Sensible discussion with TalkTalk about bufferbloat?

Hi,

Visited my parents recently and discovered their internet is suffering from truly horrendous bufferbloat, ie. latency under load. I ran a (few) tests using dslreports speedtest which reports latency under load, this revealed average download latency under load of 1.3 seconds!

http://www.dslreports.com/speedtest/31367189

A couple of graphs from thinkbroadband (where the latency peaks in yellow we can assume go much higher than 150ms!)

https://www.thinkbroadband.com/broadband/monitoring/...
https://www.thinkbroadband.com/broadband/monitoring/...

Their line is old, ratty & long, they�re lucky to get a link of 2Mbit/600Kbit, it�s never going to be higher rate (capacity) *BUT* the extra latency under load makes things appear much, much slower than they really are. It genuinely is a case of only 1 person can use the internet at one time, a simple download will slow everything else down to 'I give up' levels.

Is there any chance of speaking to someone at TalkTalk sensibly about bufferbloat?

Thanks for listening,

Kevin

All perfectly normal on ADSL/ADSL2+

With faster services you have more options e.g. operate QoS which reserves some bandwidth to allow for prioritising certain traffic

Solution is two get two lines, so if a big download is using up one they can do latency sensitive stuff on the other.

I am in a similar position with TT and getting a sensible discussion with them about anything will be a challenge.

Other than requesting your issue is raised with their CEO team and getting to speak to their UK techies, you don't have much hope I'm afraid...

The only thing I have found of any use has been getting them to agree to 'lock' my profile to 2048kbps, and then running a network level ad blocker (www.pi-hole.net) which has resulted in a 20-30% speed increase - Ookla now consistantly reporting ~1.8Mbps rather than ~1.2Mbps without

Hi Andrew,

Thanks for your response which I find disturbing more than anything else.

�All perfectly normal on ADSL/ADSL2+�. Here�s the really frustrating thing, with a sensible packet shaper in place the ISP can ensure the downlink isn�t overloaded. Andrews & Arnold make a point of running a shaper on the downlink either at or just below (95%) of the BRAS rate, thus they can achieve good latency figures even under load.

In other words it doesn�t and shouldn�t be this way. To regard it as normal is unacceptable, we should be trying harder and doing better. 1.3 seconds of additional latency under load is the equivalent of the downlink side of the exchange being on the moon!

The upstream latency is relatively well controlled by the supplied router (D-link) being around 100ms - not a perfect score but certainly an order of magnitude better than the downlink.

�With faster services you have more options e.g. operate QoS which reserves some bandwidth to allow for prioritising certain traffic�

Lower capacity links are much more in need of QoS functionality. There are some excellent shapers and active queue management algorithms out there that can maintain low latency and flow fairness. Have a look at fq_codel (and for the really brave �cake�)

For �fun� I artificially limited my own connection from 78/20 to 2/.6 with a shaper, in combination with an AQM this kept latency down to 20ms and flow fairness good. Yes there was less data flowing so downloads/uploads took more time (were slower) but general browsing was acceptable because latency was still low. Latency has much more impact on the �interactivity� of a connection than pure �speed��by which we really mean link capacity.

�Solution is two get two lines, so if a big download is using up one they can do latency sensitive stuff on the other.� There are so many practical problems with that that I�m not going to get started.

Some reading:

https://gettys.wordpress.com
http://www.edgeblog.net/2007/its-still-the-latency-s...
https://queue.acm.org/detail.cfm?id=2209336

Like I said at the start, I find �All perfectly normal on ADSL/ADSL2+� to be the most disturbing of all of this. It shouldn�t and doesn�t have to be this way, if only we could get ISPs to listen.

Cheers,

Kevin

https://www.thinkbroadband.com/speedtest does show individuals what we record for bufferbloat too and you can see the data behind this in the analysis panel

And know about queue management too, but was not aware I was talking to an advocate, was rather just giving a generic response based on someone complaining about getting a F grade result.

So yes all perfectly normal on ADSL/ADSL2+ with AAISP being the part of a small exception to these 'rules'.