[AusNOG] Contention, congestion, and link capacity planning

John Edwards jaedwards at gmail.com
Tue Sep 19 16:17:50 EST 2017


If anyone is really keen on this topic, there is a research group in
Australia with comms traffic as its focus:

http://www.trc.adelaide.edu.au/trc/

Once upon a time they had a published model for contention that mapped well
to real world dialup service data.

John


On 19 September 2017 at 09:16, paul+ausnog at oxygennetworks.com.au <
paul+ausnog at oxygennetworks.com.au> wrote:

> Thanks for all of the on and off list responses, I appreciate everybody’s
> views and advice.
>
>
>
> Unfortunately my picture may not have been painted correctly, but the
> information I received was greatly appreciated.
>
>
>
> The work I am doing is around dimensioning and modelling of backhaul links
> for sites with a varying amount of business users.
>
> The backhaul then goes to a DC for transit and peering access etc.
>
> Backhauls can be wireless or fibre depending on location and
> infrastructure available, and costs of course, but I am ultimately trying
> to get a dynamic equation that can be used to put figures into and gauge
> the costs at the end.
>
>
>
> The general consensus from the information I have received is that a
> contention ratio of around 5:1 is pretty decent on business grade
> backhauls, obviously the less users you have the less that should be, so
> working from 1:1 for a single user site and then ramping up to 5:1 by the
> time you get 10-15 users at the site seems to be pretty acceptable.
>
>
>
> The capacity per user as Ahad has suggested below, and others, works OK
> once you have the critical mass of users, when working out startup costs of
> a new site and capacity requirements unfortunately these calculations don’t
> work, it’s in these situations where I was hoping a silver bullet equation
> might be around which could help adjust costs and backhaul requirements
> based on user counts as they ramp up from 1 onwards, or even 5 onwards.
>
> I think now I will just use some basic calculations and create more
> scenarios rather than have something more dynamic, it’s no issue but I was
> just hoping I could do it an easier way.
>
>
>
> From our historical information and current link capacities we run under
> that 5:1 figure but pretty close, so using it as a figure for projections
> should accurately reflect our required bandwidth requirements as well as
> our current bandwidth allocations which are working comfortably well.
>
>
>
> Thanks again
>
>
>
> Regards
>
> Paul
>
>
>
> *From:* Ahad Aboss [mailto:ahad at swiftelnetworks.com]
> *Sent:* Monday, 18 September 2017 5:40 PM
> *To:* paul+ausnog at oxygennetworks.com.au
> *Cc:* ausnog at lists.ausnog.net
> *Subject:* Re: [AusNOG] Contention, congestion, and link capacity planning
>
>
>
> Hi Paul,
>
> When it comes to backhaul capacity planning for business grade customers,
> there is no one size fit all formula. It all comes down to the type of
> customers you have and the frequency of internet usage during business hrs
> and after hrs.
>
> Generally, the peak usage for business customer is between 8am – 6pm, MON
> - FRI. You’ll need to have enough backhaul capacity at the head-end to
> cater for occasional burst during peak hrs though not all customers
> download at full speed all at once.This is just for safety measures.
>
> For Ethernet or midband Ethernet services through Optus, TPG/AAPT or
> Telstra, you could get away with 3:1 contention on the backhaul but I
> strongly recommend that you don’t risk this contention ratio if you have
> less than 100 customers per state.
>
> Let’s say you have 100 customers, a combination of 50 x 10Mbps and 50 x
> 20Mbps, you can safely use 1Gbps backhaul per state through a single
> provider (AAPT/Telstra or Optus) and as you add more customers, you can
> closely watch the average usage across the customer base and increase the
> bandwidth as required.
>
> Just to be clear, the 3:1 contention is for best effort Ethernet services
> ONLY which is mostly used by SMEs, if you are providing guaranteed
> bandwidth 1:1, you will have to honour the contention all the way to your
> POP and internet.
>
> If you are building up your customer base slowly, be prepared for very
> slim or no margins at all as you still need to pay for the access links to
> customers, trunk port (head end or backhaul), IP transit, rack space, power
> and cross connect fees.
>
> In these circumstances, it’s best to resell these services through a
> reliable ISP until your Ethernet customer base is sizeable to justify the
> head end built.
>
> For residential grade broadband factor in 50% traffic growth every year, a
> blessing that all ISPs have to deal with while trying to maintain their
> profit margin. :)
>
> Based on industry contacts, the current average usage per SIO (IN AUS) for
> NBN and DSL are as follows;
>
> NBN: 1.3Mbps
>
> xDSL: 850Kbps
>
> Netflix and HD/UHD video streaming is changing the peak average rapidly.
>
> I hope this information helps.
>
> Cheers,
>
> Ahad
>
>
>
> On Mon, Sep 18, 2017 at 12:41 PM, paul+ausnog at oxygennetworks.com.au <
> paul+ausnog at oxygennetworks.com.au> wrote:
>
> Hi All, I was hoping to gain some thoughts from the list around contention
> and backhaul link capacity planning.
>
>
>
> We are working on some new site plans and have plenty of existing sites to
> draw usage statistics from when it comes to capacity planning, typically
> all of our backhaul links are running pretty low contention as all of our
> customers are business customers, but I am wondering if anybody has any
> formulas they have used successfully in the past.
>
>
>
> Being that we only provide business Ethernet connections planning is
> usually pretty straight forward, but in modelling some expansion plans I
> want to try and actually wrap something around the planning process for
> backhaul capacity.
>
>
>
> For example, 1 x 50M customer will clearly need 50M of backhaul from the
> POP they connect to, but what about 2, or 4, or 10 ?
>
> You could easily surmise that 2 x 50M customers don’t need 100M of
> backhaul unless they are very heavy users, so let’s say they may need 75M,
> but this requirement for backhaul is realistically a sliding scale as the
> customers and bandwidth requirements grow the backhaul is not necessarily
> going to need to grow at the same rate.
>
>
>
> I have worked this stuff out for some time now manually and had good
> results, our customers are happy, but I was hoping there would be some sort
> of calculation or formula that I could apply to some modelling figures
> which would give me a pretty close indication of requirements.
>
>
>
> Thanks in advance
>
>
>
> Regards
>
> Paul
>
>
> _______________________________________________
> AusNOG mailing list
> AusNOG at lists.ausnog.net
> http://lists.ausnog.net/mailman/listinfo/ausnog
>
>
>
> _______________________________________________
> AusNOG mailing list
> AusNOG at lists.ausnog.net
> http://lists.ausnog.net/mailman/listinfo/ausnog
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ausnog.net/pipermail/ausnog/attachments/20170919/1980ac44/attachment-0001.html>


More information about the AusNOG mailing list