Acceptable speed for a Gigabit Fibre Connection

Forums › One New Zealand (including Vodafone, WxC, Farmside) › Acceptable speed for a Gigabit Fibre Connection

1 | 2 | 3 | 4

wired

187 posts

Master Geek

#1470729 14-Jan-2016 12:27

When I was testing this, we didn't find any laptops around the office that could do more than about 600 Mbps so we had to use a high performance desk computer to get anywhere near 1 Gbps. So I agree with others about needing to check that the computers you are using can handle the speed.

We also tried the test in firefox/chrome/IE as that can vary the recorded speed significantly.

The NIC you use is important. The speedtest application does TCP offloading to the NIC. This means that the application isn't sending the normal 1500 byte packets, they are much larger, as this fragmentation is taken care of by the NIC. This means that the bandwidth for the acknowledgement packets is much smaller than would be normally required thereby reducing the effect of the reverse direction on the test traffic. TCPoffloading is found on 1 Gbps and 10 Gbps NICs because the overhead with processing TCP/IP becomes significant at the higher speeds.

When doing the testing, I would also test the router between two local computers just to see that it can handle what it is rated at.

Nebbie

458 posts

Ultimate Geek

Trusted

#1470799 14-Jan-2016 13:26

I have heard from a friend that a number of customers despite having Gig Ethernet adaptors the total throughput for the card never gets anywhere near 1gig often the downstream is affected more than the upstream. This is the downside of cheep network cards I would not be inclined to say this if more than one devices was tested with similar results. To the OP: sounds like you know people with Gig, take your PC to their house and test see if you get the better results.

---------------------------------------------------------------
Nebukadnessar

richms

28172 posts

Uber Geek

Trusted

Lifetime subscriber

#1470921 14-Jan-2016 15:10

Friend gets 22megabytes a second on their steam store stuff on a 200/200 connection on the good desktop. Only gets 14 on the laptop. Cant blame the ISP for that.

Richard rich.ms

dylanp

840 posts

Ultimate Geek

Trusted

#1471059 14-Jan-2016 16:23

richms: Friend gets 22megabytes a second on their steam store stuff on a 200/200 connection on the good desktop. Only gets 14 on the laptop. Cant blame the ISP for that.

I find it fascinating how the bottleneck has shifted from the internet connection to the local network. It's quite amazing really, that rate of internet connection speed increase we've had.

Batman

Mad Scientist
29762 posts

Uber Geek

Trusted

Lifetime subscriber

#1471063 14-Jan-2016 16:27

trig42: MB I think, not mb. /pedant
900 mb ~ 125MB

Umm incorrect. little m = milli
accepted SI abbreviation for Mega is M

madmatt

15 posts

Geek

#1472921 17-Jan-2016 00:03

Seems like a bit of misinformation going around here.

denemc: As people have suggested, please do try the following:

Asking someone to let you take your laptop around to their house and try it out on their connection (make sure you take your LAN cable too, and test that at the same time - it wouldn't be the first time a terribly wired Cat5e cable has ruined things!)
If you know someone who gets GigE at home, get them to come around and try their machine at your place as well.
Make sure your wireless is disabled - just because you plug ethernet in doesn't mean your laptop will use ethernet - I'm not sure how Windows (if that's what you're using) picks which connection to use, but in my experience it always tends to pick the worst one (possibly picks the 'longest connected' which is often wireless)
As richms stated, if you've got any security/firewall software, check or disable it for the test.

If you find your laptop isn't up to scratch, it's most likely because the NIC is passing packets to the CPU to handle, and getting the CPU to do the heavy lifting. Some onboard NICs do this, especially in laptops. You could watch out for this in Task Manager - check your CPU% as you run a speedtest for example. However you'd generally notice the laptop being sluggish if this was the case.

On the other hand, if your laptop performs fine at another house, go back to Vodafone and ask them to get the connection fault tested. Get a tech to come and visit, and get them to prove to you that you're receiving 1Gbps or equivalent with their testing hardware. You mentioned that Chorus never tested the connection when they hooked it up - while that might be true, it might be that you just didn't see them test it - but regardless, get them back in. If their testing hardware shows you're getting decent speeds, you'd likely have to pay for the callout (hence the above suggestions first), but it's not an exorbitant cost.

You mention that your laptop should be able to do better, and that other slower laptops do better than yours. Unless someone else has the exact same model laptop as you, with the same firmware revisions and hardware specs, and unless they are getting significantly better speeds than you with the same hardware, just discount that theory. There's any number of things about your laptop that might be causing issues - rule it out using the suggestions above and by other people - otherwise nobody can really help you.

michaelmurfy: A core i3 with 4gb of ram actually isnt that powerful - it is on the same scale in terms of power of my Chromebook which does have a lighter weight OS. Try a better laptop.

Also, since some speed test servers are only on gigabit connections themselves it is not a good guide on what you can get on your connection.

It's not really about the 'i3' part as much as it is about the GHz and the generation of processor. It could be a really powerful processor, but without posting specs we're just guessing in the dark.

On your other point, do you have any sources to back that rumour up? I would be incredibly surprised if any of the Speedtest servers only run on a single non-bonded gigabit connection. For one, that's amateur-ish, especially given the increasing prevalence of gigabit connections, and secondly it's a terribly idea for Spark/Vodafone/whoever to do such a thing - it costs them next to nothing, especially compared to the hundreds of complaining customers they'd get if their speed test results were all terrible because the server could only theoretically handle one customer making a speed test at any one time. My guess is that those boxes have a minimum of 2x 10Gbit NICs, at a bare minimum.

michaelmurfy

meow
13242 posts

Uber Geek

Moderator

ID Verified

Trusted

Lifetime subscriber

#1472942 17-Jan-2016 00:44

madmatt: Seems like a bit of misinformation going around here.

If you find your laptop isn't up to scratch, it's most likely because the NIC is passing packets to the CPU to handle, and getting the CPU to do the heavy lifting. Some onboard NICs do this, especially in laptops. You could watch out for this in Task Manager - check your CPU% as you run a speedtest for example. However you'd generally notice the laptop being sluggish if this was the case.

You mention that your laptop should be able to do better, and that other slower laptops do better than yours. Unless someone else has the exact same model laptop as you, with the same firmware revisions and hardware specs, and unless they are getting significantly better speeds than you with the same hardware, just discount that theory. There's any number of things about your laptop that might be causing issues - rule it out using the suggestions above and by other people - otherwise nobody can really help you.

michaelmurfy: A core i3 with 4gb of ram actually isnt that powerful - it is on the same scale in terms of power of my Chromebook which does have a lighter weight OS. Try a better laptop.

Also, since some speed test servers are only on gigabit connections themselves it is not a good guide on what you can get on your connection.

It's not really about the 'i3' part as much as it is about the GHz and the generation of processor. It could be a really powerful processor, but without posting specs we're just guessing in the dark.

On your other point, do you have any sources to back that rumour up? I would be incredibly surprised if any of the Speedtest servers only run on a single non-bonded gigabit connection. For one, that's amateur-ish, especially given the increasing prevalence of gigabit connections, and secondly it's a terribly idea for Spark/Vodafone/whoever to do such a thing - it costs them next to nothing, especially compared to the hundreds of complaining customers they'd get if their speed test results were all terrible because the server could only theoretically handle one customer making a speed test at any one time. My guess is that those boxes have a minimum of 2x 10Gbit NICs, at a bare minimum.

10Gbit is expensive. Some smaller providers (like, Woosh, Amuri, "BlahHosting", Vetta Technologies, Southloop) won't even have a 10Gbit network to work with since the business cost to get 10Gbit core routers, switches and compatible hardware just simply makes no financial sense. Also there are many Speedtest servers that are hosted on servers that may host other things (eg, a website). With all those providers I quoted I can tell they're saturated since I only get around 40Mbit to 100Mbit on my 200/200 connection.

Speedtest will often default to one of these smaller providers of which you'll get a rather ho-hum speed from based on ping alone. Sure, the boxes might be connected to a fast core etc but this doesn't mean they're on anything more than 1Gbit. You just don't know. So the Speedtest requirements call for at-least 2x 1Gbit connections but there are many other factors to consider such as peering, their core router and the hardware the Speedtest is running on at the providers end.

So as for evidence? It was a given, however if I must prove to you I am right there was once a Speedtest provider in Auckland. The guy hosting it has a rather poor reputation on here and he also quoted on his Twitter "Connected by 2x Gigabit Fibre to our core!" of which soon got deleted. He must have been getting slammed with traffic since the Speedtest server since got removed around a month later and performance to it was pretty poor.

This is why Speedtest should never be used to determine your actual internet speed. It simply isn't accurate and this has been proven many times on Geekzone. With Gigatown a Speedtest to one of these providers will effectively max it out and if there is more than one Gigatown speedtest (or even if I did a Speedtest at the same time) it has the potential to skew the results.

Now I know further information from some of the providers that have servers on Speedtest and assure me, even most of the bigger providers don't use 10Gbit ethernet to their speedtest servers. It is normally 2x 1Gbit bonded links. Companies don't make any return on Speedtest servers so why should they spend a bucketload of money on bandwidth, routers, switches and servers? Again, financial sense.

Also, my server (a HP Microserver with a dual core processor) has a NIC with hardware offloading, and also the internal NIC which is connected straight to the CPU. The dedicated NIC gets around 800Mbit throughput however the internal NIC only gets around 120mbit on a good day. This is not at all a powerful server but what everyone here is saying is the CPU can be a newer generation however it doesn't mean the NIC can handle 1Gbit. Both the NIC's on my Microserver link at 1Gbit.

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.

Sharesies kids

Free kids accounts - trade shares and funds (NZ, US) with Sharesies (affiliate link).

Talkiet

4792 posts

Uber Geek

Trusted

#1472945 17-Jan-2016 01:32

I'm utterly convinced no-one here has a clue how ookla/speedtest.net actually work. The servers get by with WAY less connectivity than you might think because of the way the tests are conducted.

Our speedtest servers seem able to reliably test gigatown connections and I see many many results (from the server reporting) from people on many ISPs getting over 900Mbps.

IF we want to keep this on topic, then we'll wait for the testing of OPs machine at another connection, and the test of a known good machine at his connection.

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.

raytaylor

4014 posts

Uber Geek

Trusted

#1473381 17-Jan-2016 20:40

denemc: We only have one pc with a gigbit ethernet port so haven't been able to test with another machine.

It's interesting that it can get 450 up, but only 250 down, which sort of indicates it's not the PC slowing it down.

You should be doing an iperf test to another host - speedtest.net is flash player based and not very reliable.

Ray Taylor

There is no place like localhost

Spreadsheet for Comparing Electricity Plans Here

raytaylor

4014 posts

Uber Geek

Trusted

#1473384 17-Jan-2016 20:44

Plus I am getting double up compared to what I am getting down indicates there is a problem.

Indicated a hardware issue to me

Is common to see that sort of thing when you run an ookla speedtest across a gigabit lan

Ray Taylor

There is no place like localhost

Spreadsheet for Comparing Electricity Plans Here

raytaylor

4014 posts

Uber Geek

Trusted

#1473385 17-Jan-2016 20:48

VodafoneDylan:
richms: Friend gets 22megabytes a second on their steam store stuff on a 200/200 connection on the good desktop. Only gets 14 on the laptop. Cant blame the ISP for that.

I find it fascinating how the bottleneck has shifted from the internet connection to the local network. It's quite amazing really, that rate of internet connection speed increase we've had.

Its interesting because I think sbibble has been saying that for years with regards to internal copper wiring.

Ray Taylor

There is no place like localhost

Spreadsheet for Comparing Electricity Plans Here

raytaylor

4014 posts

Uber Geek

Trusted

#1473388 17-Jan-2016 21:12

Talkiet: I'm utterly convinced no-one here has a clue how ookla/speedtest.net actually work. The servers get by with WAY less connectivity than you might think because of the way the tests are conducted.

Our speedtest servers seem able to reliably test gigatown connections and I see many many results (from the server reporting) from people on many ISPs getting over 900Mbps.

IF we want to keep this on topic, then we'll wait for the testing of OPs machine at another connection, and the test of a known good machine at his connection.

Cheers - N

I used to run a speedtest.net node in Napier.

And yes like other smaller providers it was on a 1gbit connection. I took it down because it kept getting slammed by people from other countries such as india and random parts of asia.

So to run a speedtest.net node all you need is a 100mbit or higher connection.

The machine you run it on can be anything - ours was an intel NUC i3 with 2gb of ram and windows 7 pro.

Ookla provides some software which runs an http server on some random port, for windows or linux. It serves up jpg files of predefined sizes and resolutions. It might have speedtest500x500.jpg .... speedtest10000x10000.jpg

The jpg files are completely randomised pixels which makes it hard to compress.

The client
==============

1) Starts by running an http GET for a 1kb text file on the server. This is how it times your pings. Its not actually a ping, its simply downloading a very small file. If the http server is under load, it may slow this down. You will probably get 5ms faster if you were to actually ping the server rather than just download the http files. But its a genius method when working within the restrictions of a flash player client.

2) The download test starts. It downloads a 100kb file and times how long that takes. You are about 10% through the speedtest by now.

3) Depending upon the time it took to download the 100kb file, it will start multiple downloads of larger jpg files, at the same time. Interesting factoid: If you are on a connection faster than 100mbits, your average speedtest consumes 1 gigabyte of data. Up at this level is where I notice your CPU and graphics capability will affect your result. Especially if the client is running via an RDP session.

4) The upload begins. The client generates random data and does an http post to the server this recieves the files and drops them from memory. The client times how long it takes to send the files, and it selects sizes based on your download speed.

5) An http post to ookla servers to report your statistics for their data mining goblins.

Also something interesting, here is my speedtest running in firefox on my dual core cpu - flash player can only access a single core of course, and you will notice that an 80mbit download uses approx 30% of the CPU core (no idea why) and the 90mbit upload uses approx 10% of the CPU core.

So thats just one reason why uploads can be faster than downloads.

Ray Taylor

There is no place like localhost

Spreadsheet for Comparing Electricity Plans Here

Coil

6614 posts

Uber Geek
Inactive user

#1473392 17-Jan-2016 21:32

@raytaylor

Big read man but well said.

So when do you want a job at Ookla.

1 | 2 | 3 | 4