[TelstraClear] Being Throttled on WarpSpeed 100Mb plan

Forums › One New Zealand (including Vodafone, WxC, Farmside) › [TelstraClear] Being Throttled on WarpSpeed 100Mb plan

1 | 2 | 3

kentnl

23 posts

Geek

#775997 7-Mar-2013 04:58

No, it's the rules of physics. Have you read the third article I linked to - it probably has the best explanation. Just ignore the fact it was written nearly 20 years ago, the concept still stands.

So every test from NZ -> US over 50Mbit is "fake" ?

So every test from Japan -> US over 50Mbit is "fake"?

You realise of course, that if "physics" is the limitation, being "the speed of light can only move so far in a certain amount of time", then "light can orbit the circumference of the earth in only 133ms" suggests that its not such a big deal.

But you don't know who did the test, and where from that got that speed. It may well be a techo at an ISP running it from a server on their core network that is going to (in terms of latency) be far closer to the test server than any real world customer will ever be. You see the odd speed test image jump up with people showing gigabit speeds from a speedtest server - these are generally from someone who works at an ISP that actually hosts the speedtest server on their network. Basically, ignore the very extreme test results that are out there - they are artificial, not real world.

Actually, I do , I have message transcripts from the person doing the tests to the test server, and the test server has no servers in nz or australia

I don't know exactly where they're doing it from in NZ, they could have some very plushy deal with some very edge network, but they are most definitely getting these speeds from NZ.

http://testmy.net/quickstats/andrewm

kentnl

23 posts

Geek

#776003 7-Mar-2013 06:23

Right, using the TCP Throughput equation:

Server: mirrors1.kernel.org ( 149.20.20.135 )

Ping : ~145ms

512kbit buffer / 0.145 => 3.5Mbit ( 441KByte ) throughput?

Ok, that makes sense; 12MB File transferred in 27 seconds seems to make sense there.

Though It would seem you could get around the speed sucking by pumping the buffer size everywhere to adjust.

ie: to deliver 50M/s throughput at that latency, having a 7250kbit buffer would make that sort of throughput viable.

Aha. Interesting.

Applying all the settings here : http://www.cyberciti.biz/faq/linux-tcp-tuning/ , I managed to get the above 12M file in only 11 Seconds instead of 27.

And multi-segmented downloading, 3 seconds => ~ 33M/s transfer.

kentnl

23 posts

Geek

#776017 7-Mar-2013 07:42

Seem about right for a 65k cache size?

jnawk

176 posts

Master Geek

#776025 7-Mar-2013 08:12

kentnl:

No, it's the rules of physics. Have you read the third article I linked to - it probably has the best explanation. Just ignore the fact it was written nearly 20 years ago, the concept still stands.

So every test from NZ -> US over 50Mbit is "fake" ?

So every test from Japan -> US over 50Mbit is "fake"?

You realise of course, that if "physics" is the limitation, being "the speed of light can only move so far in a certain amount of time", then "light can orbit the circumference of the earth in only 133ms" suggests that its not such a big deal.

This is pure gold! The concept being 20 years old is *exactly* why it no longer applies. I've not read the article, but no doubt it's going to go on about the latency limiting maximum throughput by limiting the number of packets 'on the wire' at any time? If so, the solution to that is TCP sliding windows. All hosts since about 1996 have supported it.

You might want to check out this thread.

Nick chose his countries carefully when he said US or Germany.

jnawk

176 posts

Master Geek

#776027 7-Mar-2013 08:16

RunningMan:
Try multi threading your downloads and see if you can get increased throughput.

Which will only prove that there is no problem with your connection to the host you are trying. If you see increased speeds, this is further evidence you _aren't_ getting what you are paying for.

You'll get nowhere though - TelstraClear can't figure it out themselves. I certainly got nowhere. I was upset about 1 to 3 MBit connection speeds via a single TCP session over an encrypted (SSH, not SSL) session (=non-cacheable, no transproxy interfering). Increase the threads (ie, spawn more SSH sessions), and the speed approach what is advertised on the box. Any time.

The conclusion was that TelstraClear had (have?) rubbish peering arrangements in place, and small old Xnet (WorldXchange) have better arrangements.

I sympathise with you, but you're pissing into the wind.

kentnl

23 posts

Geek

#776047 7-Mar-2013 08:48

jnawk:

This is pure gold! The concept being 20 years old is *exactly* why it no longer applies. I've not read the article, but no doubt it's going to go on about the latency limiting maximum throughput by limiting the number of packets 'on the wire' at any time? If so, the solution to that is TCP sliding windows. All hosts since about 1996 have supported it.

You might want to check out this thread.

Nick chose his countries carefully when he said US or Germany.

Though, when I applied the given math to the given ping times, I got numbers in approximate range of what speeds I was getting, so the math IS still relevant.

The problem is mostly that most hosts have their buffers optimised for LAN traffic, high throughput, but low latency, so the long ping times are not such a problem. However, to mitigate the negative cost of increased ping times, you need to pump the buffer size.

I haven't worked out where buffers need to be pumped, I found that I could pump their size locally, but only got benefits to a point. If the theory was concrete that its just RWIN / RTT = THROUGHPUT , then if RTT was too big, you could just make RWIN monsterous to match, however, when you try that, it doesn't deliver quite as expected.

jnawk

176 posts

Master Geek

#776089 7-Mar-2013 10:03

I wish I could help. The fattest pipe in NZ I've got access to is about 20mbit/sec. (I've got a much fatter one nationally, but it gets about 10mbit/sec internationally (London)).

I've got a big pipe in London too, but that won't prove anything whatsoever!

Quic Broadband

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

RunningMan

8955 posts

Uber Geek

#776109 7-Mar-2013 10:20

kentnl:
Actually, I do , I have message transcripts from the person doing the tests to the test server, and the test server has no servers in nz or australia

I don't know exactly where they're doing it from in NZ, they could have some very plushy deal with some very edge network, but they are most definitely getting these speeds from NZ.

Sorry, didn't make myself very clear. I wasn't suggesting they were testing to a server in NZ, but more along the lines of your second comment - you don't know where in NZ they are doing the test from, and that can greatly influence the result.

Ragnor

8219 posts

Uber Geek

Trusted

#776317 7-Mar-2013 15:41

This single thread throughput issue on Telstraclear upstream has been around forever, pretty bad they still haven't fixed it.

michaelmurfy

meow
13242 posts

Uber Geek

Moderator

ID Verified

Trusted

Lifetime subscriber

#777021 8-Mar-2013 17:30

Anyone else having awesome (1sec+) ping to the US on the Warpspeed plan? Our internet is going as fast as a old dog.

Michael Murphy | https://murfy.nz
Referral Links: Quic Broadband (use R122101E7CV7Q for free setup)

Are you happy with what you get from Geekzone? Please consider supporting us by subscribing.
Opinions are my own and not the views of my employer.

kentnl

23 posts

Geek

#777023 8-Mar-2013 17:45

huh. Nope. what server is this to? I'm getting <300 to most US servers still.

kentnl

23 posts

Geek

#777025 8-Mar-2013 17:47

Also, include a traceroute if you can, I find 'mtr' good for seeing where the crunchpoint is with pings in the route.

michaelmurfy

meow
13242 posts

Uber Geek

Moderator

ID Verified

Trusted

Lifetime subscriber

#777069 8-Mar-2013 20:18

Ah scrap that, looks like it might just be my Linode.

Tip: Never run ping from the "Run Command" part of PFsense, there's no way of stopping it without proper shell access as it completly locks up the web interface.

SilverwolfNZ

62 posts

Master Geek

#783002 17-Mar-2013 16:20

RunningMan:
kentnl: [
Its nice that TCL have a 100M plan, but I've yet to meet anyone on that plan who are actually getting what they're paying for. Its great to have 100M for local connections, ie: to Snaps test server .... but having it drop to 10% for ALL international traffic just makes that extra payment a joke.

If you are not getting what you are paying for, then raise it with them and complain. However, as Johnr noted, unless you are on a business plan with an SLA, then you are getting what you pay for. There is no guaranteed speed for a single international connection - this is affected by many things out of TC's control, including the server at the other end you are testing from.

Try multi threading your downloads and see if you can get increased throughput.

Multi-Thread on a Orcon plan w/ ADSL2+ 12Mbps(max local) plan is 1.6MBps

Multi-Thread on a Telstra plan w/ 98Mbps(max local) plan is 2.6 MBps.

This thread has gone off-topic, as per the previous response by Kent it was a switch somehow throttling us this end which we simply solved by "Turning it off then back on again".

1 | 2 | 3