[Make-wifi-fast] Flent traffic direction convention

[Make-wifi-fast] Flent traffic direction convention

[Tim Higgins]

[-- Attachment #1: Type: text/html, Size: 1069 bytes --]

Re: [Make-wifi-fast] Flent traffic direction convention

[Toke Høiland-Jørgensen]

(Adding flent-users@flent.org - please also include this in
Flent-specific questions)

> I might be confusing myself, but need to ask the question. What does flent
> define as upload and download?
>
> Netperf docs say:
> The TCP_STREAM test is the default test in netperf. It is quite simple,
> transferring some quantity of data from the system running netperf to the
> system running netserver.
>
> So why, when I look at tcp_1up_noping.conf does the DATA_SETS have 'test':
> 'TCP_STREAM' in the 'TCP_upload' dictionary?

Flent considers itself the client, so 'upload' is TCP_STREAM. As more
use cases have emerged, this has turned out to not always be the case,
of course; WiFi test scenarios such as you are doing being the most
prominent example. There's a --swap-up-down parameter which will switch
all instances of TCP_STREAM to TCP_MAERTS and vice-versa, though.

> Second question: Is there any documentation that can help me figure out
> exactly what traffic is running in each test type?
> I've tried looking at the .conf and .inc files, in flent/tests but can't
> figure it out.

Not really :/ There's https://flent.org/tests.html, but it hasn't really
kept up with new tests being added.

However, you can see it after the fact: The exact command lines for each
data series is stored (along with a bunch of other data) in the
SERIES_META object in the data file. Use the GUI to browse this, or
print it with 'flent -f metadata datafile.flent.gz'.

-Toke

Re: [Make-wifi-fast] Flent traffic direction convention

[Tim Higgins]

> On Mar 27, 2020, at 7:05 AM, Toke Høiland-Jørgensen <toke@redhat.com> wrote:
> 
> (Adding flent-users@flent.org - please also include this in
> Flent-specific questions)
> 
>> I might be confusing myself, but need to ask the question. What does flent
>> define as upload and download?
>> 
>> Netperf docs say:
>> The TCP_STREAM test is the default test in netperf. It is quite simple,
>> transferring some quantity of data from the system running netperf to the
>> system running netserver.
>> 
>> So why, when I look at tcp_1up_noping.conf does the DATA_SETS have 'test':
>> 'TCP_STREAM' in the 'TCP_upload' dictionary?
> 
> Flent considers itself the client, so 'upload' is TCP_STREAM. As more
> use cases have emerged, this has turned out to not always be the case,
> of course; WiFi test scenarios such as you are doing being the most
> prominent example. There's a --swap-up-down parameter which will switch
> all instances of TCP_STREAM to TCP_MAERTS and vice-versa, though.
> 
>> Second question: Is there any documentation that can help me figure out
>> exactly what traffic is running in each test type?
>> I've tried looking at the .conf and .inc files, in flent/tests but can't
>> figure it out.
> 
> Not really :/ There's https://flent.org/tests.html, but it hasn't really
> kept up with new tests being added.
> 
> However, you can see it after the fact: The exact command lines for each
> data series is stored (along with a bunch of other data) in the
> SERIES_META object in the data file. Use the GUI to browse this, or
> print it with 'flent -f metadata datafile.flent.gz'.
> 
> -Toke
> 
Thanks again, Toke. Looks like you anticipated my questions in the design! :)

Re: [Make-wifi-fast] [Flent-users] Flent traffic direction convention

[Toke Høiland-Jørgensen]

Tim Higgins <tim@timhiggins.com> writes:

>> On Mar 27, 2020, at 7:05 AM, Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>> 
>> (Adding flent-users@flent.org - please also include this in
>> Flent-specific questions)
>> 
>>> I might be confusing myself, but need to ask the question. What does flent
>>> define as upload and download?
>>> 
>>> Netperf docs say:
>>> The TCP_STREAM test is the default test in netperf. It is quite simple,
>>> transferring some quantity of data from the system running netperf to the
>>> system running netserver.
>>> 
>>> So why, when I look at tcp_1up_noping.conf does the DATA_SETS have 'test':
>>> 'TCP_STREAM' in the 'TCP_upload' dictionary?
>> 
>> Flent considers itself the client, so 'upload' is TCP_STREAM. As more
>> use cases have emerged, this has turned out to not always be the case,
>> of course; WiFi test scenarios such as you are doing being the most
>> prominent example. There's a --swap-up-down parameter which will switch
>> all instances of TCP_STREAM to TCP_MAERTS and vice-versa, though.
>> 
>>> Second question: Is there any documentation that can help me figure out
>>> exactly what traffic is running in each test type?
>>> I've tried looking at the .conf and .inc files, in flent/tests but can't
>>> figure it out.
>> 
>> Not really :/ There's https://flent.org/tests.html, but it hasn't really
>> kept up with new tests being added.
>> 
>> However, you can see it after the fact: The exact command lines for each
>> data series is stored (along with a bunch of other data) in the
>> SERIES_META object in the data file. Use the GUI to browse this, or
>> print it with 'flent -f metadata datafile.flent.gz'.
>> 
>> -Toke
>> 
> Thanks again, Toke. Looks like you anticipated my questions in the design! :)

You're welcome! And more like you are running into the same problems I
did myself originally, while running experiments quite similar to yours :)

BTW, my full data set for my own WiFi experiments are here:

https://www.cs.kau.se/tohojo/airtime-fairness/
https://www.cs.kau.se/tohojo/polifi/

The data files should also contain the Flent batch files and setup
scripts; feel free to re-use anything you need :)

-Toke

Re: [Make-wifi-fast] Flent traffic direction convention

[Erkki Lintunen]

On 3/26/20 11:22 PM, Tim Higgins wrote:
> Second question: Is there any documentation that can help me figure out 
> exactly what traffic is running in each test type?

Not exactly an answer to the question.

I found following pages very helpful when three years ago I challenged 
myself with Flent to measure WiFi network performance and max it out 
with network devices I had at hand at the time.

(One ap blew up it's two capasitors. Visually and operationally the ap 
was working as normal but measurements showed drastic drop in 
performance. Light odor and inspection of el-capasitor burned tops did 
confirm the hard to believe experience.)

<https://www.bufferbloat.net/projects/bloat/wiki/RRUL_Chart_Explanation/>

The above page provides link to this page.

<https://burntchrome.blogspot.com/2016/12/cake-latest-in-sqm-qos-schedulers.html>

Experience showed up Flent's feature to compare data sets very educating 
and powerfull, not only with the box-plot chart as the pages show. Took 
time to get mentaly bend to read and interpret the Flent charts instead 
of only numbers I was used to with iperf and other tools.

In my quest I did a change, measured it, compared to the previous set 
and reviewed, if the change showed any measurable effect. Finally I 
compared the sets of the very first and the last. I was totally driven 
by measurements on my way myth busting many my beliefs for a performant 
network.

The quest changed my definition for "getting max out of a network". Now 
the max is all quantities in effect at any time: fairness, lowest 
possible latency and highest possible bitrate. Looking at many published 
benchmarks today, I still think, max seams to mean just a maximum 
bitrate and other qualities are _supposed_ to derive from it.

Hope this is for any help.

- Erkki

Re: [Make-wifi-fast] Flent traffic direction convention

[Tim Higgins]

[-- Attachment #1: Type: text/html, Size: 2944 bytes --]