What %age of your speed do you expect to get?

As Access speeds get faster (some people are lucky enough to have 200Mbit/s+ connections) I'm starting to wonder what most peoples expectations are from their ISP in terms of the %age of the speed they expect to get at peak time.

This is of course assuming the ISP manages their capacity in such a way that latency.etc doesn't go to hell in a hand basket and screw up steaming/voip/gaming.etc

Does the expectation change with regards to where you are pulling the data from? (Given that it's usually easier/cheaper to have more local capacity though IXP's than transit to far flung destinations)

Obviously if you have a slow connection anyway it's easier for the ISP to make that bandwidth available to you, but on faster lines it becomes a lot more difficult to do it whilst providing the service at a price people would actually want to pay.

I have an FTTC connection and generally manage to max it out providing the sending end is capable of sending the data that fast, but equally I have some low end servers supposedly 1Gbit/s ports that often don't get anywhere near that, but given I'm not paying much for that service I don't really expect them to either.

I think that the ISPs bandwidth is just one aspect. If you are pulling from a server with a 1Mbs connect then that is the fastest you will get irrespective of your speed.

With regard to peak usage I went with Pulse8 as they resell TTB which apparently is prioritised higher than residential traffic on the TT network. TTB seem quite fast enabling more capacity when required.

I have noticed that BT and TT are now offering a minimum speed guarantee, I haven't read the detail to see if this applies during peak times.

As connectivity evolves I think we will see companies like Netflix, Amazon and YouTube installing caches in exchanges to reduce the traffic on the backhaul. Especially if they are charged for the bandwidth they use. They already have kit installed locally with some ISPs.

All that said I would expect to see a drop of no more than 15% in my max speed during peak...

Not in telephone exchanges however, which is how your post reads. In the UK your link for Netflix shows them providing peering links at LINX and LONAP.

No, I know they are not in telephone exchanges and I was citing my opinion as opposed to fact and was in the future, not now:

"As connectivity evolves I think we will see companies like Netflix, Amazon and YouTube installing caches in exchanges to reduce the traffic on the backhaul.

What I am saying is that at some point NetFlix and co may need to consider it if/when their consumption of bandwidth increases to the point it affects everything else because providing peering links will simply not be enough or net neutrality goes out of the window.

Edited by Skilty (Fri 07-Oct-16 09:56:23)

I think as the connections get faster the %age you will get during peak time will be less, but then again if you have a Gigabit or faster connection and only get 30%-50% of that at peak time do you actually care at that point? as 300-500Mbit/s+ is still crazy fast.


As for the caches in exchanges idea, I'm not so sure especially as that really only works for certain ISPs where they have their own network, a lot of the others the traffic is encapsulated in PPP and carried over the wholesale network (Either TT or BT) where it's dumped at the ISP's LNS which will be at a handover/pop.

Now caches in the ISP core does make some sense and I think The larger ones already do this.

Ironically most of the charges are infact the backhaul from the exchange to the POP, actual internet peering/transit is the cheaper bit for most ISP's

A lot of caching is covered by Multicast in exchanges and FTTC cabinets.

In reply to a post by dragon2611:

As Access speeds get faster (some people are lucky enough to have 200Mbit/s+ connections) I'm starting to wonder what most peoples expectations are from their ISP in terms of the %age of the speed they expect to get at peak time.

I've been with Plusnet for over three years now and I expect to get 100% 24/7. Some evenings the TBB graph looks a bit shaky but most of the time it just flatlines at 64Mb/s.

At one time I'd have accepted 75% during peak hours but since rate adaptive DSL arrived all my ISPs have delivered full speed 24/7 except when there was a fault. PN were trying my patience through the first half of last year but they admitted it was a fault and got it sorted eventually.

Edited by Andrue (Fri 07-Oct-16 19:18:21)

To be pedantic, multicast is not really caching as you can't watch something and make it start to suit you.

You mean it is only for live broadcasts?

If so my post was rubbish. I thought it was a bit more wide-ranging.

Yes, just for broadcast, like tuning into a TV channel over the air. I guess there are more uses than TV, but that's all Openreach use it for AFAIK.

Yeah I was under the impression that Netflix and on demand content uses Unicast.

BT and TalkTalk use multicast. Oddly Sky use Unicast as I understand which is why there are issues with NowTV on YouView boxes.

But to answer the question Multicast suits Live TV due to the fact it is sending the same information to (x) viewers at the same time.

Which is why I was thinking of caches in the exchanges for unicast content.

I would expect to get better than the tier below.

I think there's a lot to be said for ISPs supplying speed ranges.

https://www.comhem.se/bredband

Check those guys out. They supply speed ranges and if your performance is measurably below that due to congestion they will charge you for what you are actually getting. I reckon that's the way forward assuming the average British person can handle the concept, which is debatable.

In reply to a post by Skilty:

As connectivity evolves I think we will see companies like Netflix, Amazon and YouTube installing caches in exchanges to reduce the traffic on the backhaul. Especially if they are charged for the bandwidth they use. They already have kit installed locally with some ISPs.

All that said I would expect to see a drop of no more than 15% in my max speed during peak...

I'll run with the content providers installing caches closer to end users, however given as connectivity evolves exchanges will be retired and our connections will likely end up terminating in a relative handful of POPs that's a natural thing.

Backhaul, as in the connections out of exchanges, aren't the content providers' problem, and given the costs of power and space in them it's debatable whether putting large caches in them will be viable for them.

Whilst I think I understand your argument and point, to install caches closer to the end user than the exchanges sounds unlikely to me. How would it fit in with OR and VM infrastructure and policies?

Particularly given the physical and power constraints you mentioned.

Edited by RobertoS (Sat 08-Oct-16 13:49:35)

In reply to a post by RobertoS:

Whilst I think I understand your argument and point, to install caches closer to the end user than the exchanges sounds unlikely to me. How would it fit in with OR and VM infrastructure and policies?

Particularly given the physical and power constraints you mentioned.

Depending on what you mean by exchanges, I presume you mean the datacentres where Internet exchanges live and ASes / ISPs trade traffic.

VM already host caches in regional POPs on their network. There are some in their Seacroft POP a few miles from me for instance.

TalkTalk and Sky do the same, CDNs connected to their core network nodes.

I mean telephone exchanges, as I believe Skilty does. Which I understood you to mean as those are the ones that will be phased out.

Edited by RobertoS (Sat 08-Oct-16 14:10:00)

Netflix et al are unicast

BT live IPTV is multicast and live IPTV on TalkTalk maybe, but in cases where you start shows at your whim this will be unicast.

Too many exchanges for putting caches into be finanically sensible, core network much more sensible.

Regional IX are becoming more common and these would be the logical place for more caching hardware.

In reply to a post by RobertoS:

I mean telephone exchanges, as I believe Skilty does. Which I understood you to mean as those are the ones that will be phased out.

Given the traffic won't be IP until it reaches the 'exchange' it would be quite an achievement to install CDNs that speak PON / DoCSIS.

You seem to have gone off into some weird world nothing whatsoever to do with what you said here.

At work, I have gigabit to my laptop on a fast backbone.

I expect to be able to get at least 50M regardless of congestion on the backbone, and I'd be raising a task for Networks to investigate and fix if I couldn't routinely[1] get the full gigabit to fast destinations like GitHub. I'd have the same expectations of a home provider.

[1] Routinely would boil down to accepting up to two hours in every 24 where I don't get full speed and up to 8 hours in every 168 hours, as long as the specific problem times kept changing (i.e., it was caused by unpredictable spikes). Beyond that, I'd expect them to sort things out so that my bursty usage wasn't affected by other users. Put another way - enough leeway to allow for slowdowns due to everyone watching the same thing on Netflix, not enough to let them leave it indefinitely

Not really, what I've said makes perfect sense. If you think there's no network between where the caches are now and end users I'd recommend more use of Google.

In the case of VM Seacroft is a regional and backbone POP with a bunch of hub sites connecting to it with CMTS on them. If required caches can be colocated with those CMTS. The power and space restraints on VM are being alleviated through a combination of migration to CCAP and removal of legacy equipment. A few RUs for CDN isn't a big deal.

In the case of Sky and TalkTalk caches can be moved from the backbone network where they are right now to the local POPs where the routers that terminate the exchange backhauls live.

The natural process I was referring to was that as Openreach retire exchanges they will handle the aggregation for Sky, TalkTalk, etc, so it's a natural step to put CDN nodes where those larger handover points are.

This is nothing to do with Openreach hosting anything. The caches go in the datacentres where the current provider edge routers are.

I'm certainly not referring to hosting CDNs in the access network. That's absurd.

Edited by deleted (Sat 08-Oct-16 17:36:00)