Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Alex Atkin UK]

I thought some might be curious to see the history of my line and how little a change it takes to lose ~11Mbit of downstream speed.

I believe this is due to crosstalk (the QLN is quite telling), but certainly curious if anyone has another take on it.

Green is when I was on Digital Region syncing at 100/36, Blue just three years ago when I was still syncing at 80/20 on Zen and Red is how much its fallen since then down to 69/20.



I realise some people have it much worse, but it really sucks to have your speed decrease after experiencing such excellent speeds in 2013, with no hope of Virgin and very little of g.FAST/Fibre any time soon.  Plus its an ECI cabinet so its being hobbled by crap hardware.

[Chrysalis]

yeah when I took my very first graphs, I had many tones in the D1 section at max 15 bit tones. Now not a single one, even on a 50db loop loss adsl1 line I had some tones at 15bits per tone.

On my current sync I have lost about 40% of my signal since that day (although am now on diff pair, but original pair got as low as under 50mbit/sec a whopping 60% drop).

My D3 hasnt weakened much, its in D1 its took a real hammering and a modest hit in D2 as well.

[kitz]

I've lost well over 40Mbps from my headline speed purely down to cross-talk

[burakkucat]

A non-provocative question: Other than noting the numbers (because you can) does the degradation actually affect your usage of the service provided on your circuit? 

[Ronski]

In my case yes, when I first got VDSL I had an upload of around 12Mbps, when I changed to VM some 6 years later my upload was around 6Mbps and would fluctuate up and down over time. So my cloud backups would take much longer, other than that not really.

[Weaver]

@Burakkucat I suppose that one scenario might be where there are multiple users in a household, and someone finds that certain concurrent activities are no longer possible, an example might be several people watching streaming TV at once.

Low speed means that activities have to be coordinated. I have found that an upload that is a backup to the Apple iCloud fails sometimes if one or more downloads are in progress at the same time. That is very surprising to me, given that the downloads should not be putting too much load on the upstream direction, since all that is needed is the transmission of ACKs upstream.

I don’t do big downloads from Netflix when my wife is trying to browse the web because it slows her down too much and makes things unbearable for her. Wish I could fix that. But I don’t have proper QoS (grrr) and in any case, I don’t know whether or not the traffic belonging to Netflix or Amazon download is QoS-marked, and if not then I don’t know of a way to distinguish it and therefore be able to mark it. I really wish that I could de-prioritise these kinds of bulk download activities and prioritise manual web browsing and especially DNS.

[Chrysalis]

burakkucat no, I am lucky that the starting signal strength was enough for a 105mbit sync speed and I am still able to get a mid 70s sync speed now, so most of the loss above the openreach speed cap.  But other people with similar % drops on much lower sync speeds wont be so lucky.  Truthfully I could probably drop to a 55/10 or 40/10 service and not have my usage pattern really affected much.

Weaver you can possible identify netflix traffic via ASN, but if aaisp run local netflix caching servers those ip's wont be in netflix ASN range.

I thought aaisp already prioritised small packets? which should help you, otherwise just shoving everything through a generic fq_codel queue may even work, but remember to make the pipe limit lower than the actual connection throughput capability.  The downside of QoS you lose some headline speed for it to work at its best.

[Ronski]

I really wish that I could de-prioritise these kinds of bulk download activities and prioritise manual web browsing and especially DNS.

Can you not just prioritise traffic to and from her devices?

[niemand]

I don’t do big downloads from Netflix when my wife is trying to browse the web because it slows her down too much and makes things unbearable for her. Wish I could fix that. But I don’t have proper QoS (grrr) and in any case, I don’t know whether or not the traffic belonging to Netflix or Amazon download is QoS-marked, and if not then I don’t know of a way to distinguish it and therefore be able to mark it. I really wish that I could de-prioritise these kinds of bulk download activities and prioritise manual web browsing and especially DNS.

Do you not have a way of prioritising the first few MB of a transfer via a burst parameter? No need to mark anything that way and naturally prefers browsing, etc.

Obviously stuffed for DNS but would at least help with shaping TCP - can you not even just set a minimum bandwidth of a few kilobits per second dedicated to port 53 and have the kit shape everything else to keep those few kilobits free?

To answer your other question, no, traffic from Netflix and Amazon isn't marked by those guys. To properly QoS you would need something similar to my work set up, where traffic goes to a DC and is shaped outbound from there to my home office. It also identifies millions of different applications and websites so I can shape by domain, application, protocol, etc, etc.

[Chrysalis]

Thats very interesting Carl, So you use a VPN or something to route all your traffic and then shape on outbound from the VPN?

[j0hn]

Thats very interesting Carl, So you use a VPN or something to route all your traffic and then shape on outbound from the VPN?

https://forum.kitz.co.uk/index.php/topic,22482.0.html

[Weaver]

> Do you not have a way of prioritising the first few MB of a transfer via a burst parameter? No need to mark anything that way and naturally prefers browsing, etc.

That’s an interesting idea. Would a system like that reset into state 1 after either a short idle timeout or a period of low traffic rate, and then a period of sustained high traffic puts it into state 2, something like that ?

I don’t have the tools to prioritise Mrs Weaver. I can rate-limit certain flows by MAC address or by IP traffic type. I use the rate limiting system on ‘guests’ who are defined as anyone who is not a member of a whitelist, defined by MAC addresses. This is a pretty rubbish way of doing it though, because it limits users even when there is no need because no ‘important’ users or flows need the bandwidth.

[niemand]

> Do you not have a way of prioritising the first few MB of a transfer via a burst parameter? No need to mark anything that way and naturally prefers browsing, etc.

That’s an interesting idea. Would a system like that reset into state 1 after either a short idle timeout or a period of low traffic rate, and then a period of sustained high traffic puts it into state 2, something like that ?

Couple of ways to do it. Either prioritise the first few MB of a flow, a five-tuple of source and destination IP and port alongside IP protocol, regardless of how long-lived the flow is, or have a timer that resets after traffic drops below some threshold for a time.

This has been done on entire networks. Some broadband connection types running over shared media have max burst parameters that do not rate limit the first x MB of a transfer at all, they allow that end user device to use all the available bandwidth for a period, and others monitor utilisation and modify priority of end user stations depending on their usage of the network.

It's not difficult to implement in software and is very simple indeed if the software is already monitoring per flow and has shaping modules in place already - when the burst criteria have been exceeded force that flow back through the classification engine on a lower priority, likely its original one rather than the expedited burst one.

[niemand]

Thats very interesting Carl, So you use a VPN or something to route all your traffic and then shape on outbound from the VPN?

Absolutely not all of it, just the stuff to and from the home office, and even then only a subset as I choose to.

Could certainly push all traffic in the entire network through it but there's no real point just yet. Doing that would require me to either suck up the electricity bill of a server or buy a new one, alongside paying AWS bandwidth and CPU charges.

It actually functions as an offload and overflow for the most part - I compensate for the lack of upload capacity on VMB by overflowing traffic onto it once I reach a certain level. My office is permitted 15 Mb/s of the VMB line before traffic starts to waterfall onto LTE.

Using a VPN for everything is a bit silly.

[Weaver]

I wish the Firebrick had real QoS, observance of existing marking and rules to categorise stuff and / or apply marking. And state machines that you could set up.

It has these things called ‘profiles’ which are bool states that can be set by various triggers, but traffic levels have not been included.