[Snap] UFB International speed

Forums › 2degrees (including Slingshot, Orcon, Flip, Stuff Fibre, MyRepublic, 2talk and Vocus) › [Snap] UFB International speed

karizzma

2 posts

Wannabe Geek

#116539 1-May-2013 23:11

Just wondering if I could some international speedtest results on snap UFB 100/50. Been looking for some but I haven't been able to find any.

1 | 2

487 posts

Ultimate Geek

#809956 2-May-2013 08:46

What locations internationally?

Sydney:

karizzma

2 posts

Wannabe Geek

#810221 2-May-2013 15:04

What about west coast USA? LA, California?

Noodles

487 posts

Ultimate Geek

#810262 2-May-2013 15:43

LA:

Have seen this much higher though. With 100Mbit you get much better throughput with multi threaded downloads (which speedtest does not do). I can easily get 12MB/s from the US.

sidefx

3714 posts

Uber Geek

Trusted

#810285 2-May-2013 15:50

Speedtest will do up to 4 threads I think...

"I was born not knowing and have had only a little time to change that here and there." | Octopus Energy | Sharesies
- Richard Feynman

hads

392 posts

Ultimate Geek

Trusted

#810528 2-May-2013 22:46

They seem unusually high for International servers.

speedtest.net is often not particularly accurate, best to use a real world transfer.

nicegear.co.nz - VoIP, DVB and other Linux compatible hardware

Zeon

3918 posts

Uber Geek

Trusted

#810546 2-May-2013 23:11

Don't trust those speed tests. a single TCP stream to the US west coast is usually around 3mbps on even a $100,000 a month connection due to latency and TCP.

Speedtest 2019-10-14

dwl

371 posts

Ultimate Geek

#810630 3-May-2013 09:29

Speedtest by default runs two threads but providers can invoke up to 8 threads. I have seen single thread speeds from a Canadian ISP limited to 3 Mbps by TCP settings (window size) get 18 Mbps from the same server on speedtest where 6 threads were enabled. Downstream speedtest should be fairly accurate as it is just transferring large random JPG files but upstream tests can be less reliable.

Users of UFB services from providers such as Snap and Orcon have seen much better single thread speeds than 3 Mbps from many international sites but only if TCP is tuned for high bandwidth delay product (should be default on modern OS). In addition, the congestion will vary by ISP and time of day. A setting that can have a big impact is Selective Acknowledgement (SACK) as if it isn't enabled (client or server) any minor loss event will have a much greater slowdown impact on TCP.

Based on posts late 2012 (see SNAP forum section for postings on "WOW! Got International Bandwidth!?" and "UFB Poor Performance" with last post Jan 2013) SNAP users were getting high international speeds once an interface issue with Chorus was sorted.

This is a complex topic as there are many elements involved and I have yet to see a good answer on how to measure the international traffic performance of an ISP. There is only limited bandwidth into NZ and even users with different more aggressive TCP algorithms can supress other flows, potentially giving one happy user and one unhappy.

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.

CYaBro

4590 posts

Uber Geek

ID Verified

Trusted

#810635 3-May-2013 09:46

WHat sort of speeds do people get using speedof.me?

I'm busy doing a big upload so can't do any tests ATM.

Opinions are my own and not the views of my employer.

dwl

371 posts

Ultimate Geek

#810890 3-May-2013 15:47

speedof.me is an interesting tool, with real time feedback, but has some limitations for this check. It seems single threaded so may not show as fast as speedtest.net (but does relate to real world).

Being CDN based it is likely it will only measure to Sydney - source is shown on the graphic - and I don't know how it could be used to any other destination. International speeds to the US seem to be a significantly bigger issue than to Australia which is understandable due to the higher delay (extra 100ms - harder for TCP). I wait with interest for results.

unsignedint

69 posts

Master Geek

#811352 4-May-2013 21:27

I'm on UFB 30/10 and international speeds are great. Quite easy to max out my line, with a single thread. Below is a download of a linux ISO form California, US @ 3.2MB/s on WiFi.

[code]
$ wget http://mirrors.kernel.org/centos/6.4/isos/x86_64/CentOS-6.4-x86_64-netinstall.iso -O /dev/null
--2013-05-04 21:21:16-- http://mirrors.kernel.org/centos/6.4/isos/x86_64/CentOS-6.4-x86_64-netinstall.iso
Resolving mirrors.kernel.org... 149.20.20.135, 149.20.4.71
Connecting to mirrors.kernel.org|149.20.20.135|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 241172480 (230M) [application/octet-stream]
Saving to: ‘/dev/null’

39% [====================================> ] 95,683,368 3.26MB/s eta 46s ^C
[/code]

hads

392 posts

Ultimate Geek

Trusted

#811566 5-May-2013 13:48

I can max out my 100Mbit USB connection to a NZ based server but with real world transfers (rsync/HTTP) from the US I see a max of around 2-2.5MB/s.

Pulling that same file just now shows around 1.8MB/s;

hads@cobalt:~$ wget http://mirrors.kernel.org/centos/6.4/isos/x86_64/CentOS-6.4-x86_64-netinstall.iso -O /dev/null
--2013-05-05 13:43:01-- http://mirrors.kernel.org/centos/6.4/isos/x86_64/CentOS-6.4-x86_64-netinstall.iso
Resolving mirrors.kernel.org (mirrors.kernel.org)... 149.20.4.71, 149.20.20.135
Connecting to mirrors.kernel.org (mirrors.kernel.org)|149.20.4.71|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 241172480 (230M) [application/octet-stream]
Saving to: `/dev/null'

100%[=============================================================================================>] 241,172,480 1.53M/s in 2m 5s

2013-05-05 13:45:08 (1.83 MB/s) - `/dev/null' saved [241172480/241172480]

hads@cobalt:~$

nicegear.co.nz - VoIP, DVB and other Linux compatible hardware

Publius

276 posts

Ultimate Geek

#811584 5-May-2013 14:33

People shouldn't expect to max out 100Mbit/sec on UFB for a single connection.

The real reason to buy 100mbit is for total aggregation of a lot of users, not for a single user.

There are many websites that will never give you 100mbit/s downloads, and thats normal.

hads

392 posts

Ultimate Geek

Trusted

#811586 5-May-2013 14:34

Indeed, I'm trying to show that that's not what actually happens.

nicegear.co.nz - VoIP, DVB and other Linux compatible hardware

mercutio

1392 posts

Uber Geek

#813499 8-May-2013 09:17

Publius: People shouldn't expect to max out 100Mbit/sec on UFB for a single connection.

The real reason to buy 100mbit is for total aggregation of a lot of users, not for a single user.

There are many websites that will never give you 100mbit/s downloads, and thats normal.

most web sites hosted overseas are on gigabit connections, but tcp/ip ramp up can take a few seconds, and most web browsing won't go that fast for throughput, but having a faster connection can mean that you're mostly sitting on application bottlenecks as it should be.

things like php, databases etc can really slow down web sites.

and it's not unreasonable to expect 100megabit/sec to australia on 10mb files i reckon.

mercutio

1392 posts

Uber Geek

#813500 8-May-2013 09:18

hads: I can max out my 100Mbit USB connection to a NZ based server but with real world transfers (rsync/HTTP) from the US I see a max of around 2-2.5MB/s.

Pulling that same file just now shows around 1.8MB/s;

hads@cobalt:~$ wget http://mirrors.kernel.org/centos/6.4/isos/x86_64/CentOS-6.4-x86_64-netinstall.iso -O /dev/null
--2013-05-05 13:43:01-- http://mirrors.kernel.org/centos/6.4/isos/x86_64/CentOS-6.4-x86_64-netinstall.iso
Resolving mirrors.kernel.org (mirrors.kernel.org)... 149.20.4.71, 149.20.20.135
Connecting to mirrors.kernel.org (mirrors.kernel.org)|149.20.4.71|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 241172480 (230M) [application/octet-stream]
Saving to: `/dev/null'

100%[=============================================================================================>] 241,172,480 1.53M/s in 2m 5s

2013-05-05 13:45:08 (1.83 MB/s) - `/dev/null' saved [241172480/241172480]

hads@cobalt:~$

with linux, i've noticed that using more recent kernels (3.8 etc) on the sender can improve throughput, and on the receiver having at least 3.0 kernel can help.

also sometimes tweaking /proc/sys/net/ipv4/tcp_rmem can help.

1 | 2