[Bloat] [Cerowrt-devel] marketing #102 - giving netperf-wrapper a better name?
David P. Reed
dpreed at reed.com
Fri Mar 20 20:19:45 EDT 2015
Drag is an fluid dynamic term that suggests a meaning close to this... flow rate dependent friction.
But what you really want to suggest is a flow rate dependent *delay* that people are familiar with quantifying.
Fq_codel limits the delay as flow rate increases and is fair.
The max buffer limits delay due to Little's lemma also.
The actual delay limit in practice is what one measures in most cases. Codel is just softer than the hard limit of a small buffer.
So there are two qualitative measures - delay limit in units of milliseconds and softness or stiffness in units of milliseconds per queue depth I'd guess. Softness gives the recovery rate after a burst.
You should divide delay limit by elephant packet size in milliseconds. Based on channel rate.
I'd think the scaled delay limit should
On Mar 20, 2015, "Bill Ver Steeg (versteb)" <versteb at cisco.com> wrote:
>I was kidding about "sucks-less", and forgot the smiley in my initial
>We do need a metric with an end-user-friendly name, though. Most people
>understand "lag", and understand that lower numbers are better. You
>could probably explain "lag-while-loaded" to most users (particularly
>people who care, like gamers) in a manner that got the point across.
>From: Jonathan Morton [mailto:chromatix99 at gmail.com]
>Sent: Friday, March 20, 2015 4:26 PM
>To: Bill Ver Steeg (versteb)
>Cc: Rémi Cardona; bloat; cerowrt-devel at lists.bufferbloat.net
>Subject: Re: [Bloat] marketing #102 - giving netperf-wrapper a better
>> On 20 Mar, 2015, at 22:08, Bill Ver Steeg (versteb)
><versteb at cisco.com> wrote:
>> We should call the metric "sucks-less". As in "Box A sucks less than
>Box B", or "Box C scored a 17 on the sucks less test".
>I suspect real marketing drones would get nervous at a
>My idea - which I’ve floated in the past, more than once - is that the
>metric should be “responsiveness”, measured in Hertz. The baseline
>standard would be 10Hz, corresponding to a dumb 100ms buffer. Get down
>into the single-digit millisecond range, as fq_codel does, and the
>Responsiveness goes up above 100Hz, approaching 1000Hz.
>Crucially, that’s a positive sort of term, as well as trending towards
>bigger numbers with actual improvements in performance, and is thus
>more potentially marketable.
> - Jonathan Morton
>Cerowrt-devel mailing list
>Cerowrt-devel at lists.bufferbloat.net
-- Sent with K-@ Mail - the evolution of emailing.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Bloat