctGw02wMBl81nbjabGxkDg6GL%2F8QhE%3D&reserved=0>.
IOAM doesn't specify how the packets are being selected. To my opinion it would make sense to
refer in Section 7.4 of draft-ietf-ippm-ioam-deployment to RFC 5474 and RFC 5475 which are
describing packet selection and sampling techniques, both widely used with IPFIX.
The benefit of sampling at the IOAM encapsulation node and
flag with IOAM-DEX versus sampling on all nodes is that the
amount of flows being exported is the same, but in case of
IOAM-DEX we know the forwarding path in addition without
additional overhead in the data export.
[ZQ] Yes, agree that packets have to be selected for iOAM
but must be flow based and must not be further sampled for
the selected flows in my understanding because some underlay
supporting mechanisms such as alternative marking require
consecutive packets. Usually we configure sampling rate for
IPFIX, 1:5000 for example. But we can not insure that the
packet sampled on node1 will be exactly sampled at the next
node on its traverse path. Thus it is difficult to get the
delay metric when working in this way.
[TG] Exactly. The difference to traditional IPFIX is that in
IOAM only samples/selects once, at the edge on the IOAM
encapsulation node.
·The iOAM metrics, especially the delay and packet loss,
should be exported in time, for example in one second. I
don't think IPFIX is the right protocol.
For packet loss, you are correct, IPFIX is not the right
protocol since packet loss is observed at the edges of the
observation domain (black box monitoring). Packet loss is
observed with probing where YANG push is the best choice for
data export. IPFIX is being used to visualize where the
packets are being dropped for which flow with IE89
ForwardingStatus.
[ZQ] Again, it is better to export all the data plane iOAM
metrics by one protocol if possible.
IPFIX can export metrics without aggregation where flows are
immediately being exported at ingress or egress. It is
correct that aggregation introduces delay. Therefore it
should only be applied to use cases which make sense
https://www.rfc-editor.org/rfc/rfc7015#section-3
<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.rfc-editor.org%2Frfc%2Frfc7015%23section-3&data=05%7C01%7CThomas.Graf%40swisscom.com%7C502beb5de8ee458ffbcb08daf214f30f%7C364e5b87c1c7420d9beec35d19b557a1%7C0%7C0%7C638088469032680658%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=D2o7UpmX5sG5cvM9auSL%2F5%2BiENpfkZuNICY4UzRBZis%3D&reserved=0>.
What I like to highlight is that in IPFIX supports an inactice timeout window. This allows the
data export to be exported as quickly as possibly without waiting to finish the active timeout
window.
[ZQ] As I know, there is no immediate export working mode
for IPFIX. IPFIX exports flow information when the flow is
aged or over, when the configured export interval expires.
At present, the export interval is in minute which can not
satisfy the time requirement of iOAM.
[TQ] As described previously described IPFIX indeed supports
it by default. RFC 7011 doesn't mandate aggregation. RFC
7015 describes this in section 4.2 figure 4
https://www.rfc-editor.org/rfc/rfc7015#section-4.2
<https://www.rfc-editor.org/rfc/rfc7015#section-4.2>. Here
an example how this is configured in Cisco IOS XR
(https://www.cisco.com/c/en/us/td/docs/routers/asr9000/software/netflow/command/reference/b-netflow-cr-asr9k/netflow-commands.html#wp1580058501
<https://www.cisco.com/c/en/us/td/docs/routers/asr9000/software/netflow/command/reference/b-netflow-cr-asr9k/netflow-commands.html#wp1580058501>).
There are many other vendor implementations.
Therefore using aggregation depends on the use case, the
objective. Speaking for a network operator with a large
scale Network Telemetry pipeline one of the key objectives
is to have a wide view across the network while minimizing
the amount of data as much and as early in the pipeline as
possible. For automated traffic verification of several
thousands of L3 VPN's a highly aggregated data set makes
more sense then when troubleshooting of a particular
customer flow across the network is needed. Here cardinality
counts. Therefore we use with RFC7015 a two step data
aggregation concept. One where the aggregation is performed
on the network node without loosing cardinality. Where at
data-collection through replication a second aggregation is
performed where cardinality is reduced to a minimum.
Allowing latency queries on the data set.
Looking forward to your comments.
Best wishes
Thomas
*From:*li_zhenqi...@hotmail.com
<mailto:li_zhenqi...@hotmail.com> <li_zhenqi...@hotmail.com
<mailto:li_zhenqi...@hotmail.com>>
*Sent:* Friday, January 6, 2023 3:51 PM
*To:* Graf Thomas, INI-NET-VNC-HCS <thomas.g...@swisscom.com
<mailto:thomas.g...@swisscom.com>>; Tianran Zhou
<zhoutianran=40huawei....@dmarc.ietf.org
<mailto:zhoutianran=40huawei....@dmarc.ietf.org>>; ippm
<i...@ietf.org <mailto:i...@ietf.org>>
*Cc:* opsawg@ietf.org <mailto:opsawg@ietf.org>
*Subject:* Re: RE: [ippm] 回复: FW: WG Adoption Call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01
Hello Thomas,
As summarized in table 1 in RFC9232, gRPC is an application
protocol, which can be used to export all the telemetry
metrics for Management Plane, Control Plane, Forwarding
Plane and External Data. By using GPB encoding, gRPC is
widely used to export real time performace metrics, such as
link delay, jitter, packet loss measured by TWAMP or STAMP.
Almost all our vendors support this way to export passport
iOAM metrics and the test results in our lab are good. We
have plan to deploy it in our field network soon.
In my understanding, the aggregation benifit of IPFIX is not
applicable for iOAM metrics, such as delay and packet loss,
because iOAM metrics are different for different flows.
Neither sampling, how can you insure that the flow you want
its iOAM metrics will be sampled? The iOAM metrics,
especially the delay and packet loss, should be exported in
time, for example in one second. I don't think IPFIX is the
right protocol.
I do not object this doc to move forward if we can reach
this consensus.
------------------------------------------------------------------------
Best Regards,
Zhenqiang Li
li_zhenqi...@hotmail.com <mailto:li_zhenqi...@hotmail.com>
*发件人:*thomas.g...@swisscom.com
<mailto:thomas.g...@swisscom.com>
*发送时间:* 2023-01-04 00:20
*收件人:*li_zhenqi...@hotmail.com
<mailto:li_zhenqi...@hotmail.com>;
zhoutianran=40huawei....@dmarc.ietf.org
<mailto:zhoutianran=40huawei....@dmarc.ietf.org>;
i...@ietf.org <mailto:i...@ietf.org>
*抄送:*opsawg@ietf.org <mailto:opsawg@ietf.org>
*主题:* RE: [ippm] 回复: FW: WG Adoption Call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01
Dear Zhenqiang,
Thanks a lot for your feedback.
I presume with gRPC you are referring to YANG push (RFC
8639, RFC 8641, draft-ietf-netconf-udp-notif,
draft-ietf-netconf-https-notif). gNMI (gRPC is the
transport of gNMI) has been proposed
(draft-openconfig-rtgwg-gnmi-spec) in 2018 but not
standardized within IETF as a transport protocol for
YANG push. It did not find enough traction. A good
overview about the state of the union about YANG push in
the industry is a presentation I gave at the IEPG, slide
3-6,
http://www.iepg.org/2022-11-06-ietf115/slides-115-iepg-sessa-network-operator-challenges-in-network-telemetry-data-mesh-integration-thomas-graf-00.pdf
<https://eur03.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.iepg.org%2F2022-11-06-ietf115%2Fslides-115-iepg-sessa-network-operator-challenges-in-network-telemetry-data-mesh-integration-thomas-graf-00.pdf&data=05%7C01%7CThomas.Graf%40swisscom.com%7C502beb5de8ee458ffbcb08daf214f30f%7C364e5b87c1c7420d9beec35d19b557a1%7C0%7C0%7C638088469032680658%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=9usAacXd2s7rhe1h2xxMnjdXVzbqSU%2BgOf57k01jO5I%3D&reserved=0>.
RFC 9232 gives a good overview about Network Telemetry.
In section 3.1.3
(https://datatracker.ietf.org/doc/html/rfc9232#section-3.1.3
<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fdoc%2Fhtml%2Frfc9232%23section-3.1.3&data=05%7C01%7CThomas.Graf%40swisscom.com%7C502beb5de8ee458ffbcb08daf214f30f%7C364e5b87c1c7420d9beec35d19b557a1%7C0%7C0%7C638088469032680658%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=HGGNAm4pzdE2cV0BNuy2Kzjq1G8xysgqizEQ7E1LSNI%3D&reserved=0>)
is a summary of the protocols for forwarding plane data collection. As you pointed out, it makes
sense to use one common Network Telemetry protocol. However today we have 3, IPFIX for data
plane, BMP for BGP routing control-plane and YANG push for management-plane. These 3 protocols
have different unique mandatory characteristics. IPFIX allows data reduction with sampling
(RFC5476) an aggregation (RFC7015). While BMP allows to mirror BGP PDU's (lightweight) and add
device dimensions such as peering, RIB and route-policy (context). And YANG push allows through
its sophisticated data modelling to have configurational and operational metrics within a single
data model.
In OPSAWG presentation of IETF 115
(https://datatracker.ietf.org/meeting/115/materials/slides-115-opsawg-export-of-on-path-delay-in-ipfix-00.pdf
<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fmeeting%2F115%2Fmaterials%2Fslides-115-opsawg-export-of-on-path-delay-in-ipfix-00.pdf&data=05%7C01%7CThomas.Graf%40swisscom.com%7C502beb5de8ee458ffbcb08daf214f30f%7C364e5b87c1c7420d9beec35d19b557a1%7C0%7C0%7C638088469032680658%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=YZ99wLc%2BMdCmXbYk1LLgyPWYFy5dVuLig8TdR1FIZzs%3D&reserved=0>)
I described in slide 2 and 3 the benefit of adding the delay measurements to IPFIX. Having one
single protocol for data-plane data collection. The ability to perform data correlation and
aggregation with an existing proven IPFIX protocol. Does that resonate with you?
The use cases are described in section 5 of the draft
(https://datatracker.ietf.org/doc/html/draft-tgraf-opsawg-ipfix-on-path-telemetry-01#section-5
<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fdoc%2Fhtml%2Fdraft-tgraf-opsawg-ipfix-on-path-telemetry-01%23section-5&data=05%7C01%7CThomas.Graf%40swisscom.com%7C502beb5de8ee458ffbcb08daf214f30f%7C364e5b87c1c7420d9beec35d19b557a1%7C0%7C0%7C638088469032680658%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=g4f1cnIkv6OWD%2BSFlxq%2F43xhgyBCiP2%2BcgXQtr5YFpI%3D&reserved=0>).
Able to see for a given egress interface, BGP next-hop or SRv6 traffic engineered path or active
segment how much delay at which node we accumulate. With YANG push this would be rather difficult
and expensive to achieve. This was also confirmed by large network operators in their first
on-path delay measurement deployments.
Best wishes
Thomas
*From:*ippm <ippm-boun...@ietf.org
<mailto:ippm-boun...@ietf.org>> *On Behalf Of
*li_zhenqi...@hotmail.com <mailto:li_zhenqi...@hotmail.com>
*Sent:* Tuesday, January 3, 2023 4:51 PM
*To:* Tianran Zhou
<zhoutianran=40huawei....@dmarc.ietf.org
<mailto:zhoutianran=40huawei....@dmarc.ietf.org>>; ippm
<i...@ietf.org <mailto:i...@ietf.org>>
*Cc:* opsawg@ietf.org <mailto:opsawg@ietf.org>
*Subject:* [ippm] 回复: FW: WG Adoption Call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01
Hi Tianran and all,
Why not use one protocol, such as grpc, to export all
the iOAM metrics? It is ok to export one way delay in
IPFIX. If other metrics, such as queue depth, buffer
occupancy, etc, have to be exported in grpc, it is not
necessory to export one way delay in IPFIX.
------------------------------------------------------------------------
Best Regards,
Zhenqiang Li
li_zhenqi...@hotmail.com <mailto:li_zhenqi...@hotmail.com>
*发件人:*Tianran Zhou
<mailto:zhoutianran=40huawei....@dmarc.ietf.org>
*发送时间:* 2022-12-22 10:40
*收件人:*'IETF IPPM WG' <mailto:i...@ietf.org>
*抄送:*opsawg@ietf.org <mailto:opsawg@ietf.org>
*主题:* [ippm] FW: WG Adoption Call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01
Hi IPPM,
The OPSAWG just started a WG adoption call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01.
https://datatracker.ietf.org/doc/draft-tgraf-opsawg-ipfix-on-path-telemetry/
<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fdoc%2Fdraft-tgraf-opsawg-ipfix-on-path-telemetry%2F&data=05%7C01%7CThomas.Graf%40swisscom.com%7C502beb5de8ee458ffbcb08daf214f30f%7C364e5b87c1c7420d9beec35d19b557a1%7C0%7C0%7C638088469032680658%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=e8tWpkK4w3QDdinb4j5Bg2QaXomgUAfzWAkLbXteqzg%3D&reserved=0>
There are performance metrics registrations within
this draft.
We would really appreciate your comments from IPPM.
Thanks,
Tianran
*发件人**:*OPSAWG [mailto:opsawg-boun...@ietf.org
<mailto:opsawg-boun...@ietf.org>] *代表 *Tianran Zhou
*发送时间:* 2022年12月22日 10:26
*收件人:* opsawg@ietf.org <mailto:opsawg@ietf.org>
*抄送:*
draft-tgraf-opsawg-ipfix-on-path-teleme...@ietf.org
<mailto:draft-tgraf-opsawg-ipfix-on-path-teleme...@ietf.org>
*主题:* [OPSAWG] WG Adoption Call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01
Hi WG,
This mail starts a WG Adoption Call for
draft-tgraf-opsawg-ipfix-on-path-telemetry-01.
https://datatracker.ietf.org/doc/draft-tgraf-opsawg-ipfix-on-path-telemetry/
<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fdoc%2Fdraft-tgraf-opsawg-ipfix-on-path-telemetry%2F&data=05%7C01%7CThomas.Graf%40swisscom.com%7C502beb5de8ee458ffbcb08daf214f30f%7C364e5b87c1c7420d9beec35d19b557a1%7C0%7C0%7C638088469032680658%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=e8tWpkK4w3QDdinb4j5Bg2QaXomgUAfzWAkLbXteqzg%3D&reserved=0>
Please reply your supports or objections.
We would really appreciate your comments.
Since there are holidays, this call will last for 3
weeks, and end on Thursday, Jan 12, 2023.
Cheers,
Tianran (as co-chairs)
_______________________________________________
ippm mailing list
i...@ietf.org <mailto:i...@ietf.org>
https://www.ietf.org/mailman/listinfo/ippm
<https://www.ietf.org/mailman/listinfo/ippm>
_______________________________________________
OPSAWG mailing list
OPSAWG@ietf.org
https://www.ietf.org/mailman/listinfo/opsawg