Interestingly, I can't complete an Speedtest upload test to 2D LA at all today!

nPerf completes, but is still sub-par:

Hi All,

I have the team investigating this, will provide an update shortly on it.

Nick.

Thanks @NickMack, I appreciate you and your team's responsiveness as always! :)

SamF:

 

Thanks @NickMack, I appreciate you and your team's responsiveness as always! :)

 

All good - We need some feedback from our upstream provider around how they are policing with the additional capacity we recently added, should have more information tomorrow.

Nick.

Update

Yesterday afternoon we made an adjustment to the metrics on our US links to smooth out traffic patterns, especially upload, which has proven to be effective from the testing that we have carried out so far. There are always going to be some peaks and troughs due to TCP windowing, but after our change we have seen decent improvements.

We also have a change in the pipeline to fine-tune our shaping on these US links, we are working through configuration, QA and scheduling on this at the moment.

Nick.

Thanks @NickMack!

However, upload isn't showing any significant improvement for me:

In terms of real-world applications; I notice that it starts off at full speed, but after a few seconds drops to 2-3mbits/sec and then just sits there.

I guess this might be part of your 'shaping fine-tuning'?

SamF:

 

Thanks @NickMack!

 

However, upload isn't showing any significant improvement for me:

 

 

 

 

 

In terms of real-world applications; I notice that it starts off at full speed, but after a few seconds drops to 2-3mbits/sec and then just sits there.

 

I guess this might be part of your 'shaping fine-tuning'?

 

Can you test via bootable linux to rule out all other variables - https://www.geekzone.co.nz/forums.asp?forumid=85&topicid=239862

Nick.

Sure, will do and then I'll get back to you with the results.

Ok @NickMack, here are the results from the Ubuntu testing:

So definitely faster on the upload with nperf, and significantly faster on the download with speedtest, which is interesting.

Switching to Chrome on my Windows box I can replicate the nperf results:

Regardless, real-world uploads are still stuck at 2-3mbits/sec per thread.

SamF:

Ok @NickMack, here are the results from the Ubuntu testing:

So definitely faster on the upload with nperf, and significantly faster on the download with speedtest, which is interesting.

 

Switching to Chrome on my Windows box I can replicate the nperf results:

 

Regardless, real-world uploads are still stuck at 2-3mbits/sec per thread.

Pretty decent:

SamF:

Pretty decent:

 

I am in Auckland.

Those last 2 tests were from my normal browser in Windows.

@NickMack I actually agree with @SamF here - have noticed that international over to the US side is a little lacking as of late and have only noticed this due to my Crashplan backups (which actually go via the US for my main backup) only doing around 20Mbit currently where I've seen 400Mbit+ a little over 2 months ago. It is something I've raised and I believe somebody at 2degrees was looking into it. This coincides with what my SamKnows probe is seeing regarding international speeds not breaking 20-100Mbit on anything outside of NZ / Australia where a few months ago (when my Crashplan backups were actually completing) SamKnows was showing an average of 500Mbit to most places outside of NZ.

I completed these tests inside a Docker container on a machine that can indeed achieve (well over) Gigabit speed due to NIC teaming - download speeds are normally great.

I know this is peak time however it doesn't seem to matter what time I run these tests. My upload has about 30Mbit on it currently with parents watching something from my Plex server so isn't really utilized. Also tested with a Fritz!Box and get the same results. Apart from Speedtest not being too accurate, I've found it to actually be quite accurate in what I've tested with iperf3 to some VM's I've spawned up for testing with one of your engineers and actually more accurate than nperf which is very much CPU bound.

michaelmurfy:

 

@NickMack I actually agree with @SamF here - have noticed that international over to the US side is a little lacking as of late and have only noticed this due to my Crashplan backups (which actually go via the US for my main backup) only doing around 20Mbit currently where I've seen 400Mbit+ a little over 2 months ago. It is something I've raised and I believe somebody at 2degrees was looking into it. This coincides with what my SamKnows probe is seeing regarding international speeds not breaking 20-100Mbit on anything outside of NZ / Australia where a few months ago (when my Crashplan backups were actually completing) SamKnows was showing an average of 500Mbit to most places outside of NZ.

 

I completed these tests inside a Docker container on a machine that can indeed achieve (well over) Gigabit speed due to NIC teaming - download speeds are normally great.

 

   

 

I know this is peak time however it doesn't seem to matter what time I run these tests. My upload has about 30Mbit on it currently with parents watching something from my Plex server so isn't really utilized. Also tested with a Fritz!Box and get the same results. Apart from Speedtest not being too accurate, I've found it to actually be quite accurate in what I've tested with iperf3 to some VM's I've spawned up for testing with one of your engineers and actually more accurate than nperf which is very much CPU bound.

 

Hi mate,

Who yanked your chain :-) As per my comments earlier, we are investigating, seems to potentially coincide with adding more 10 gig lags to our AKL->USA direct express route. We are working with our upstream provider to investigate as well as making some tuning changes on our side. 

Once I know more, I'll share (as I do..) ;-)

Nick.