Re: [Make-wifi-fast] Bufferbloat on Norwegian train wifi

Re: [Make-wifi-fast] Bufferbloat on Norwegian train wifi

[Toke Høiland-Jørgensen]

> I've always noticed that the wireless network on Norwegian Vy trains
> around Oslo can be close to useless with page load time in excess of
> 10 seconds and frequent timeouts. That's why I decided to bring along
> the laptop to run an RRUL BE test, and the results are, as expected,
> pretty terrible. Mind you, I ran this test while the train was in a
> tunnel, but I don't think that should matter all that much since
> there's good 4G service in the tunnel.
>
> I've attached the flent data file. Average ping is around 800 ms at
> the highest, while ICMP spikes above 1000 ms at the highest. I get
> well below 1 Mbps of download bandwidth, and the upload bandwidth is
> considerably higher but quite variable at 3 to 15 Mbps.

Yeah, that seems pretty bad. Although my personal record for bloat on a
train was around ~30 seconds ;)

Guess you could try complaining to the operator, but it may be difficult
to get hold of anyone with enough of a clue to actually do something
about this. And since the service is likely to be oursourced, you'll
need to get the train operator to apply pressure to the company
providing the service to get them to care. Tried to get Icomera (the
company providing the internet in Swedish and German trains, among
others) to care about bufferbloat some years back; but was totally
unsuccessful :/

-Toke

Re: [Make-wifi-fast] Bufferbloat on Norwegian train wifi

[Simon Barber]

My record was on a Gogo inflight inernet connection. Ping time of 12 minutes!!!! Perhaps I should have complained that I only got 48 minutes out of my 1 hour session?

Simon


> On Jun 22, 2020, at 5:51 AM, Toke Høiland-Jørgensen <toke@redhat.com> wrote:
> 
>> I've always noticed that the wireless network on Norwegian Vy trains
>> around Oslo can be close to useless with page load time in excess of
>> 10 seconds and frequent timeouts. That's why I decided to bring along
>> the laptop to run an RRUL BE test, and the results are, as expected,
>> pretty terrible. Mind you, I ran this test while the train was in a
>> tunnel, but I don't think that should matter all that much since
>> there's good 4G service in the tunnel.
>> 
>> I've attached the flent data file. Average ping is around 800 ms at
>> the highest, while ICMP spikes above 1000 ms at the highest. I get
>> well below 1 Mbps of download bandwidth, and the upload bandwidth is
>> considerably higher but quite variable at 3 to 15 Mbps.
> 
> Yeah, that seems pretty bad. Although my personal record for bloat on a
> train was around ~30 seconds ;)
> 
> Guess you could try complaining to the operator, but it may be difficult
> to get hold of anyone with enough of a clue to actually do something
> about this. And since the service is likely to be oursourced, you'll
> need to get the train operator to apply pressure to the company
> providing the service to get them to care. Tried to get Icomera (the
> company providing the internet in Swedish and German trains, among
> others) to care about bufferbloat some years back; but was totally
> unsuccessful :/
> 
> -Toke
> 
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast

Re: [Make-wifi-fast] Bufferbloat on Norwegian train wifi

[Michael Yartys]

Wow, those are some truly crappy networks!

I'll try to contact Vy to get them to do something about it. I wonder how much of this might be due to the equipment on their trains though. I decided to measure the bufferbloat on the LTE network in my neighbourhood, and on average I got 300 ms above baseline while running an 8-stream TCP download. Then again, I ran this test with my phone acting as a hotspot, and the results might be affected by that. Does anybody know if this methodology produces reliable results? I presume that even the very short peak of 86 Mbps shouldn't really cause much WiFi-related bufferbloat.

Michael

Re: [Make-wifi-fast] Bufferbloat on Norwegian train wifi

[Toke Høiland-Jørgensen]

Michael Yartys <michael.yartys@protonmail.com> writes:

> Wow, those are some truly crappy networks!

Yeah, 'bloat-bragging' seems to have become quite the sport on these
mailing lists ;)

> I'll try to contact Vy to get them to do something about it. I wonder
> how much of this might be due to the equipment on their trains though.
> I decided to measure the bufferbloat on the LTE network in my
> neighbourhood, and on average I got 300 ms above baseline while
> running an 8-stream TCP download. Then again, I ran this test with my
> phone acting as a hotspot, and the results might be affected by that.
> Does anybody know if this methodology produces reliable results? I
> presume that even the very short peak of 86 Mbps shouldn't really
> cause much WiFi-related bufferbloat.

Hard to say from first principles. 300ms doesn't sound unrealistic for
bloat on an LTE network and if you have a good WiFi connection on your
phone the WiFI link *shouldn't* be a bottleneck. But it's hard to know
for sure; just too many variables.

I guess you could try running a speedtest on your phone while you have a
ping running from your tethered laptop. Not sure how effective the
app-based speedtests are at inducing bloat, though; they're certanly not
measuring it...

-Toke

Re: [Make-wifi-fast] Bufferbloat on Norwegian train wifi

[Michael Yartys]

‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Tuesday, 23 June 2020 12:11, Toke Høiland-Jørgensen <toke@redhat.com> wrote:

> Michael Yartys michael.yartys@protonmail.com writes:
>
> > Wow, those are some truly crappy networks!
>
> Yeah, 'bloat-bragging' seems to have become quite the sport on these
> mailing lists ;)
>
> > I'll try to contact Vy to get them to do something about it. I wonder
> > how much of this might be due to the equipment on their trains though.
> > I decided to measure the bufferbloat on the LTE network in my
> > neighbourhood, and on average I got 300 ms above baseline while
> > running an 8-stream TCP download. Then again, I ran this test with my
> > phone acting as a hotspot, and the results might be affected by that.
> > Does anybody know if this methodology produces reliable results? I
> > presume that even the very short peak of 86 Mbps shouldn't really
> > cause much WiFi-related bufferbloat.
>
> Hard to say from first principles. 300ms doesn't sound unrealistic for
> bloat on an LTE network and if you have a good WiFi connection on your
> phone the WiFI link shouldn't be a bottleneck. But it's hard to know
> for sure; just too many variables.
>
> I guess you could try running a speedtest on your phone while you have a
> ping running from your tethered laptop. Not sure how effective the
> app-based speedtests are at inducing bloat, though; they're certanly not
> measuring it...

Thanks for the suggestion! It seems like I get results that are in line with what I got when running the flent test. I'll contact my mobile service provider as well to check if they're aware of the bufferbloat issue.

Also, if anyone's interested, I found an interesting paper about bufferbloat experienced during uploading on 4G networks: https://www-users.cs.umn.edu/~fengqian/paper/bufferbloat_imc16.pdf

>
> -Toke