Tony,
Comments In-Line
Thanks
Jim Uttaro
-----Original Message-----
From: Tony Li <[email protected]> On Behalf Of Tony Li
Sent: Sunday, April 2, 2023 12:20 PM
To: UTTARO, JAMES <[email protected]>
Cc: Gyan Mishra <[email protected]>; Robert Raszuk <[email protected]>;
RTGWG <[email protected]>; [email protected]
Subject: Re: TTE
Hi Jim,
Thank you for your comments.
> On Apr 2, 2023, at 5:19 AM, UTTARO, JAMES <[email protected]> wrote:
>
> Tony,
>
> I am unsure of the problem you are solving with this draft.. In the
> introduction "Unforseen and/or dynamic events, can skew..." Certainly
> unforeseen events are not accounted for in the estimate based on historical
> demands, what is meant by "dynamic changes" ? and how does it differ from
> "unforeseen" ?
No difference, just a bit of poetic license.
Fundamentally, the problem that we’re trying to address are loads that change
in unanticipated ways and at a higher frequency than global path optimization
can support.
> The draft goes into addressing predictable patterns of network utilization
> using the example of "follow the sun". This is predictable and should be
> captured and be part of the historical pattern.. This may be captured and
> addressed in a set of changes that 'migrate" traffic to under utilized links.
> Of course, this may impact latency or some other metric that is used in SLAs.
>
> How does the solution deal with placing flows on a "backup path" which then
> may become congested creating a "ping pong " effect within a sub-graph of the
> topology?
As previously discussed, TTE assumes that there is adequate capacity engineered
into backup paths. Such capacity should already be present to account for link
failures.
[Jim U>]The capacity of the backup topology may include one or more equal or
unequal cost paths to accommodate the additional load that is shunted. Are you
suggesting that this is a pre-defined set such that the backup set still meets
some percentage of SLA/SLOs? This is a valid tradeoff; I have had similar
discussions in re BGP Persistence and whether or not it should be applied for
different FOUs.
> There is no discussion in re the longevity of a given set of flows as input
> to the "flow selection", an example you use " a singer my announce...", I
> would think this would result in many small flows congesting the link which
> are most likely short lived.. This is opposed to long lived elephant flows
> used in the middle of the night for machine-2-machin type applications.
That’s a fair point, thank you. Flow selection is an interesting topic, to be
sure.
Tony
_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg
