Geekzone: technology news, blogs, forums
Guest
Welcome Guest.
You haven't logged in yet. If you don't have an account you can register now.


View this topic in a long page with up to 500 replies per page Create new topic
1 | 2 | 3 
raytaylor
4014 posts

Uber Geek

Trusted

  #986872 13-Feb-2014 16:54
Send private message

RuralJohnny:

One thing I noticed on testmy.net is that download speeds vary by file size.  Performance comes with larger files!  That has to be a function of network design.

This speed-file size relationship was confirmed at nzspeedtest.com but again, with tests to servers in NZ and Aus, the reported speeds were significantly less than ookla's.



Speedtest.net does a quick sample test with a ~200kb download. If the file comes in nice and fast, it then starts up 4 to 5 parallel downloads of different file sizes ranging from 500kB to 100mB. This is where you often see the speed test jump in speed after a second or so into the test.

So this means that they can test a dialup modem, and a fiberoptic conneciton.

The TCP protocol sits underneath all this. It has a speed control mechanism where
The first packet is sent at a slow speed, and as responses come in, it will start sending the packets faster and faster until the speed of the responses levels off. This indicates the maximum speed between the server, through all the pipes, to the end user.
As incoming packet responses slow down, it slows down the sending of the packets. The idea of this is so that packets dont get dropped or disappear in congested pipes along the route.

It takes time to speed up the file transfer because it has to take incremental samples. So on a reasonably fast connection (1 megabit+) the file could have been transfered by the time the incremental speed increases have reached the maximum throughput. As you get more and more distant (network hops and network distant, not always physical distance) the time it takes for the speed to increase will be longer.
So multiple files are transfered at once with the speedtest.net service. This means that each one will be slowly speeding up, but the total throughput can be measured by adding the current speed of each transfer together.

Underneath that, is the cell tower TDMA.
As your client device (mobile phone, 3g stick, rbi connection) starts transfering files, the cell tower will start to allocate more timeslots to it. This is slightly delayed by the fact the tcp connection isnt running at full speed, and so the cell tower also takes measurements and will increase the time slots as required.

With more devices attached to a cell tower, it takes longer to allocate those time slots and pull them from devices that are not transmitting or receiving at high speed so they dont need as many slots. This further delays the speed that the tcp transfer can take place.

In the end, big file transfers have more time to get up to speed.

See here http://en.wikipedia.org/wiki/Slow-start




Ray Taylor

There is no place like localhost

Spreadsheet for Comparing Electricity Plans Here




Talkiet
4793 posts

Uber Geek

Trusted

  #986892 13-Feb-2014 17:29
Send private message

raytaylor:
Lias: While we have this vast pool of combined knowledge posting in here.

Given everything you've mentioned, what would you suggest is the best way for an end user to get a clear picture of how well their connection is performing for international data(which is how must users perceive "internet performance")?



That is difficult.
All popular youtube videos are served up from caches within new zealand. Akamai runs local CDN nodes that distribute content for their customers (windows, apple updates etc)
So you never know if a website you are visiting is going to serve the content from NZ or overseas.

Anyhow, the best way I can suggest direct international performance is to pick a bunch of speedtest servers on the western seaboard of the united states and average the results.


That's only going to test performance to the Western Seaboard of the US. There's an AWFUL lot of traffic that goes the other way (Through Aussie and Malaysia)

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.


RuralJohnny

33 posts

Geek


  #986910 13-Feb-2014 17:47
Send private message

raytaylor: ...

The TCP protocol sits underneath all this. It has a speed control mechanism where


Thanks Ray, I am learning a lot.

I understand that TCP was designed for a wired world where congestion was more of an issue than dropped packets due to a noisy or mobile environment.  Congestion is something that is within the network designers control to some extent. For example, I wonder if the congestion window has been resized (increased) to enable a faster start?

But even so, when I see speedtest.net's pics showing the startup ramp being 6% of the transmission time, I do wonder.  Surely it does not take seconds for the congestion window to ramp up to full speed?

And if it is about congestion, then the network design could provide for a reduced contention ratio to improve the service quality.



JohnButt
374 posts

Ultimate Geek

Trusted

  #986912 13-Feb-2014 17:51
Send private message

raytaylor: 

It takes time to speed up the file transfer because it has to take incremental samples. So on a reasonably fast connection (1 megabit+) the file could have been transfered by the time the incremental speed increases have reached the maximum throughput. As you get more and more distant (network hops and network distant, not always physical distance) the time it takes for the speed to increase will be longer. 
So multiple files are transfered at once with the speedtest.net service. This means that each one will be slowly speeding up, but the total throughput can be measured by adding the current speed of each transfer together.


Ray, this may be a bit of an exaggeration for a mobile test running at typical speeds of 1-5Mb/s.  Our experience is that 1MB files (8Megabit file) is that rampup is complete inside the first 100kB

Here is a 5MB file showing the rampup completed well inside the first quartile



And here is a 300kB file, i.e. 2.4Mb file on Vodafone Network in Wellington, which clearly finishes rampup before 75kB is downloaded.  This chart is typical of our mobile downloads.

1 | 2 | 3 
View this topic in a long page with up to 500 replies per page Create new topic





News and reviews »

Air New Zealand Starts AI adoption with OpenAI
Posted 24-Jul-2025 16:00


eero Pro 7 Review
Posted 23-Jul-2025 12:07


BeeStation Plus Review
Posted 21-Jul-2025 14:21


eero Unveils New Wi-Fi 7 Products in New Zealand
Posted 21-Jul-2025 00:01


WiZ Introduces HDMI Sync Box and other Light Devices
Posted 20-Jul-2025 17:32


RedShield Enhances DDoS and Bot Attack Protection
Posted 20-Jul-2025 17:26


Seagate Ships 30TB Drives
Posted 17-Jul-2025 11:24


Oclean AirPump A10 Water Flosser Review
Posted 13-Jul-2025 11:05


Samsung Galaxy Z Fold7: Raising the Bar for Smartphones
Posted 10-Jul-2025 02:01


Samsung Galaxy Z Flip7 Brings New Edge-To-Edge FlexWindow
Posted 10-Jul-2025 02:01


Epson Launches New AM-C550Z WorkForce Enterprise printer
Posted 9-Jul-2025 18:22


Samsung Releases Smart Monitor M9
Posted 9-Jul-2025 17:46


Nearly Half of Older Kiwis Still Write their Passwords on Paper
Posted 9-Jul-2025 08:42


D-Link 4G+ Cat6 Wi-Fi 6 DWR-933M Mobile Hotspot Review
Posted 1-Jul-2025 11:34


Oppo A5 Series Launches With New Levels of Durability
Posted 30-Jun-2025 10:15









Geekzone Live »

Try automatic live updates from Geekzone directly in your browser, without refreshing the page, with Geekzone Live now.



Are you subscribed to our RSS feed? You can download the latest headlines and summaries from our stories directly to your computer or smartphone by using a feed reader.