[AusNOG] Why not Symmetric ingress and egress?

Sun Jun 20 11:37:51 EST 2010

On Thu, 17 Jun 2010 20:54:33 +0930
John Lindsay <JLindsay at internode.com.au> wrote:

> You know I don't actually follow your argument.
> 

The issue I'm highlighting is that 

(a) to avoid congestion (if that is your goal) you have to provision
link capacity based on peak traffic load.

(b) other than at the edge of the network, the link technologies used
in the network are most commonly symmetric.

(c) the cost of the link, the technologies on the ends of it, or the
costs of the technologies to deliver congestion free traffic to or from
it, are usually proportional to the peak traffic load on the link.

(d) typically traffic peak occurs in only one of the directions on a
symmetric link, so you have to size the symmetric link based on the
peak load in that single direction. By definition, you get equivalent
capacity in the opposite direction on the symmetric link.

(d) asymmetric link technologies commonly allow or encourage peak
traffic loads in one direction to be significantly higher than in the
other direction, with ADSL being a good example.

(e) if you're using symmetric links to aggregate traffic from
asymmetric links, your symmetric links will have to be sized based on
the traffic load that the aggregated asymmetric links can generate.

f) because symmetric bandwidth is obviously symmetric, your
downstream asymmetric links will never be able generate traffic that
utilises the symmetric link bandwidth in one of the symmetric link
directions.

(g) if you're not using a downstream technology that forces asymmetry
(unlike cable or ADSL), and can convince people to trade download for
upload capacity, e.g. instead of 100/8, be happy with 50/50 in the FTTH
case, you're lowering the capacity required on your symmetric backhaul
links, which will result in reduced costs. In some cases that cost
saving may not be on the symmetric link itself (i.e. reducing 4Gbps peak
to 2Gbps peak on a 10GigE won't reduce the cost of the 10Gbps circuit),
but it will occur somewhere else in the network where you buy
(symmetric) bandwidth to service an asymmetric load being created at
the edge.

> Internode's customers generally want the fastest possible downstream they can get and want to be able to download as fast as possible.

That's why I was initially surprised at that unprompted Whirlpool
survey. My expectations of people on Whirlpool is that they're the
very vocal minority, and most of the time don't represent the
views of the majority of ADSL customers (when you read something
there, take off your techo hat, and think whether your sister, brother,
parents, punter in the street would say the same thing). The fact that
they spend time on Whirlpool, rather than going to the football, playing
golf, watching TV etc., indicates that they're both more interested in
the "Internet/Broadband" as a topic, and are also more technically
informed. Currently, out of 579 votes, 378 (65.3%) would prefer 50/50
over 100/8. So even the "educated and selective" consumers of broadband
would prefer symmetry. I doubt the majority of ADSL customers know or
care, as long as the experience they get from ADSL is good enough, and
better than what they got with their ADSL modem.

Which customers have you surveyed? Have Internode done a (e)mail out
survey to it's "normal" (not Whirlpool :-)) customers to find out what
their service expectations are, whether they're all being met, or
whether they're being disproportionally met (i.e. undershot or
overshot)?

>Some experimenting has shown that the limiting factors are
>  backchannel bandwidth which needs to be in the ratio of 100/7 or
<  better (as I recall)

That's the TCP effect described in that RFC.

> and CPE which needs to not be < $100 junk.  The
>  Apple Airport Extreme / Timecapsule and Cisco boxes have proven to be
>  excellent for FTTH whereas everything else, pretty much, has proven
>  to crap out above 10 ~ 20 Mbps.
>

That seems to suggest that 100Mbps was made available as a product
before the market was ready for it, as if the market was ready for it
then there would be CPE far more commonly available at a market
acceptable price that can provide those 100Mbps speeds. 

There's this psychological trait of humans called "perceptual contrast".
It means we judge things by what has, usually recently, come beforehand.
For example, a 5KG weight will feel light if we've just held a 10KG
weight, yet if we hold the 5KG weight first, then relative to the
things we normally hold in daily life, the 5KG weigh will feel heavy.
The same thing occurs with price. If we buy a $2000 suit first, then we
perceive a $300 jumper to be relatively cheap. Flip it around and the
$300 jumper sets our reference point, and relative to most clothes
that's an expensive jumper, therefore making our perception of the suit
to be very expensive. 

I believe that rather than offering 100Mbps plans over FTTH, as long as
ISPs ensured that there was a noticeable contrast between ADSL
and their FTTH plans, they could have offered far lower. For example,
*guaranteed* 30/30Mbps. I think customers would have been quite happy
with that, as they know that peak ADSL2+ speeds aren't assured (well,
those that care about them do) or would have experienced better peak
performance than their prior ADSL experience, and the symmetry would
have also been a selling point as to why FTTH is better than ADSL. That
would have avoided the CPE issue, and would have also increased the
likelyhood that people would see their 30Mbps being fully utilised as
they used the Internet.

Here's an interesting advert for Verizon Fios (i.e. FTTH). It's
interesting to see one of their selling points as to why it is better
than cable.

http://www.youtube.com/watch?v=MiHsxQJ9ZOo

> Our experience, and we have hundreds of FTTH customers on other networks, is that the vast majority of customers buy 25Mbps access because it's cheap.  If they do buy a faster connection and don't buy a good router they complain about not achieving line speed.
> 

Exactly. I'm starting to be convinced that the only thing all us humans
have in common is the desire to get value for money. Sure, if you're on
100Mbps you might be able to impress your friends by downloading off of
your ISPs file mirror or speedtest at that speed, but for the rest of
the time, when accessing the rest of the Internet, you're probably
consciously or unconsciously saying to yourself, "I didn't really need
this, and could be spending the money on something else".

> Our experience, particularly with a couple of hundred thousand ADSL subs is that asymmetric access results in fairly balanced in/out traffic at the head end presumably because quite a lot of the back channel is P2P and it will use whatever it can find.
> 

I'd suggest, as Sam Silvester said, that is because a lot of Internode's
plans don't count upstream. I think that you're right, it probably is
P2P upstream, and also would be a relatively small percentage of your
customers who're actively trying to maximise their use of the unmetered
upstream. By not counting upstream traffic, Internode probably have a
disproportionately larger number of "P2P upstreamers" than other ISPs.

I understand that in general P2P users are rewarded with increased
downstream performance by the how much they contribute upstream. It'd
be interesting to know if Internode's customers who're choosing
Internode because of the unmetered upstream are achieving their
fundamental goal of increased downstream. IOW, are they able to fill
their downstream capacity, because they're able to fill their upstream
without any consequence. If they're not, or aren't able to
very often, then those customers would also probably be likely to trade
some downstream capacity for increased upstream.

> Generally in Internode's network we send other networks more traffic than they send us because we are a large content source.  We don't need any more outbound traffic although we do have a little more headroom on the international links.
> 

That certainly would help mitigate some of the disparity caused by an
asymmetric edge, however it would only be helping get better efficiency
out of your transit and peering links. To help mitigate the effect on
your backhaul to your ADSL subscribers, you'd need to put those content
sources at the ends of those backhaul links i.e. the exchanges. That's
probably a bit impractical.

> What's your experience Mark?
> 

I've worked with a few ISPs, with at least one having it's own ADSL
infrastructure. 

Regards,
Mark

> jsl
> --
> John Lindsay - GM Regulatory and Corporate Affairs - Internode and Agile
> Street: 150 Grenfell St, Adelaide
> Postal: P.O. Box 284, Rundle Mall, Adelaide, S.A. 5000
> Direct: ph +61 8 8228 2965 - fx +61 8 8235 6965 mb/cell +61 4 0357 7711
> 
> On 17/06/2010, at 7:39 PM, Mark Smith wrote:
> 
> > On Thu, 17 Jun 2010 09:29:45 +0930
> > John Lindsay <JLindsay at internode.com.au> wrote:
> > 
> >> It's mostly a way of differentiating business services from domestic services.
> >> 
> >> Access seekers will be paying by the megabit up and down.
> >> 
> >> When I say differentiating what I mean of course is "enabling charging more for".
> >> 
> >> Based on ISP experience asymmetric service keeps traffic in better balance.
> >> 
> >> There are also some limits in the FTTH network architecture that mean there is less total bandwidth in the back channel.
> >> 
> >> It is easy to prioritise the traffic from one head end to hundreds of end users.  It is much harder to control those end points when they each want to transmit over one shared path.  The network uses much the same protocol as WiFi for controlling this.
> >> 
> > 
> > I think it would actually be more economic if symmetric was used.
> > 
> > Lets say you have 100 customers on 100 Down/8 Mbps Up. If, at your
> > peak traffic time, 50% of them were fully utilising their 100Mbps down,
> > to avoid congestion, your backhaul will need to be 5 Gbps towards
> > the customers. As you end up buy symmetric backhaul, you'll also be
> > buying 5Gbps upstream bandwidth. Even with all 100 customers utilising
> > their 8Mbps up, you'll only ever utilise 800Mbps of the upstream 5Gbps.
> > What a waste of 4.2 Gbps upstream backhaul!
> > 
> > If, OTOH, you can convince those customers to accept 50 Mbps/50 Mbps,
> > you only have to buy 2.5 Gbps symmetric bandwidth for backhaul, and will
> > possibly be able to fully utilise the upstream backhaul you've paid
> > for. So you're buying less backhaul bandwidth in total, and gaining
> > better efficiency out of what you've bought. This efficiency carries
> > right through your network to your expensive transit links.
> > 
> > How hard would it be to convince people to accept 50/50? It seems 66%
> > of the (debatably) technically savvy Whirlpooleans would be fine
> > with it -
> > 
> > http://forums.whirlpool.net.au/forum-replies.cfm?t=1462456 
> > 
> > It'd also avoid the performance problems that TCP has with asymmetry -
> > 
> > RFC3449 - "TCP Performance Implications of Network Path Asymmetry"
> > 
> > http://tools.ietf.org/html/rfc3449
> > 
> > My guess is that ADSL was chosen as the broadband technology to use,
> > rather than a symmetrical DSL technology (not in Australia, I'm
> > talking by the broadband groups who standardise it i.e. the Annex M
> > people), because it sounded right for the way people were using the
> > Internet at the time (consumers rather than producers), rather than
> > understanding that the Internet protocols have operated over symmetric
> > links for most of their life and it is therefore an unstated design
> > assumption. If that wasn't the case, I don't think the above RFC would
> > exist and be a Best Current Practice RFC.
> > 
> > Regards,
> > Mark.
> > 
> > _______________________________________________
> > 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