New DSLReports speedtester

Sigh ...

[Kushan]

Originally Posted by qasdfdsaq

Must be more to it than that, I picked "fibre" on every machine and it used 1 upstream on one machine and 24 upstreams on another.

Originally Posted by Ignitionnet

I think it does a quick test with a small amount of data and from there decides on a number of streams.

Yup, I think you're both right. A few more tests show not a lot of difference on my connection.

[Ignitionnet]

Actually I'm not sure on that one now.

I just used the DSL setting and received 8 streams each way.

Oh well, still maxed out the connection so all good.

[qasdfdsaq]

Well there's always the manual setting tab

[pip08456]

Originally Posted by qasdfdsaq

Must be more to it than that, I picked "fibre" on every machine and it used 1 upstream on one machine and 24 upstreams on another.

Thought you were on FTTC quas?

[qasdfdsaq]

Originally Posted by pip08456

Thought you were on FTTC quas?

VM D1 10/0.5 2008-2009
VM D3 50/1.5 2009-2011
BT FTTC 80/20 2011-2014
BT ADSL 11/1 2014-2015
BT FTTC 80/20 Last week till present.

[Chrysalis]

for reference

no QoS on router, using plusnet pppoe.

buffer bloat C-E it varied.

no QoS on router, using sky dhcp auth

buffer bloat usually B but sometimes C

egress QoS on router using plusnet buffer bloat usually A but sometimes B
same on sky buffer bloat always A

QoS on router for ingress and egress on both isp's the buffer bloat is either A or A+ every time.

Very good tester.

Generally for most activities downstream QoS probably isnt needed as a low amount of threads on downstream will generally increase latency by a moderate amount only, whilst on upstream even a single threaded upload can make latency sky rocket. However I observed e.g. on steam downloads using no QoS on ingress has a pretty nasty effect causing packet loss for all other traffic. Steam bombard's the connection with 20+ connections during downloading and it seems the buffer gets completely swamped, doing QoS with 2mbit topped of my download speed stops the packet loss.

Other observations.

When restricting upload speeds and reducing upload buffers generally uploading is still able to flatline at the configured limit even single threaded, however doing the same on downstream will cause the download speeds to bounce around a bit at a bit under 100% utilisation (sometimes less depending on the congestion control etc.) which some people may not find acceptable (I expect if is any isp congestion speeds would plummet as the connection with reduced buffers will be much more "polite"), however multithread downloads can still typically flatline at the configured speed (e.g. steam).

[qasdfdsaq]

Originally Posted by Chrysalis

doing the same on downstream will cause the download speeds to bounce around a bit at a bit under 100% utilisation (sometimes less depending on the congestion control etc.) which some people may not find acceptable (I expect if is any isp congestion speeds would plummet as the connection with reduced buffers will be much more "polite")

Your buffer settings would have no effect on ISP congestion because the buffering occurs within the ISP. Your window and ACK settings may have an effect though.

[Chrysalis]

the isp wont buffer if you move the bottleneck.

here is an example of ingress QoS.

[qasdfdsaq]

Why would you deliberately bottleneck yourself if the ISP is congested?

[adduxi]

Just tried this out, Wireless, Windows 8.1 64bit, IE11, no changes to Test parameters.
Granted sitting about 20 feet from Access Point and looks like it used my Infinity circuit for uploading.

[Chrysalis]

Originally Posted by qasdfdsaq

Why would you deliberately bottleneck yourself if the ISP is congested?

if the isp is congested the QoS goes out the window as then is no consistent line capacity.

If you were suggesting the line would be not worse affected under congestion, then yes maybe. But depends on the congestion I think, if the congestion hits after the buffer has grown then the larger buffers will probably give more throughput before they shrink. Thats my reckoning anyway.

In short router side QoS is pointless and a bad idea if there is frequent isp congestion.

I cannot figure out how to get the images people are using above, is that a different tester ?

[joglynne]

The two tests are similar enough for me to stick with ookla as I like the way my results are illustrated. I guess the technical information contained in the DSLReports is more useful to members who understand what it means.

So I figured out how to get the graphic to display, and ran another test.



My download speed is limited by the fact my router(s) only run 100M (Duplex) ports.

The upload is VirginSky.