Hi Randy,

You are right, the primary use case is data-center network, with its specific 
devices: e.g. multiple shared-memory-buffer chips connected in Clos topology. 
The typical MTU I've seen so far there is still 1500, even though enabling 
jumbo framing is generally not a problem. However, the use of GRO and interrupt 
coalescing on host side made the need for jumbo MTU less pressuring.

The idea behind the "template" field was to inform the network device of the 
TLVs that are being "requested" by the originator of the message, along with 
the full spec of the field sizes. The device would then fill in the "slots" it 
can fill in, e.g. the device identifier, ingress/egress ports. This allows the 
originator to request for "lots of stuff", but get back the info supported by 
each individual device. I did receive feedback from some of the folks working 
on silicon side that parsing the template might be "somewhat" expensive, though 
not impossible. I'm continuing that conversation to get closer to "real" 
possibilities of the upcoming silicon. 

I mainly started the document to bring the in-band telemetry to the attention 
of the general network community. To us (facebook neteng) the ability to embed 
device state in data-plane packets is an important enabler for real-time fault 
isolation in the network (our current methodologies are based on some 
variations around fast/wide traceroute).

Regards,

Petr
________________________________________
From: Randy Bush [[email protected]]
Sent: Friday, March 18, 2016 8:30 PM
To: Petr Lapukhov
Cc: opsawg
Subject: Re: [OPSAWG] Call for Agenda Items for Buenos Aires

> https://urldefense.proofpoint.com/v2/url?u=https-3A__datatracker.ietf.org_doc_draft-2Dlapukhov-2Ddataplane-2Dprobe_&d=CwIBAg&c=5VD0RTtNlTh3ycd41b3MUw&r=LU_vJaM_EQu1Ssm35j2xlA&m=uBG5ONHmU7YzsFZ9R25b6TL-l0JZUYFvwDLEAWPqEnA&s=IAtJOy3uPSUJUcEYqZI85xYPTWOcuSQEvcxRddmryAs&e=

given the heavy per-hop payload, do you have estimates on how many hops
with an mtu of 512?  or is this really meant for use within an
operator/datacenter, and maybe one with junbo frames?

i do not understand the relationship of the Telemetry Record Template to
the Telemetry Record.  are these two separate list structures with
reduncancy?  or is it one list and the Telemetry Record Template (which
has TL but no V) is just there to confuse me?

randy

_______________________________________________
OPSAWG mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/opsawg

Reply via email to