Speedtests on VM

So, I've always gotten poor results from the TBB speedtest. I have always just swallowed the line of it being due to lack of thread depth on the test.

This is what I got just now on the TBB speedtest.

https://www.thinkbroadband.com/speedtest/15038359294...

And, this is speedtest.net, within a few minutes.

http://beta.speedtest.net/result/6573749352

Steam consistently gets ~380mbit (downloads up to ~43mbytes/sec), but the argument has always been that both of these two are heavily threaded.

That doesn't really bother me, as most of my downloads are from steam anyway, so, whatever.

Today I happened to download the Server 2012R2 eval iso, just in chrome, with no extensions, so will be a single source download:

http://imgur.com/PiTBGnm

This was within minutes of the TBB test above.

So it seems that just about everything else I do, even single source downloads from capable servers, can hit 380mbit, just not the TBB test....

Is this just poor peering from VM to TBB?

May be nothing to do with peering but to do with how traffic gets across the Virgin core to the links to outside world, no congestion on any point we control

http://www.speedtest.net/my-result/6574311351

http://www.speedtest.net/my-result/6574308277



Seems the same for most of us VM customers with regards to low threads . This is on business 350.

TBH i wouldn't lose too much sleep over the TBB speedtest results. I'm also getting strange TBB speedtest results on my 330/30 FTTP Fluidone connection over the BT Wholesale 21CN network - even multithread TBB speedtest is < 300 Mbps. Yet Speedtest.net gives me 310 Mbps day & night. My file downloads are a constant ~ 40 MB/s at all times. No issues at all streaming multiple 4K youtube streams on my connection and IPTV streams on my 2 x MAG 256 boxes are as smooth as a baby's bottom



http://www.speedtest.net/my-result/6574378891

Edited by deleted (Sun 27-Aug-17 18:55:57)

Looking at traffic fine leaving us and Linx but latency is rising inside the VM core network, i.e. not as bad as earlier in 2017 but maybe heading that way.

Our figures using median speeds rather than mean/top 25% tend to emphasis issues, plus if latency is playing up the ookla algorithm is much more likely to pick the server on the VM network.

If it helps whenever I test on speed test i make sure I specify the server and not a VM one.

And if the issue is down to congestion in the virgin core unless this other server is also affected you won't be affected

Now I could dumb down the speed test for Virgin Media users, i.e. remove graphs and report the top 20% speed only which will usually be within a few % of the Ookla if that would make people happier?

In reply to a post by jdigz7:

Seems the same for most of us VM customers with regards to low threads . This is on business 350.

Seems that speedtest.net uses 9 connections. Tested on 60mbit FTTC, connection count increasing over first four seconds. It could be that high speed connections get more eg. the test increases connections beyond 9 if the result is improved - I don't have a faster line to test with.

But assuming 9 is the limit it's still > 6 that TBB use.

I wonder whether any single thread tests use BBR (A tcp congestion control method) yet, that would be better than "normal" TCP CC as it's less sensitive to random packet loss.

And yet my single connection HTTP download from Microsoft hits 380mbit.

BIg ISP with many routes across and out of and into its network...

Remember when they said that the total loss of BQM was due to a 1% drop in capacity on their network...

TCP is more fun as routing may be across different links for the two directions, so ACK delays can impact things.