UFB International Bandwidth Speed Test

Forums › New Zealand Broadband › UFB International Bandwidth Speed Test

1 | 2 | 3

487 posts

Ultimate Geek

Trusted

#1216751 17-Jan-2015 23:51

plambrechtsen: And here is the problem with TCP and long ping times. Without tweaking of your TCP stack or multiple threads you will never get fast speeds to the other side of the world.

If you really wanted a good speed test you should use UDP and measure how much data is dropped or what is the maximum speed it can arrive at.

I doubt you will see anything faster than 15mb ever to the UK or NL. It's just how TCP works on long round trips.

Yep however that is exactly my intention.
I am streaming over HTTP (TCP packets) so it makes no sense to measure UDP.
The server has an upload cap of 9mb anyway, so I am not looking for the highest throughput. This is hosted on a domestic cable connection.

My first experiment was to find out what the "viewable" stream quality would be from NL to NZ.
I can now conclude that pretty much anywhere in NZ would be able to obtain SD quality streams compressed at 1.3mbps, and so far everybody responding to this thread could pull at least 3mbit in HD.

My second experiment is to find out if there are any notable differences on the various ISPs, the results are too few to be conclusive at the moment but I'm quite surprised by the Orcon reading being faster than what I am getting here in the UK!

Thanks so far for all contributions - everybody posting a reading will get a +1 :)

Haere taka mua, taka muri; kaua e wha.

Talkiet

4793 posts

Uber Geek

Trusted

#1216753 17-Jan-2015 23:56

plambrechtsen: And here is the problem with TCP and long ping times. Without tweaking of your TCP stack or multiple threads you will never get fast speeds to the other side of the world.

If you really wanted a good speed test you should use UDP and measure how much data is dropped or what is the maximum speed it can arrive at.

I doubt you will see anything faster than 15mb ever to the UK or NL. It's just how TCP works on long round trips.

Except speedtest.net (and possibly other speedtests) fudge their tests and estimate line speeds. It's only partially related to the actual traffic they exchange as part of the test. And they use multiple threads, and they use a non centrally tended subset of results and then they add a fudge factor for 'overheads'

It's a miracle that speedtest.net results ever match expectations.

Cheers - N

Please note all comments are from my own brain and don't necessarily represent the position or opinions of my employer, previous employers, colleagues, friends or pets.

kiwikiwi

455 posts

Ultimate Geek

#1216763 18-Jan-2015 01:06

ScuL:
plambrechtsen: And here is the problem with TCP and long ping times. Without tweaking of your TCP stack or multiple threads you will never get fast speeds to the other side of the world.

If you really wanted a good speed test you should use UDP and measure how much data is dropped or what is the maximum speed it can arrive at.

I doubt you will see anything faster than 15mb ever to the UK or NL. It's just how TCP works on long round trips.

Yep however that is exactly my intention.
I am streaming over HTTP (TCP packets) so it makes no sense to measure UDP.
The server has an upload cap of 9mb anyway, so I am not looking for the highest throughput. This is hosted on a domestic cable connection.

My first experiment was to find out what the "viewable" stream quality would be from NL to NZ.
I can now conclude that pretty much anywhere in NZ would be able to obtain SD quality streams compressed at 1.3mbps, and so far everybody responding to this thread could pull at least 3mbit in HD.

My second experiment is to find out if there are any notable differences on the various ISPs, the results are too few to be conclusive at the moment but I'm quite surprised by the Orcon reading being faster than what I am getting here in the UK!

Thanks so far for all contributions - everybody posting a reading will get a +1 :)

Since I'm still awake at stupid hour o clock and I have a Orcon UFB connection, is there anything else you would like investigated?
Other people with their alternative ISPs should ask the same too.

You can also follow me on twitter here @kiwifortw I do twitch streams every now and then at twitch.tv/kiwiforthewin :)

HTTP 404 Sarcasm not found.

kiwikiwi

455 posts

Ultimate Geek

#1216764 18-Jan-2015 01:07

Talkiet:
plambrechtsen: And here is the problem with TCP and long ping times. Without tweaking of your TCP stack or multiple threads you will never get fast speeds to the other side of the world.

If you really wanted a good speed test you should use UDP and measure how much data is dropped or what is the maximum speed it can arrive at.

I doubt you will see anything faster than 15mb ever to the UK or NL. It's just how TCP works on long round trips.

Except speedtest.net (and possibly other speedtests) fudge their tests and estimate line speeds. It's only partially related to the actual traffic they exchange as part of the test. And they use multiple threads, and they use a non centrally tended subset of results and then they add a fudge factor for 'overheads'

It's a miracle that speedtest.net results ever match expectations.

Cheers - N

It would be a miracle if Speedtest was ever correct.

You can also follow me on twitter here @kiwifortw I do twitch streams every now and then at twitch.tv/kiwiforthewin :)

HTTP 404 Sarcasm not found.

ScuL

487 posts

Ultimate Geek

Trusted

#1216766 18-Jan-2015 01:47

Continue as in the starting post :) The more results the better :D Cheers

Haere taka mua, taka muri; kaua e wha.

linw

2849 posts

Uber Geek

#1216807 18-Jan-2015 08:47

Orcon UFB 30/10. 5.83/0.53. Plimmerton 0845.

coffeebaron

6235 posts

Uber Geek

Trusted

Lifetime subscriber

#1216816 18-Jan-2015 09:18

I got 4-6Mbps down and 0.4 up on RBI Wireless through http://www.ultimatebroadband.co.nz/ so even RBI it is possible. The test does seem a bit short though, as it didn't fully ramp up.
Perhaps a large file for us to download and post how long it takes?

Rural IT and Broadband support.

Broadband troubleshooting and master filter installs.
Starlink installer - one month free: https://www.starlink.com/?referral=RC-32845-88860-71
Wi-Fi and networking
Cel-Fi supply and installer - boost your mobile phone coverage legally

Need help in Auckland, Waikato or BoP? Click my email button, or email me direct: [my user name] at geekzonemail dot com

Quic Broadband

Move to New Zealand's best fibre broadband service (affiliate link). Free setup code: R587125ERQ6VE. Note that to use Quic Broadband you must be comfortable with configuring your own router.

haywardb

84 posts

Master Geek

ID Verified

Subscriber

#1216822 18-Jan-2015 09:32

Sun Morning @ 9:30am on Orcon 100/20 connection.
6.04 / .53

ScuL

487 posts

Ultimate Geek

Trusted

#1216823 18-Jan-2015 09:35

coffeebaron: I got 4-6Mbps down and 0.4 up on RBI Wireless through http://www.ultimatebroadband.co.nz/ so even RBI it is possible. The test does seem a bit short though, as it didn't fully ramp up.
Perhaps a large file for us to download and post how long it takes?

cool that's interesting to know.

Extending the duration would increase the throughput yep but so far the averages seem to be sufficient for streaming quality so the end result will only be better :)
Also, loading a larger file would increase the risk of more than one test running simultaneously plus increase server load/bandwidth.
On paper it's better but for what I need it to do it's fine like this :)

Haere taka mua, taka muri; kaua e wha.

plambrechtsen

1948 posts

Uber Geek
Inactive user

#1216827 18-Jan-2015 09:57

ScuL: Yep however that is exactly my intention.
I am streaming over HTTP (TCP packets) so it makes no sense to measure UDP.
The server has an upload cap of 9mb anyway, so I am not looking for the highest throughput. This is hosted on a domestic cable connection.

My first experiment was to find out what the "viewable" stream quality would be from NL to NZ.
I can now conclude that pretty much anywhere in NZ would be able to obtain SD quality streams compressed at 1.3mbps, and so far everybody responding to this thread could pull at least 3mbit in HD.

My second experiment is to find out if there are any notable differences on the various ISPs, the results are too few to be conclusive at the moment but I'm quite surprised by the Orcon reading being faster than what I am getting here in the UK!

Thanks so far for all contributions - everybody posting a reading will get a +1 :)

The problem is you are using flawed testing methodology. A speed test isn't really going to find out if there are any differences between ISPs.

I would safely assume all ISPs should be able to deliver a 5+mb sustained rate from the other side of the world. So I am really wondering what you want to test. As the baron suggested perhaps hosting a 500mb file and asking how long it took to download using wget may be a better way to test as that is single threaded, real world transfer rate.

I would say trace routes to see which direction of the world the traffic travels in depending on the upstream provider would be of more use, plus ping times as it's the TCP round trip which kills you. The issue with ICMP Pings is that they are often de-prioritised so it too won't give you a real world view.

In short, the only way to find out what you want to do will work... is to do it yourself. As there are literally hundreds of different factors that could impact streaming speed. And just because it's working today, doesn't mean it won't get broken tomorrow by whoever you choose as an ISP.

So I am really wondering about the whole reason to bother doing this? Unless the link is booby trapped with some malware payload and we are getting pwned.

Worthwhile ways to test streaming:

1) Have a large file to download and see how long that takes.
2) Have multiple streams setup of a creative commons video that is HD, and re-encode using VLC on your hosting server so people can test different speeds and see if they work (or not).
3) Do something similar to NetFlix Example short: http://www.reddit.com/r/NetflixBestOf/comments/23m338/example_short_23976_2010_11m_test_your_netflix

ScuL

487 posts

Ultimate Geek

Trusted

#1216909 18-Jan-2015 12:53

I am going to have to argue some of your points :)
First and foremost I wanted to establish if the minimum speed of 1.3mbit/s for SD quality would be attainable. I had no idea if this would work prior to starting this thread and thanks to the responses given it seems that regardless of being on UFB, cable or even rural broadband the bandwidth is sufficient to get a picture.

Secondly I had to select a method that would be straight forward and quick to execute for any willing participant. If you ask "strangers" on a forum to do something for you but also ask them to install one more applications or complicated configurations then the amount of feedback one can expect would be severely reduced.

I currently don't have access to any client side internet connection in NZ so the only way I can test this is by asking people to test it for me.

When it comes to ICMP/ping signals I agree with you hence why I didn't ask for any results specifying ping.
You are right again when it comes to traceroutes but again I didn't want to complicate things. Number of hops could be a valid testing method. If you have a recommendation how this could easily be conducted feel free to mention it.
Would you suggest it would make any sense for me to run traceroutes to servers within the network of the various ISPs from my end?

Comments on the various options
1) Yes I agree it would be better with a single threaded larger file, however as said, hosting this on a domestic connection it means that only one "speed test" could run simultaneously. The only ways I can imagine doing that would be #1 to contact people individually and ensure they're not testing at the same time or #2 limit the apache server to serve to only 1 client and give the others attempting it a 403 busy notice. Also I don't want to flood my parents cable connection by needlessly pumping through dozens of 500 meg files, albeit with the time difference most of the testing would occur at night time.
2) Again I would have to do this on a per person basis
3) similar to 2

Thanks for thinking with me though!

Haere taka mua, taka muri; kaua e wha.

plambrechtsen

1948 posts

Uber Geek
Inactive user

#1216937 18-Jan-2015 14:09

I would personally be deeply concerned if any ISP in NZ couldn't sustain a 2mb connection to NL or UK. That is definitely at the lower end of speeds these days. 15mb is the fastest anyone can get due to 330ms tcp round trip or so. So that would be the upper end of any single tcp thread without tweaking at both ends.

Most large ISPs count their international connection rates in the 10gb/s or more.

Most often if you hit ISPs servers they use for hosting. Especially with the larger ones. That infrastructure is completely different from the Internet connections for their customers so I am not sure of the value in doing it.

I can normally sustain a 900KB on a 14mb ADSL connection to my VPS in the UK. So that is good for normal SD and HD at a pinch. Again tweaking of my tcp window was required to achieve it.

Internet connection in NZ is on average pretty good to excellent. The only areas are rural or someone connected to an ASAM or conklin as they have backhaul congestion issues that are not going to be solved any time soon.

KiwiSurfer

1462 posts

Uber Geek

ID Verified

Lifetime subscriber

#1217030 18-Jan-2015 17:37

I know it's not UFB but just for your information I have Telecom/Spark ADSL2 here and I get 2.74mbps down and 0.20mbps up. Not bad I think given that the DSL line speed is only 14.71mbps.

While I can see plambrechtsen's point I still think this is a interesting exercise just to get an idea of what is possible. I sometimes stream videos from a UK site and they stream pretty well considering we are on the other side of the world! It's not very consistent though -- it can be very poor from time to time but mostly it streams fine with minimal buffering (usually just at the start).

- James

BlackHand

131 posts

Master Geek

#1217111 18-Jan-2015 20:28

Orcon, 100/50, Silverdale/Auckland = 6.35 / 0.53 @ 20:25

dwl

371 posts

Ultimate Geek

#1217922 19-Jan-2015 22:13

On Spark rural ADSL I get 4.2 Mbps down which is not much less than my max of about 5.5 Mbps.

A note about speedtest mini - this is using two threads for the download (bandwidth share between two downloads so might mislead) but if I try a single thread with wget and a bigger file version from your server with say: wget http://scul.zapto.org/mini/speedtest/random3000x3000.jpg I see peak speed of 5.3 Mbps and average of 4.1 Mbps for this 17 MB file so TCP seems quite happy with single thread over this distance (in case that is what your app uses).

1 | 2 | 3