Print

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

[Scottie]

I'm helping a rural community consider the feasibility of a locally-built broadband infrastructure.  Scottish Highlands, so pretty low density.  While local distribution of broadband looks practical and affordable, any kind of uplink to (say) Linx, in Edinburgh (~150 miles), is looking really expensive, the more so because those costs for the aggregate signal will have to be shared between a relatively small number of users.  I wondered whether we're looking at things the right way?

The community envisaged using its own ISP providing a basic connectivity service.  From a traffic point of view, a router interfacing with Linx and, downstream from it, a few level 2 switches dispersed in the community to aggregate circuits from clusters of premises.  Basing the router in the community, we'd seemed to need a connection from it to Linx at Edinburgh.  We weren't sure what uplink service would be appropriate, or from whom.

In the Highlands, Openreach may be the only practical provider and, looking at their website, are we correct to think of using their Ethernet Backhaul Direct offering?  OR has recently discounted these prices, public release here:

https://www.openreach.co.uk/orpg/home/updates/briefings/ethernetservicesbriefings/ethernetservicesbriefingsarticles/eth04816.do

Using these prices, rental (only) of 1 Gbps to Linx from 150 miles away (240 kms) would seem to cost:

EBD  £4,098  (is this per annum?)
EAD main link( £0.276 x 240,000) £ 66.240
If these are the correct components from the price list, the uplink to Linx would be around £70k pa, per Gbps.  At concurrent peak traffic of, for example, 25 Mbps per customer circuit (independent streaming to 2 HD devices per circuit, plus other low-level traffic), that £70k could only support 40 users and would account for £1,750 of their annual fees, £146 per month - just for uplink!

Clearly, that isn't going to fly.

Are we looking at things the right way?  Router in the community, and dedicated GbE connection directly to Linx?  We're not sure exactly which OR services could be used -  are EAD and EBD what would be needed in this scheme?

Thanks for any comments.  Really, any comments.  I volunteered to help with this and I'm really sad at what it seems to be costing.  What have we missed?

Scottie

[S.Stephenson]

Pretty sure the BT cabinets that offer 80/20 to 200+ customers only have a 1gbit link.

How spread out is the community because BT has done deals before to roll out FTTC.
http://www.communityfibre.bt.com

[Scottie]

Pretty sure the BT cabinets that offer 80/20 to 200+ customers only have a 1gbit link.

Interesting, S.  So those 200 + customers will never concurrently receive 80 Mbps.  We were hoping to avoid that situation, by ensuring sufficient uplink capacity to the IX.

I saw this report, yesterday, (it may have been linked from Kitz' site):

http://stakeholders.ofcom.org.uk/binaries/research/infrastructure/2014/next-gen.pdf

and it mentions two things of interest.

First, B4RN seems to use dark fibre (from an entrant provider, not from BT) between itself and Linx Manchester.  Lighting it itself with DWDM, it can deploy as much capacity as it needs.  And B4RN's FAQ page mentions that their network is 1Gbps symmetric, and uncongested, even if everybody streams at the same time - because they have the capacity on their uplink.  That's what we wanted to do, too.

Secondly, the report cites several NGA entrants who mention that Openreach's EAD circuit is prohibitively expensive.  That's what we're finding in our cost review.

You are probably aware that Ofcom recently required BT to make a Dark Fibre service available, and some provisional information has been released about that:

https://www.openreach.co.uk/orpg/home/products/darkfibreaccess/darkfibreaccess.do

In one of those documents, Openreach mention that their price for the Dark Fibre (DFA) service has been derived using the formula agreed with Ofcom; though the price is similar to the EAD/EBD price for 1 Gbps circuit.  DFA would be much better value for an AltNet, though, because much, much more than 1 Gbps could be transmitted through it.

Thanks for the note about the customers, and uplink capacity, on BT's 80 Mbps service.  I might explore some calculations of average vs peak demands that we expect. 

Grateful, Scottie

[Weaver]

Would it be worth contacting B4RN directly and asking them if they would mind letting you pick their brains?

[gt94sss2]

Interesting, S.  So those 200 + customers will never concurrently receive 80 Mbps.  We were hoping to avoid that situation, by ensuring sufficient uplink capacity to the IX.

I assume you have checked this area won't be under a BDUK scheme or that residents wouldn't prefer to wait a few years for the 10mb USO.

BT FTTC cabinets have 4 fibres, of which 1 is connected by default. As you have gathered this means your plans for bandwidth capacity will be excessive given the benefits of contention (for an ISP).

[tickmike]

Re 'EBD  £4,098  (is this per annum?)' I think it's per month !.

you need to do some searches for any dark fibre near you.

What area do you live. ?

[Weaver]

In the Highlands, the issue of maintenance is also a killer. Some of my neighbours have complained about a local wireless ISP having extended downtime after serious weather events, so says Mrs Weaver. So will need to budget for contractors to do reactive timely maintenance on a retainer unless there are a lot of both skilled, always-available and very dedicated locals about.

[aesmith]

In general any Internet access network will be over subscribed.  We were involved in a project for a village in Aberdeenshire, using wireless distribution.  They took a commercial Internet connection as their main link.   As of my last involvement last summer, they had 55 subscribers.  Each subscriber, depending in tariff, had a nominal data rate of 5, 7 or 30meg download.   Totting that all up it was over subscribed by around 6:1 being supplied by a 100meg main Internet connection.

By the way, in your Openreach costs you need to factor in whether there will or will not be Excess Construction Charges for your end of the tail.  These can be quite substantial.

Also, if you're talking about taking an Ethernet point to point as your backhaul, you need to factor in actual IP connectivity costs at Linx, and peering costs.   Will you be providing authentication, dns, address allocation?

[niemand]

Interesting, S.  So those 200 + customers will never concurrently receive 80 Mbps.  We were hoping to avoid that situation, by ensuring sufficient uplink capacity to the IX.

No need for that and it would be crazy to do so.

Trying to deliver enough uplink to allow every customer to max out at the same time would leave tons of capacity unused and leave you guys spending a bunch of money unnecessarily.

Don't aim for no contention, aim for no visible contention. Whether you are contending at 5:1 or 500:1 as long as the link is never full the customer experience will be that of a 1:1 link.

Average punters are using a couple of megabits a second at peak, rising at 40-60% per year. Scale accordingly and you should be fine.

[Weaver]

@ignitionnet wrote:
> Average punters are using a couple of megabits a second at peak, rising at 40-60% per year.

Won't TCP simply max out each pipe? When you say “a couple of megabits a second”, do you mean taking the average over all users over some time window of duration x, where x is of the order of minutes?

[niemand]

That's the peak across entire customer bases.

[aesmith]

It's perfectly normal to have that sort of oversubscription.  It's based on the assumption that peak loads from individual subscribers is unlikely to coincide (similar to "diversity" as applied to electrics).   The larger the number of subscribers the lower the chance of their loads coinciding.   

Just did a rough calc on my home connection.  4M synch speed and 3.5M profile equates to just over 400K bytes per second.  Flat out that would be 995 Gig in a month.  In reality my usage is around 50-60 per month, even if I was hitting the A&A 100gig cap that would still be an average rate of only 300K.   (Unless a decimal point has slipped somewhere in my calculation)

That could be refined by considering only peak time traffic, let's say 100% of my load came in the four busiest hours, that makes the average 1.8M.   We do something similar for voice loads, considering the traffic in terms of call attempts during the "busy hour" to calculate chances of trunk oversubscription.