Oliver Hohlfeld wrote:
>> In ascii art, it might look like this:
>> ======++++++++++++
>>                -                              +
>>                 -                                +
>>                  -                                  +
>>                  -                                    +
> What exactly does this plot represent and on what measurement
> data is it based on? What is your definition of "good" and "bad"?
> Just response times? Where does it's shape comes from?
>
> Before (again) diving into debates on how bufferbloat marketing plots
> should look like, I think we need to make sure that the marketing
> is backed up by empirical data.
>
> --Oliver
It's the inverse of a response time versus load graph for two cases, one
with a very early degradation, the other with a normal one.  It's a
bowdlerized version of some real measurements of a REST-based system
with and without a bug.  I see the list software put in a link to the
gif, so look at that: it's far better than my hack at "art".

Good and bad are indeed defined in terms of response time for a
request-replay pair, with the appropriate units being part of my question.

The normal shape of a response-time curve is a "hockey stick", like
"_/".  Technically it's a hyperbola with it's legs asymptotic to a pair
of straight lines, one line horizontal at the value for load=1, and the
second slanted at an angle that depends on the response time, the
bottleneck time (if different), the think time between requests and the
load.  The right-side-up graph uses normal time units.

I tried the experiment of inverting it and asking Dave and the list for
an opinion on units: in my actual example they were normalized by
dividing by the minimum response time, making it non-dimensional.

--dave

-- 
David Collier-Brown,         | Always do right. This will gratify
System Programmer and Author | some people and astonish the rest
[email protected]           |                      -- Mark Twain
(416) 223-8968

_______________________________________________
Bloat mailing list
[email protected]
https://lists.bufferbloat.net/listinfo/bloat

Reply via email to