Evening congestion - you won't like this

Various people moaned when I did the httpx6 blog and said tbbx1 analysis was on the way.

Just ran it for the first time, a bug for pre 2013 data that I need to fix (we changed speed test a lot then hence I know what needs doing there) and while the analysis behaves as one would expect for Virgin Media, PlusNet has the oddity where evening speeds are showing as higher than daytime for single thread tests.

BT, Sky and TalkTalk wibble wobble a bit as I would expect, so not a maths problem.

Dare one suggest that the figure someone bandied around that congestion was affecting 1 in 10 means that unless you plot JUST those affected the majority are not affected mask it out and traffic management is helping to keep things looking good for most.

Will share more when I've fixed my bugs and looked more at data volumes and some more checks but so far looks like the HTTPx6 blog will see a repetition.

well its not I dont like it, I actually like that you ran the analysis.

Traffic management plusnet side probably has zero impact on a BT wholesale issue.

Also in regards to my own tbb tests, I generally made the effort to hop gateways when I was on a bad one, meaning most of my tests had good results.

Make no mistake tho there is a definite pattern of severe performance dropoff on plusnet for multiple users, so far of all the ones who have declared a move away from plusnet their problems were resolved by changing isp's.

Not trying to be a clever dick but if we all decide we and the rest of the family are going to stream, download and game in high def via our broadband connection then something has to give somewhere

I used to manage quite nicely on an allowance of 5Gb per month. I've been taught by others that downloading catch up stuff is convenient (and it is) and Netflix let me download a whole series of House of Cards in High Def for £6.99. If I do it and the rest of the road does it you can't wonder the broadband system is going to reach breaking point and, in some areas, some ISPs it's doing so already

I'm not sure what system Plusnet uses to speed up its connections in the evenings. Either it defies logic or someone, somewhere, must be getting a mighty slow service that isn't being recorded in the figures

That's the bit that is making me go back over the figures, will be clearer once I share the blog with the tbbx1 stuff.

Had looked at this as a comparison very quickly ages ago and found httpx6 was a reasonable model for widespread problems. Probably will lose any friends at Virgin Media once its published though.

The way issues like this affect some and not others highlights how while we could stuff software into a hardware box and ship them out if you don't have any in bad areas you won't see anything, and the volume of testing will often be noticeable by a provider who might....

and bandwidth usage will only continue to grow.

windows 10 allows full 1080p in both browser and windows app, windows 7 is 720p only via browser, lots of win7 users it seems are moving to win10.

likewise xbox360 was 720p only, and xbox one supports 1080p netflix.

however when netflix pricing plans end for legacy accounts, we may see people renewing on lower packages to keep the cost down which restricts the resolution.

In reply to a post by Chrysalis:

and bandwidth usage will only continue to grow.

however when netflix pricing plans end for legacy accounts, we may see people renewing on lower packages to keep the cost down which restricts the resolution.

Not quite sure about that. When my shiny new 4k ultra high def Sony arrived recently I gave serious thought to using the 4k option with Netflix. What's that going to do to my monthly download and will Plusnet cope?

In reply to a post by Chrysalis:

we may see people renewing on lower packages to keep the cost down which restricts the resolution.

I think more people will cancel sky and keep Netflix and Amazon streaming for the flexibility.

In reply to a post by 961a:

When my shiny new 4k ultra high def Sony arrived recently I gave serious thought to using the 4k option with Netflix.

So did I but it requires a bit of 'jiggery pokery' if you are in the UK if I understand the situation correctly. I couldn't be bothered and settled for HD which I find very acceptable. I used to wonder how on earth people managed to exceed the 40GB limit that I was on before changing to unlimited but now I understand 8^)
p.s. If congestion causes packet loss then every late afternoon and evening I see the results in my BQM

Edited by longedge (Tue 21-Jul-15 21:00:39)

When you see high pings are you using the connection? It could just be your own usage causing it if streaming rather than congestion on the ISP.

But saying that your BQM doesn't appear to me to be that different at any point in the day.

In reply to a post by ian72:

When you see high pings are you using the connection?

Not necessarily - a week or so ago I switched off every device on the network except for my iPhone and I switched wifi off on that. The latency spike showers continued throughout the night so if it's my problem it can only be my router. Day on day I see very little or no packet loss from midnight onwards and it starts again around midday and increases through the afternoon and evening. OK it's not a huge amount of red but until a couple of months ago I didn't see any at all.

Yeah I had a connection bounce sometime last night after midnight and lost my nice connection. One bounce later and it's back. Eight months on the problem is definitely still there.

http://community.plus.net/forum/index.php/topic,1408...

I think I'm nearly at the end of the my contract so might as well look around for an alternative. Trouble is I'd rather not get tied into a long term contract when at least I know how to fix the PN issue when it strikes.

yeah, for that reason pulse8 might be worth a shot as there is no tie in.

Pulse8 FTTC is MPF. Full LLU. But still no tie-in, as you say.

Edited by RobertoS (Thu 23-Jul-15 13:32:11)

http://blog.thinkbroadband.com/2015/07/another-look-...

Graphs and figures are now on the blog, not that it helps. Had to drop EE as not enough data from single thread testing, i.e. as with lots of people the faster to run web based test rather than the full flash test is preferred.

Perhaps it is one for a Plusnet Rep to answer. Their chance to sell themselves on their great evening service (or own up as to why speed tests do well in the evening).

Have you got any non port-80 analysis?

thanks for the analysis, I am baffled but it is what it is.

As that was analysis up to and including June 2015 the tbbx1 test was over tcp port 8095

i.e. this was before the single thread http tester appeared

Hi

This issue of single threaded and multi-threading testing is something I've seen just recently. My connection during peak time over the last week (around 8 to 10pm) drops from the usual ~74Mbps down to ~10Mbps on the single thread test, but usually manages to maintain > 50Mbps on the multi-thread test (10 * 6 = 60Mbps)

As I understand it, for the multi-thread test it is like me plus another 5 people, we each get around 10Mbps for this test out of the congested share so get a speed test result around 6 times better. Of course this means everyone else gets a little less bandwidth to service what looks like another 5 connections, but an extra 5 people for the duration of the test out of thousands sees a negligible reduction overall, so it works to give a higher aggregated speed.

When I'm browsing and downloading files however, I'm a single user, so see the reduced speed.

Annoyingly the BT Wholesale speed checker for reporting speed issues only performs a multi thread download, so gives the higher reading, and comes back that the line is performing within good parameters, although I don't consider downloading at 10Mbps good for 80/20 line.

Would it be possible to chart the difference between the two tests taken at the same time for each ISP. If congestion is a problem, this should see an increasing difference between the two tests during peak times, if charted with y axis as the difference and x axis as time, would should a trend upwards during peak times.

Browsing is not strictly a single thread, browsers will fire off multiple threads downloading the various components for a web page.

The numbers are in the blog for what you propose so someone can do the maths if they want.

I really think the issue for identifying the issues with plusnet is it is not affecting everyone, where for example congestion is pretty obvious on Virgin Media either because it affects more people or the impact on people across the board is a lot higher.

Tempted to try and add the AAISP, Zen, IDNet and other smaller WBC providers together in a 'super ISP' and see if the pattern repeats. If so its something in the BT Wholesale network most likely.

Another option is splitting down into the regions, but every time you split the data you make it more prone to noise.

I got an idea but I expect it may be too complex of a request.

Is it possible to only use users who have plotted speedtest's across off peak and peak?

My theory for the higher peak time results is simple, people with higher sync speeds are testing more often during peak. Also notice the average result is pretty low and people with lower sync speeds wont feel it as much.

Possible but unsure sample size will be enough

You might also benefit from getting info on typical user habits, such as streaming or P2P. Perhaps a simple poll. I'll wager that over time Plusnet have discouraged the heavy users elsewhere.

Or maybe their traffic management systems really are that good...then why isn't BT using them?

A simple poll has many flaws in that respect as it will draw in people from other providers, and even if you say vote if only a PlusNet user others will - experience tells me that.

Then how do we know that poll of people is representative of the PlusNet user base.

Of course everything is possible, just a case of effort versus payback and having lots of other things that need doing that will help a wider body of people.

Having thought about this for a bit I've just come on here to make exactly the same suggestion. Unless you do that the results are absolutely useless at telling if there is a peak time slowdown as you've no way of knowing if it is a valid comparison.

At a guess, I'd suggest that many of the people who are prepared to pay for for a higher speed connection are working and hence would only be at home running speed tests during peak time.

Explain Virgin Media result then?

I'm not familiar with their speeds or pricing so I couldn't comment.

If you'd said there was no noticeable difference between peak and off-peak I'd have believed it. But, when the Plusnet graphs (before Plusnet pulled them) always showed far higher traffic levels during the evenings, higher peak speeds only make sense if the people running the tests during peak times have higher speeds to start with than the people who typically run tests during off-peak.

It may be the mix of customers is different on Plusnet to Virgin.

Edited by jelv (Sat 25-Jul-15 21:48:16)

Another point. If users were experiencing slower speeds during the day than during the evening don't you think we'd have seen posts to that effect?

perhaps VM has enough congestion that it overides or is some other reason.

But I very much doubt the higher peak results are anything to do with what plusnet or BTw are doing, thats clearly down to who is doing the tests at specific times of the day.

For some reason tho you not keen on showing results for a small subset of users.

So for this reason to give credibility to time of day results you should only be showing results for users who are across all the time periods you want to analyse.

Give me 200 PlusNet users willing to leave a PC on that will run 20 seconds of testing every 30 minutes so long as browser window open.

Only did the original plots because people kept saying the speed test results would show the problem.