Re: [tsvwg] I-D Action: draft-ietf-tsvwg-nqb-08.txt

[Sebastian Moeller <[e-mail]>]

Hi Ruediger,

10% of which rate exactly? WiFi is variable rate and the upstream side really has no idea about the instantaneous capacity....

The NQB rate limiting really needs to be performed by nodes in control of variable rate links, which in turn means that the NQB wifi side-effect should be opt-in instead of the current opt-out approach. A point I have made pretty consistently. As long as the default NQB dscp maps to AC_VI/O NQB is unsafe for deployment, ni ifs and no buts. As long as active intervention is required to keep NQBs side effects under control there is an obvious safety failure mode right there.


Regards
        Sebastian




On 26 October 2021 09:10:24 CEST, [e-mail] wrote:
>Hi Greg,
>
>[RG] - Bob suggests 10% of the access bandwidth as a limit, so to say. A rule of thumb says the resource demand for flows below 10% interface bandwidth can be characterized by their average bandwidth, whereas above 10%, the peak bandwidth better characterizes the flows required resources. So above this boundary burstiness of such a flow impacts other flows (and therefore peak bandwidth should be used to define the required resources).
>
>To me the suggestion of Bob makes sense. In my recollection, I've read this rule of thumb in a publication of Gunnar Karlsson, a researcher who's been working on measurement based admission control.
>
>Regards,
>
>Ruediger
>
>
> ######### you wrote ###########  
>
>In looking back at the mailing list just now, I realized that I did not address Bob Briscoe's comment: 
>https://mailarchive.ietf.org/arch/msg/tsvwg/ntGanYqcGP8GqlgzCPSQ1W6rg1U/
>
>Apologies for that. 
>
>His suggestion was to replace:
>
>"If the application's traffic exceeds more than a few packets per RTT, or exceeds approximately 1 Mbps on an instantaneous (inter-packet) basis, the application SHOULD NOT mark its traffic with the NQB DSCP."
>
>With something along the lines of:
>
>"If the application's traffic exceeds a single digit percentage of the global average access link capacity at the time, the application SHOULD NOT mark its traffic with the NQB DSCP.  At the time of writing this document, the global average access link capacity is 63 Mbps down and 13 Mbps up for mobile networks, 113 Mbps down and 62 Mbps up for fixed networks [REF https://www.speedtest.net/global-index or some other reference]."
>
>His rationale can be found in the mailing list posting linked above.
>
>I'm interested in opinions on this, as well as suggestions on the wording.
>
>-Greg
>
>

-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.