false reading on speedtest.net

cholcobo3

Why does my 4mbit internet connection read as 12 or more on speedtest.net?  Doesn't happen on other VPN providers.  I can get the real number by using testmy.net

Is this an error?  Or dishonesty?

Also, PrivateInternetAccess seems to disconnect/reconnect if you've been online for too long.  This morning I woke up and my test was complete: doesn't happen with AirVPN.

OmniNegro

Most speed test sites use burst speed as the single measure they bother with. This gives bigger numbers that imply a better speed, but it does not pan out in the long run. I will be copying this into the other thread you posted about this in.

cholcobo3

OmniNegro no they don't.  In fact speedtest.net discards the top 10% of readings and bottom 30% of readings to ignore burst speed.  Also, testing for more than one second would average out burst speed errors to nothing in no-time.  It doesn't even need a long-run as you say.  speedtest.net does use multithreaded, which is a subtle difference from testmy.net's single-threaded test, but I don't think you'd understand the difference if I tried to explain it, because of your burst-speed comment indicating that it would be over your head.

OmniNegro

OmniNegro no they don't.  In fact speedtest.net discards the top 10% of readings and bottom 30% of readings to ignore burst speed.  Also, testing for more than one second would average out burst speed errors to nothing in no-time.  It doesn't even need a long-run as you say.  speedtest.net does use multithreaded, which is a subtle difference from testmy.net's single-threaded test, but I don't think you'd understand the difference if I tried to explain it, because of your burst-speed comment indicating that it would be over your head.

You sir/madam are trying to irritate me. No luck for you. I do understand mutithreading quite well. And I guarantee you would be surprised at how much I know on this subject. But let me pose a question for you:

You say that multithreading should make speedtest.net better than testmy.net. Fine. How much bandwidth are you pushing? Enough to matter for even a cheap multicore netbook to matter is doubtful.

Just how much overhead do you think is involved in the networking side of things? 1% of one core is more than enough for an unencrypted connection maxing out a gigabit Ethernet setup. How about with the VPN running AES-256 as the cipher, SHA-256 as the hash, and RSA-2048 as the authentication?(Not that the authentication would matter more than once per session unless you lost your connection to the VPN.) You could in that case choke a single core on a slow CPU. But in that case your statement of a multithreaded setup working better would be correct.

But that is dependent upon higher speed networks than most people have access to.

cholcobo3

I'm still confused... why does speedtest.net show a speed three to five times faster than the actual, real speed?

OmniNegro

I'm still confused... why does speedtest.net show a speed three to five times faster than the actual, real speed?

This is not anything I can back up, but I believe they may have noticed the fact that people use their speedtest more often than others because it shows erroneously high results. Just like how people prefer to drive cars that can reach speeds higher than they can legally drive anywhere. They cannot use it, but they like to think they could.

This is just a hypothesis. Please feel free to ignore it.

cosmoxl

I've seen this before with my own connection.

All I can think of is that the data is sent in such a way that the header compression that PIA uses in the openvpn connection is accounting for the large increase.

If many udp packets are sent with very short length, such that the header makes up a significant percentage of the size of the packet, could header compression make it seem as though speed is that much higher?

OmniNegro

LZO is certainly better than deflate for compression. It is a lean compression method, but it does a good job and has virtually no overhead. If you use LZO to compress a text file with 1000 bytes of junk repeated over and over, then it will compress it to just over 1000 bytes total size.

So yes. Without a doubt if many tiny packets are sent, the compression could easily make it look like you had many times the bandwidth.
(I believe LZO is the same actual compressor as LZMA. If not then I could easily be overstating it's potency.)

*Edit* Nope. I somehow missed it the first search. Here is LZO on Wikipedia.

Ryan123

Just sharing my own experience. A few weeks ago I had to check my internet speed with speedtest.net, but the readings was not correct. I thought it was the issue with My ISP. Then I had to contact my service provider http://acanac.com in Ontario. But they informed me that the reasons are burst speed and some network issues. Yesterday also I found the same issue in speedtest.net.