Buffering distorts test result

[Skitter]

My broadband goes from a happy average of say 8Mbps to seizing completely. I immediately retest but, because of the seizure, the test cannot even start (buffers) until the seizure is cleared. Then the internet recovers so the retest often shows an improved download speed (!) and completely fails to illustrate the  total lack of download at the beginning of the test because it does not reflect the time it takes for the test to start when there is absolutely no download volume at all.

 

The test needs to be modified to include how long it takes to start the test and that time should then be factored into the download test time which would give a much more accurate picture of overall performance.

[CA3LE]

I see that you're using the multithread test.  If you disable multithread the progress bar will load much more smoothly.

 

That progress bar on multithread is more jittery than the linear test because of how the data is loaded.  Each thread has a larger chunk of data, those chunks can lag behind together, finish simultaneously... or one by ..... ... one.  As each chunk completes it updates its percentage of the overall task.  If you observe the data flow with your computer's networking tool you may see a smooth flow when the progress seems to be updating erratically, or having a long start. 

 

I have more programming functionality at my disposal these days so I'll love to revisit how that multithread test works in the future.  I should at least attempt to show you what's happening... give you an idea why you might not be seeing updates to the progress right away.

 

When it has a long start like that it's because it may be opening 8 threads, before the test can see that anything has happened it needs at least one of those threads to finish and say, "I'm done with that X kB chunk of data ... update the overall percentage X %"

 

Again, watch a networking tool during the test, try different test options, it may surprise you and give you a better feel for how TestMy.net works under the hood. 

 

The linear test loads more smoothly because I'm serving you the data more predictably.  See My Settings to disable/enable multithread.

 

[Skitter]

@CA3LE

 

Thank you for getting back so quickly. I have disabled the Multithread option. Basically I am trying to create a testbed report that demonstrates to my ISP the seizures that I mentioned above so any suggestions as to the best way to set up the test would be gratefully received. I have also altered the frequency to every 5 mins, as the hourly test was often missing the lowpoints in performance.

 

Again, my thanks for your help.

[CA3LE]

Seeing your results come in now..

 

TestMy.net Test ID : n70ThccLf

 

TestMy.net Test ID : DwT~fPmp1

 

See how the classic test also has more resolution.  You can see more about how the transfer flowed.  Under My Results scroll to the details and click the Test_ID.  

 

To fluctuate between 6 and 18 Mbps... that's pretty heavy.

 

 

Site note: 

 

I was curious when the multithread test started, ended up being a year earlier than I remember.  June 2013.  

 

https://web.archive.org/web/20130619022018/testmy.net/multithread

Version 13.37  

 

I've made no changes to the core of how that test works since it was established.

[Skitter]

I just wanted to say that the  testpoint at 12:14 took over 3 mins to ocmplete due to a bottleneck during which is stood still. So 19Mbps is just not giving the whole picture. I checked the Test iD which does not really help.

 

Incidentally is the 94% variance in the Latency Test an indication of the volatility of performance?

 

Thanks

 

Kit

TestMy Latency.pdf

[Skitter]

Sorry I should have stated  94% variance in London.