On 05/09/07, paul at clubi.ie <paul at clubi.ie> wrote:
> On Tue, 4 Sep 2007, Thomas Bridge wrote:
> > Typically the issue with high bandwidth consumers and cost in a
> > broadband environment is the impact on the circuits interconnecting
> > the Telco
> Right... So you're acknowledging the premise - external links are
> suffering from contention, even some internal links you say. (And P2P
> is /the/ major user of bandwidth, by most accounts).
Contention is such a bad word really - of course theres "contention".
What I'm saying is that for ISPs - at least in Ireland - typically
the biggest cost in supplying DSL is going to be the STM-1 circuit
connecting you to the telco's network to receive the L2TP traffic, and
not the cost of the international network.
> > The cost of transit is not the most significant cost. And given
> > current transit prices,
> If I used 'transit' loosely, apologies. I, and the OP (I think),
> meant external.
You mean traffic leaving the ISP to go to other ISPs, be they Irish or
That's exactly what I meant by transit.
> > Yes I have - its irrelevant.
> Yet you acknowledge load on external links is becoming a problem?
No I don't - you're either not understanding or I'm not making clear -
there typically isn't an issue buying the international transit
required to support a service.
> > You've had a bee in your bonnet for some years about the fact that
> > in some situations Eircom took your PPP frame and wrapped it up in
> > an IP header to give to your ISP,
> Yep. It's far from ideal.
You never had a problem when the packet was being wrapped up in a nice
little ATM header all the way (or being ATM, split into about 40
different cells each with their own nice little ATM header). It
makes very little difference and is actually easier for the ISP and
the telco to manage the L2TP way.
> > Unless you're suggesting that all ISPs should run IP to the DSLAM -
> > but that's horrendously cost prohibitive compared to the wholesale
> > solutions offered by the Telcos.
> I'm not a business person,
Must ... resist... temptation.
Nope - can't do it.
My response is that "i'd never have guessed".
> but the current situation is technically
> /awful/. If you think otherwise, please agree below to the 'scenic
> routing' comment.
I think you're getting so obsessed with network purity, you're
bemoaning the fact that MPLS, L2TP and BGP are all protocols designed
to deal with different administrative domains.
I don't regard it as particularly technically awful that my packet may
hvae to travel 50 or even 100 miles around Ireland to get to another
user on the same local exchange. This is a possibility even in the
Telco world if I'm using one carrier to call another carrier.
> A saner model likely would require different business arrangements,
> and I can imagine you would disapprove.
That's exactly where I have the issue - the existing model has evolved
to fit the "business" requirements. That's been a feature of IP from
day one - if you want a purist model, then the OSI stack is just over
By the way I've administered a CLNP network - it's the ugliest thing
I've experienced since someone put an image of John Prescott having an
affair with Anne Widdecombe into my head.
> If the wholesale-DSL-from-state-telco model is as great as you claim
> it to be, how come there is no thriving market of reseller-ISPs? Why
> is the market converging on a small number of players, each of which
> appears keen to control their infrastructure as 'close' to the
> customer as they can get?
Economies of scale. And it's the nature of many industries that they
start off with lots of smaller companies and then finally consoldidate
into many big ones.
In any case - it's a different discussion.
> Anyway, no matter the business model: Scenic routing sucks..
It's the nature of the world we live in. Its the nature of networks
that traffic doesn't always take the physically shortest path.
> (Yes, IPv4 is costly to administrate, IPv6 hopefully less so, but it
> still sucks too there. Granted that is a factor).
At the risk of starting a whole new thread by it self - how is IPv6
"less costly" than IPv4 to administer?
Hint: The issue isn't the allocation of IP addresses.
> > If you're going to start blathering on about "native" IP access, you
> > need to define what you mean by it. You also need to acknowledge
> > that regardless of how it's done, there's a whole bunch of layer 2
> > protocols underlying that connection.
>> I don't have a problem with L2 protocols (like PPP). I have a problem
> with encapsulation of layers of L2 protocols over L3 protocols
> (L2TP/IP, then again ATM).
But why? Other than aesthetics?
> > The whole point of an IP layer is to abstract the physical topology.
> > It's the entire point of having a layered networking model.
> The layered networking model is to abstract the software
> /implementation/ of a network stack.
> It's NOTHING to do with abstracting interconnections the topology of
> nodes in a network. That's quite a unique interpretation, you should
> patent it.
My viewpoint is hardly unique - the entire point of an IP layer is
that I route IP - I don't worry about what the guys at the lower
layers do. In the broadband aggregation world for instance, I
simply want a lot of point to point links to my client sites.
In $LASTJOB, we had a clear and distinct management domains between
the various layers of the network - we had the fibre guys, the SDH
guys, and the IP guys. We didn't worry too much about what the
other layers were doing. The OSI model is more than purely software
implementation abstraction - it's also about administrative domains.
It frees the mind of the IP administrator from worrying about
pernickity issues like SDH or ATM configuration. It may not have
been the intent of the model to cause that, but that's exactly how
it's worked out.
> > Within that, a number of compromises are typically made based on
> > cost.
> And politics.
You say to-mato, I'll say tom-ato.
> That's a choice borne of the present model, a lot of which is to do
> with politics.
> Do you really, /really/ think the ideal is for Joe and Jack to have
> to use bandwidth on the exchange<->BAS, BAS<->ISP and ISP<->ISP
Yes. That's called "administrative domains". If Joe and Jack
choose different domains, they deal with the consequences.
> Are you really saying scenic routing, at least as the result of these
> franken-networks of tunneled protocol over tunneled protocol, are a
> good thing? :)
There's two seperate issues - the existence of "scenic" routing, and
how its done.
Assume that instead of L2TP and a BAS, $TELCO runs ATM direct to your
ISPs BAS. That's still scenic routing. It does however remove the
ugliness of tunnel protocol over tunnel protocol. Even two users
connected to the same ISP will have to send their traffic to the
central site to exchange.
> The problem is we're sending packets all around Ireland, when we
> could just switch them in one of those huge boxes in (say) Limerick
> city exchange..
And when Magnet and BT unbundle the exchanges in Limerick, and stop
using the BAS, the traffic will still travel all the way around
Ireland so that they can cross over to Eircom at INEX.
> > You also have better control over traffic allocation and bandwidth
> > allocation on a per service basis that way.
> Bollocks :)
No - ATM's bandwidth allocation is much nicer than IPs.
> > If you're going to descibe something as crap just because it's
> > designed to address Layer 8 issues fair enough - I just find it
> > ironic coming from a BGP developer.
> BGP is a good example...
Yes - it's designed to ensure that instead of taking the technically
best path, it takes the path most appropriate based on administrative
domains, cost and politics.
> away by BGP - yet IP (e.g. a packet) still sees the /full/ topology.
> Local considerations (at some given hop) can still therefore apply.
This does raise existential questions about the nature of an IP
packet, but in fact, a packet won't see the full topology. It will
only see the hops that exist on the "best" path (as determined by the
various the routing protocols along the way). It certainly doesn't
see the full topology of the network.
> We /do/ need to make compromises, to reduce information through
> abstraction - but we can do that conceptually in the control plane,
> *without* having to abstract the topology of the actual transport.
> Abstracting the link-topology because the IP control plane sucks (in
> addition to whatever political/competitive considerations there are
> around state-telcos) is not good.
You know - I've read this three times - and I'm still not sure what
you're talking about.
If I buy a broadband connection, I buy a IP link from my house to my
ISPs BAS. How that link is delivered, be it over ATM, ethernet,
L2TP, MPLS, wet string, IP over Avian carrier or the magic pixies
carrying the bits, isn't relevant. It's a link.
In exactly the same way that a few years we sold two leased lines to
two businesses within a couple of doors of each other - that meant
their traffic had to take the scenic route via our data centre. You
know what, it's the nature of networks.
> And yes, a good part of why that is done is cause IP and BGP suck. It
> (like most IP routing protocols) requires manual input of several
> magic parameters at each host (arg!), it's pretty complex, it has a
> wealth of knobs to tweak.
> That doesn't make it good.
> Sadly, many in the network-ops community seem to relish the challenge
> of IP routing being as difficult as possible, and vendors are happy
> to give it to them.
And this is relevant to the issue of scenic routing how? It actually
gets harder to route IP - in either it's v4 or it's v6 incarnation -
if you want to interconnect everything at every exchange and still
have more than one ISP.
> Then, due to the complexity, vendors are even
> more happy to give network-ops tunneling methods to help deal with
> control-plane complexity...
> It needn't be so, but hey..
If you can come with an easy, workable, cheaper solution that allows
multiple ISPs (and therefore administrative domains) to send their
traffic to other users in the same exchange, then I'm interested in
hearing what it is.
Of course, the fact remains that one of the reasons the ISPs aren't
particularly interested in fixing that problem is that the amount of
traffic that is exchanged between users within the same exchange is
Maintained by the ILUG website team. The aim of Linux.ie is to
support and help commercial and private users of Linux in Ireland. You can
display ILUG news in your own webpages, read backend
information to find out how. Networking services kindly provided by HEAnet, server kindly donated by
Dell. Linux is a trademark of Linus Torvalds,
used with permission. No penguins were harmed in the production or maintenance
of this highly praised website. Looking for the
Indian Linux Users' Group? Try here. If you've read all this and aren't a lawyer: you should be!