[Bloat] Open Loaded latency data from ookla

[Bloat] Open Loaded latency data from ookla

[Dave Taht]

Now available:

https://www.ookla.com/articles/announcing-loaded-latency-ookla-open-data

--
A pithy note on VOQs vs SQM: https://blog.cerowrt.org/post/juniper/
Dave Täht CEO, TekLibre, LLC

Re: [Bloat] [Rpm] Open Loaded latency data from ookla

[Sebastian Moeller]

Hi All,

We are getting there...

'The average latency of all tests performed in the tile, represented in milliseconds'

This is infinitely better than no under-load numbers. However, the latency distribution is neither 'normal'* nor fully symmetric (it can not, there is a hard minimum, the minimal path RTT and a hard maximum (icmp timeout) even though the distribution will likely be closer to the minimum than the maximum)**.

Next improvement possibility: they report averages and number of tests, but no measure of variance... 

To put a positive spin on it, reporting luli numbers at all is great, as is the potential to improve the reporting quality with small additions!

Regards
        Sebastian

*) Which is not a showstopper per se as long as the distribution is unimodal and not too skewed then the average is still intuitively interpretable, but is it?
**) My gut feeling is a beta distribution could be fitted to the data and that functions control parameters reported to better model the expected distribution shape.

P.S.: This time without links to test what triggers getting blocked...

Re: [Bloat] [Rpm] Open Loaded latency data from ookla

[Taraldsen Erik]

Considering how much the C level management weights the level us ISP's ranks in Ooklas test - weaknesses and all, this is very good news.  At least here in Norway the ranking in Ookla is used in marketing and is considered very important.  If ISPs (such as Telenor who I work for) starts to slip in the rating due to bad latency numbers - we may finally get some traction to improve the situation.  

-Erik 



On 24/02/2023, 08:43, "Bloat on behalf of Sebastian Moeller via Bloat" <bloat-bounces@lists.bufferbloat.net <mailto:bloat-bounces@lists.bufferbloat.net> on behalf of bloat@lists.bufferbloat.net <mailto:bloat@lists.bufferbloat.net>> wrote:


Hi All,


We are getting there...


'The average latency of all tests performed in the tile, represented in milliseconds'


This is infinitely better than no under-load numbers. However, the latency distribution is neither 'normal'* nor fully symmetric (it can not, there is a hard minimum, the minimal path RTT and a hard maximum (icmp timeout) even though the distribution will likely be closer to the minimum than the maximum)**.


Next improvement possibility: they report averages and number of tests, but no measure of variance... 


To put a positive spin on it, reporting luli numbers at all is great, as is the potential to improve the reporting quality with small additions!


Regards
Sebastian


*) Which is not a showstopper per se as long as the distribution is unimodal and not too skewed then the average is still intuitively interpretable, but is it?
**) My gut feeling is a beta distribution could be fitted to the data and that functions control parameters reported to better model the expected distribution shape.


P.S.: This time without links to test what triggers getting blocked...
_______________________________________________
Bloat mailing list
Bloat@lists.bufferbloat.net <mailto:Bloat@lists.bufferbloat.net>
https://lists.bufferbloat.net/listinfo/bloat <https://lists.bufferbloat.net/listinfo/bloat>