[Make-wifi-fast] Uplink vs downlink latency

[Make-wifi-fast] Uplink vs downlink latency

[Tim Higgins]

[-- Attachment #1: Type: text/html, Size: 2546 bytes --]

[-- Attachment #2: netgear_rax15_latency_up_vs_dn_ofdma_on.jpg --]
[-- Type: image/jpeg, Size: 79767 bytes --]

[-- Attachment #3: netgear_r7800_latency_up_vs_dn_ofdma_off.jpg --]
[-- Type: image/jpeg, Size: 77582 bytes --]

[-- Attachment #4: netgear_rax15_latency_up_vs_dn.jpg --]
[-- Type: image/jpeg, Size: 80082 bytes --]

Re: [Make-wifi-fast] Uplink vs downlink latency

[Isaac Konikoff]

[-- Attachment #1: Type: text/plain, Size: 2802 bytes --]

Hi Tim,

I know this will probably be in your write-up, but I am curious if you are
sniffing the traffic to verify that OFDMA is actually happening by pinning
the aid to the monitor interface and using a wireshark display filter
radiotap.he.data_1.ppdu_format==0x2 to see the HE_MU frames. From my
testing it is not something that is automatically in use all the time...it
is used as needed per each AP's decision algorithm. I seem to be able to
cause OFDMA to happen sometimes when using small payloads with bursty
traffic and periods of quiet, but it is not something that I can turn on
and just say OFDMA is happening.

Also, in my flent tcp_download vs tcp_upload tests I see similar latency
values that are both relatively low. I will have to run the rtt_fair_var to
see how it compares.

Nice use of flent! Looking forward to your write-up.
Isaac


On Wed, Apr 29, 2020 at 2:40 PM Tim Higgins <tim@smallnetbuilder.com> wrote:

> Hi all,
>
> I finally have my testbed working the way I want and am starting to run
> tests to see if OFDMA does anything useful.
>
> This will all be covered in detail in an upcoming SmallNetBuilder article.
> But I wanted to sanity check something with this esteemed group.
>
> The tests are basically the flent rtt_fair_var up and down tests ported to
> the octoScope platform I use for WiFi testing.
> The initial work was done on flent, with a lot of hand-holding from Toke.
> (Thank you, Toke!)
>
> Using 4 Intel AX200 STAs on Win10. iperf3 is running traffic using TCP/IP
> with unthrottled bandwidth. I've taken Bjørn's idea and have each STA using
> a different DSCP priority level, but with TCP/IP traffic, not UDP. I'm
> sticking to using CS0-7 equivalents and confirmed that the iperf3 --dscp
> values properly translate to the intended WiFi priority levels.  Each STA
> has a different priority, either CS0,3,5 or 6 (best effort, excellent
> effort, video and voice).
>
> Ping is used to measure latency and always runs from AP to STA. Only
> TCP/IP traffic direction is reversed between the down and uplink tests.
>
> One thing that jumps out immediately is that uplink latencies are *much*
> lower than downlink, with either OFDMA on or off. Attached are three
> examples. The CDFs are average latency of the 4 STAs.
>
> The NETGEAR R7800 is a 4x4 AC Qualcomm-based. I'm using this as a baseline
> product.
>
> The NETGEAR RAX15 is 2x2 AX Broadcom-based. You can see what I mean when I
> say OFDMA doesn't help.
>
> Does this much difference between up and downlink latency pass the sniff
> test?
>
> ===
> Tim
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast

[-- Attachment #2: Type: text/html, Size: 4056 bytes --]

Re: [Make-wifi-fast] Uplink vs downlink latency

[Dave Taht]

throughput and latency are interrelated, whats the throughput?

On Wed, Apr 29, 2020 at 2:40 PM Tim Higgins <tim@smallnetbuilder.com> wrote:
>
> Hi all,
>
> I finally have my testbed working the way I want and am starting to run tests to see if OFDMA does anything useful.
>
> This will all be covered in detail in an upcoming SmallNetBuilder article. But I wanted to sanity check something with this esteemed group.
>
> The tests are basically the flent rtt_fair_var up and down tests ported to the octoScope platform I use for WiFi testing.
> The initial work was done on flent, with a lot of hand-holding from Toke. (Thank you, Toke!)
>
> Using 4 Intel AX200 STAs on Win10. iperf3 is running traffic using TCP/IP with unthrottled bandwidth. I've taken Bjørn's idea and have each STA using a different DSCP priority level, but with TCP/IP traffic, not UDP. I'm sticking to using CS0-7 equivalents and confirmed that the iperf3 --dscp values properly translate to the intended WiFi priority levels.  Each STA has a different priority, either CS0,3,5 or 6 (best effort, excellent effort, video and voice).
>
> Ping is used to measure latency and always runs from AP to STA. Only TCP/IP traffic direction is reversed between the down and uplink tests.
>
> One thing that jumps out immediately is that uplink latencies are *much* lower than downlink, with either OFDMA on or off. Attached are three examples. The CDFs are average latency of the 4 STAs.
>
> The NETGEAR R7800 is a 4x4 AC Qualcomm-based. I'm using this as a baseline product.
>
> The NETGEAR RAX15 is 2x2 AX Broadcom-based. You can see what I mean when I say OFDMA doesn't help.
>
> Does this much difference between up and downlink latency pass the sniff test?
>
> ===
> Tim
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast



-- 
Make Music, Not War

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-435-0729

Re: [Make-wifi-fast] Uplink vs downlink latency

[Bob McMahon]

[-- Attachment #1: Type: text/plain, Size: 2890 bytes --]

I'm thinking ping may not be ideal for benchmarking OFDMA effects on
latency.  Also, the end/end latency preferred seems to me the socket
write() to final socket read() per that write(). Also, for TCP, there are
the connect times. I realize network stack guys focus on stack related
measurements, e.g. RTT, but the latencies users experience include the
application level and system level os interactions.

Just some food or thought.

Bob


On Wed, Apr 29, 2020 at 5:07 PM Dave Taht <dave.taht@gmail.com> wrote:

> throughput and latency are interrelated, whats the throughput?
>
> On Wed, Apr 29, 2020 at 2:40 PM Tim Higgins <tim@smallnetbuilder.com>
> wrote:
> >
> > Hi all,
> >
> > I finally have my testbed working the way I want and am starting to run
> tests to see if OFDMA does anything useful.
> >
> > This will all be covered in detail in an upcoming SmallNetBuilder
> article. But I wanted to sanity check something with this esteemed group.
> >
> > The tests are basically the flent rtt_fair_var up and down tests ported
> to the octoScope platform I use for WiFi testing.
> > The initial work was done on flent, with a lot of hand-holding from
> Toke. (Thank you, Toke!)
> >
> > Using 4 Intel AX200 STAs on Win10. iperf3 is running traffic using
> TCP/IP with unthrottled bandwidth. I've taken Bjørn's idea and have each
> STA using a different DSCP priority level, but with TCP/IP traffic, not
> UDP. I'm sticking to using CS0-7 equivalents and confirmed that the iperf3
> --dscp values properly translate to the intended WiFi priority levels.
> Each STA has a different priority, either CS0,3,5 or 6 (best effort,
> excellent effort, video and voice).
> >
> > Ping is used to measure latency and always runs from AP to STA. Only
> TCP/IP traffic direction is reversed between the down and uplink tests.
> >
> > One thing that jumps out immediately is that uplink latencies are *much*
> lower than downlink, with either OFDMA on or off. Attached are three
> examples. The CDFs are average latency of the 4 STAs.
> >
> > The NETGEAR R7800 is a 4x4 AC Qualcomm-based. I'm using this as a
> baseline product.
> >
> > The NETGEAR RAX15 is 2x2 AX Broadcom-based. You can see what I mean when
> I say OFDMA doesn't help.
> >
> > Does this much difference between up and downlink latency pass the sniff
> test?
> >
> > ===
> > Tim
> > _______________________________________________
> > Make-wifi-fast mailing list
> > Make-wifi-fast@lists.bufferbloat.net
> > https://lists.bufferbloat.net/listinfo/make-wifi-fast
>
>
>
> --
> Make Music, Not War
>
> Dave Täht
> CTO, TekLibre, LLC
> http://www.teklibre.com
> Tel: 1-831-435-0729
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast

[-- Attachment #2: Type: text/html, Size: 3814 bytes --]

Re: [Make-wifi-fast] Uplink vs downlink latency

[Tim Higgins]

[-- Attachment #1: Type: text/html, Size: 7057 bytes --]

Re: [Make-wifi-fast] Uplink vs downlink latency

[Tim Higgins]

[-- Attachment #1: Type: text/html, Size: 3174 bytes --]

[-- Attachment #2: netgear_rax15_throughput_up_vs_dn.jpg --]
[-- Type: image/jpeg, Size: 95432 bytes --]

[-- Attachment #3: netgear_r7800_throughput_up_vs_dn.jpg --]
[-- Type: image/jpeg, Size: 92616 bytes --]

Re: [Make-wifi-fast] Uplink vs downlink latency

[Tim Higgins]

[-- Attachment #1: Type: text/html, Size: 6394 bytes --]

Re: [Make-wifi-fast] Uplink vs downlink latency

[Dave Taht]

It would be kind of cool if you added a ubnt uap-pro mesh router to
your mix. And (eventually) - reflashed it with our latest openwrt
stuff. that's a 3x3 802.11ac product that I know for sure has most of
our stuff in the default firmware nowadays.

On Thu, Apr 30, 2020 at 7:18 AM Tim Higgins <tim@smallnetbuilder.com> wrote:
>
> I take your point, Bob, and agree single-ended latency measurements like you produce with iperf would be more technically correct.
>
> I write and review for a consumer audience, where ping is the standard for latency aka lag measurement. So that's why I'm using ping.
> That, and the fact that iperf isn't integrated into octoScope's toolset yet. But they're working on it and all their STA instruments are properly time-synced, so the measurements will be accurate.
>
> Thank your for all your work in iperf, BTW. The features you've added are a welcome improvement.
> ===========
> Tim
> On 4/29/2020 8:32 PM, Bob McMahon wrote:
>
> I'm thinking ping may not be ideal for benchmarking OFDMA effects on latency.  Also, the end/end latency preferred seems to me the socket write() to final socket read() per that write(). Also, for TCP, there are the connect times. I realize network stack guys focus on stack related measurements, e.g. RTT, but the latencies users experience include the application level and system level os interactions.
>
> Just some food or thought.
>
> Bob
>
>
> On Wed, Apr 29, 2020 at 5:07 PM Dave Taht <dave.taht@gmail.com> wrote:
>>
>> throughput and latency are interrelated, whats the throughput?
>>
>> On Wed, Apr 29, 2020 at 2:40 PM Tim Higgins <tim@smallnetbuilder.com> wrote:
>> >
>> > Hi all,
>> >
>> > I finally have my testbed working the way I want and am starting to run tests to see if OFDMA does anything useful.
>> >
>> > This will all be covered in detail in an upcoming SmallNetBuilder article. But I wanted to sanity check something with this esteemed group.
>> >
>> > The tests are basically the flent rtt_fair_var up and down tests ported to the octoScope platform I use for WiFi testing.
>> > The initial work was done on flent, with a lot of hand-holding from Toke. (Thank you, Toke!)
>> >
>> > Using 4 Intel AX200 STAs on Win10. iperf3 is running traffic using TCP/IP with unthrottled bandwidth. I've taken Bjørn's idea and have each STA using a different DSCP priority level, but with TCP/IP traffic, not UDP. I'm sticking to using CS0-7 equivalents and confirmed that the iperf3 --dscp values properly translate to the intended WiFi priority levels.  Each STA has a different priority, either CS0,3,5 or 6 (best effort, excellent effort, video and voice).
>> >
>> > Ping is used to measure latency and always runs from AP to STA. Only TCP/IP traffic direction is reversed between the down and uplink tests.
>> >
>> > One thing that jumps out immediately is that uplink latencies are *much* lower than downlink, with either OFDMA on or off. Attached are three examples. The CDFs are average latency of the 4 STAs.
>> >
>> > The NETGEAR R7800 is a 4x4 AC Qualcomm-based. I'm using this as a baseline product.
>> >
>> > The NETGEAR RAX15 is 2x2 AX Broadcom-based. You can see what I mean when I say OFDMA doesn't help.
>> >
>> > Does this much difference between up and downlink latency pass the sniff test?
>> >
>> > ===
>> > Tim
>> > _______________________________________________
>> > Make-wifi-fast mailing list
>> > Make-wifi-fast@lists.bufferbloat.net
>> > https://lists.bufferbloat.net/listinfo/make-wifi-fast
>>
>>
>>
>> --
>> Make Music, Not War
>>
>> Dave Täht
>> CTO, TekLibre, LLC
>> http://www.teklibre.com
>> Tel: 1-831-435-0729
>> _______________________________________________
>> Make-wifi-fast mailing list
>> Make-wifi-fast@lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>
>
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast



-- 
Make Music, Not War

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-435-0729