|
List Info
Thread: more-specifics via IX
|
|
| more-specifics via IX |
  United States |
2007-10-15 00:09:08 |
I have a few customers' customers, who appear at a local IX.
Due to
the MLPA-like nature of the IX, I hear their prefixes both
at the IX
and via my own transit customers. I normally use localpref
to prefer
customer advertisements over peers' advertisements.
There is a customer's customer who is advertising
more-specifics at the
IX (and using a different source AS, to boot). I can think
of a couple
ways to prevent hearing these, but thought I should ask for
suggestions
first.
-Bradley
|
|
| Re: more-specifics via IX |
  Spain |
2007-10-15 04:49:04 |
On 15-okt-2007, at 7:09, Bradley Urberg Carlson wrote:
> There is a customer's customer who is advertising
more-specifics at
> the IX (and using a different source AS, to boot). I
can think of
> a couple ways to prevent hearing these, but thought I
should ask
> for suggestions first.
What exactly is the problem?
|
|
| Re: more-specifics via IX |
  United States |
2007-10-15 07:33:13 |
On Oct 15, 2007, at 7:41, Wolfgang Tremmel
<wolfgang.tremmel de-
cix.net> wrote:
>
> Am 15.10.2007 um 07:09 schrieb Bradley Urberg Carlson:
>
>>
>> I have a few customers' customers, who appear at a
local IX. Due
>> to the MLPA-like nature of the IX, I hear their
prefixes both at
>> the IX and via my own transit customers. I
normally use localpref
>> to prefer customer advertisements over peers'
advertisements.
>>
>> There is a customer's customer who is advertising
more-specifics at
>> the IX (and using a different source AS, to boot).
I can think of
>> a couple ways to prevent hearing these, but thought
I should ask
>> for suggestions first.
>>
>
> you should honor your customers routing policy and
simply accept the
> routes.
Whilst it is nice to accept a downstream of a downstream's
routing
policy like that I don't think it is your place to say that.
The other
response asking what the problem is also is a good example
of the
misunderstanding of problems with the shim6 solution
although at a
different place in the network. If MY policy is to send all
customer
traffic through my customer connections, I should be able to
do that.
To answer the OP's question I'd be looking at manually
filtering the
more specifics if they are also sending the aggregates
through the IX.
|
|
| Re: more-specifics via IX |

|
2007-10-15 08:27:04 |
|
The problem is you're announcing the aggregate prefixes of your customer39;s customers to your upstream providers and the traffic from your upstreams to those networks will be routed through the IX (instead of your customer connection) because of the longer prefix effect and so you're not charging the traffic and you're losing revenue!!!
I think in this case, you have to filter all those prefixes learnt from the IX.
Che-Hoo
On 10/15/07, Wolfgang Tremmel < wolfgang.tremmel de-cix.net">wolfgang.tremmel de-cix.net> wrote:
Am 15.10.2007 um 07:09 schrieb Bradley Urberg Carlson:
I have a few customers' customers, who appear at a local IX. Due to the MLPA-like nature of the IX, I hear their prefixes both at the IX and via my own transit customers.
I normally use localpref to prefer customer advertisements over peers' advertisements.
There is a customer39;s customer who is advertising more-specifics at the IX (and using a different source AS, to boot). I can think of a couple ways to prevent hearing these, but thought I should ask for suggestions first.
you should honor your customers routing policy and simply accept the routes.
Wolfgang
--
DE-CIX Management GmbH Phone: +49 69 1730 902-26
Lindleystr. 12, 60314 Frankfurt Mobile: +49 173 340 4833
Geschaeftsfuehrer Harald A. Summa Fax: +49 69 4056 2716
Zentrale: Lichtstr. 43i, 50825 Koeln
|
| Re: more-specifics via IX |
  United States |
2007-10-15 08:48:29 |
On Mon, 15 Oct 2007, Bradley Urberg Carlson wrote:
> I have a few customers' customers, who appear at a
local IX. Due to
> the MLPA-like nature of the IX, I hear their prefixes
both at the IX
> and via my own transit customers. I normally use
localpref to prefer
> customer advertisements over peers' advertisements.
>
> There is a customer's customer who is advertising
more-specifics at the
> IX (and using a different source AS, to boot)
Time to time you will see this.
You could also hear the more specifics from another peer
that is one of
their transit providers or you could hear them via one of
your transit
providers.
> I can think of a couple ways to prevent hearing these,
but thought I
> should ask for suggestions first.
You can do all kinds of things to other network's routes
especially when
those routes aren't from your customers and what you are
doing doesn't
break connectivity (or solves a capacity problem and
improves
connectivity). However, if you tweak routes of a paying
customer then you
will need to consider what your answer to your customer will
be for
overriding their traffic engineering.
Mike.
+----------------- H U R R I C A N E - E L E C T R I C
-----------------+
| Mike Leber Wholesale IPv4 and IPv6 Transit
510 580 4100 |
| Hurricane Electric Web Hosting Colocation
AS6939 |
| mleber he.net http://he.net |
+-----------------------------------------------------------
------------+
|
|
| Re: more-specifics via IX |
  Nepal |
2007-10-16 05:32:05 |
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Bradley Urberg Carlson wrote:
>
> I have a few customers' customers, who appear at a
local IX. Due to the
> MLPA-like nature of the IX, I hear their prefixes both
at the IX and via
> my own transit customers. I normally use localpref to
prefer customer
> advertisements over peers' advertisements.
>
> There is a customer's customer who is advertising
more-specifics at the
> IX (and using a different source AS, to boot). I can
think of a couple
> ways to prevent hearing these, but thought I should ask
for suggestions
> first.
I have seen the opposite of this as well. ISP X announces an
aggregate
at the local IX, but has some more specifics announced to
the transit
providers for TE needs. To avoid sending/receiving traffic
over transit
links and prefer peering route, they were suggested to also
announce
their more specific to their peers.
In your case, if your customer's customer is multihomed,
they might be
announcing more specific in line with their own routing
policies. If you
don't like it - then you'd setup a specific filter,
resulting most
probably in asymmetrical routing.
thanks
- gaurab
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (Darwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
iD8DBQFHFJMlSo7fU26F3X0RAqYLAJ9suyoqQ5Q9qU6lJKaaH0imAznsUACg
xKxk
CDM0aL0Ij32L5By3lnVjFuQ=
=CgYZ
-----END PGP SIGNATURE-----
|
|
| Re: more-specifics via IX |
  United States |
2007-10-16 10:48:04 |
|
owner-nanog merit.edu wrote on 10/15/2007 01:09:08
AM:
>
> I have a few customers' customers, who appear at a local IX. Due
to
> the MLPA-like nature of the IX, I hear their prefixes both at the
IX
> and via my own transit customers. I normally use localpref to
prefer
> customer advertisements over peers' advertisements.
>
> There is a customer's customer who is advertising more-specifics at
the
> IX (and using a different source AS, to boot). I can think of
a couple
> ways to prevent hearing these, but thought I should ask for suggestions
> first.
>
> -Bradley
>
>
>
Sounds like misconfiguration. Have you tried contacting them and
explaining the problem? |
| Re: more-specifics via IX |
  United States |
2007-10-17 18:06:48 |
On 15 Oct 2007, at 03:49, Iljitsch van Beijnum wrote:
>
> On 15-okt-2007, at 7:09, Bradley Urberg Carlson wrote:
>
>> There is a customer's customer who is advertising
more-specifics
>> at the IX (and using a different source AS, to
boot). I can think
>> of a couple ways to prevent hearing these, but
thought I should
>> ask for suggestions first.
>
> What exactly is the problem?
well.. the problem of course is that you pull in the traffic
from the
aggregate transit prefix which costs you $$$ but then you
offload it
to the customer via a peering link for which you are not
being paid
its a pain but you cant stop the customer from doing it..
you can
however filter your customers prefix at the IX (an ASN
filter would
be easiest)
if you think it is malicious, you may want to hit them with
something
official (IANAL)
Steve
|
|
| Re: more-specifics via IX |
  United States |
2007-10-17 21:55:25 |
Thanks for the suggestions.
On Oct 17, 2007, at 6:06 PM, Stephen Wilcox wrote:
> well.. the problem of course is that you pull in the
traffic from the
> aggregate transit prefix which costs you $$$ but then
you offload it
> to the customer via a peering link for which you are
not being paid
A bigger problem is that my IX peer pays less to my customer
for
transit. If my customer notices that transit traffic has
been going
around him, he may be grumpy. I prefer happy customers.
> its a pain but you cant stop the customer from doing
it.. you can
> however filter your customers prefix at the IX (an ASN
filter would be
> easiest)
In this case, the IX peer had let their transit provider (my
customer)
source the routes, then later advertised their own routes at
the IX
using their own ASN (so inconsistent source-as, and my
as-path filter
missed them). I don't think they were trying to steal
bandwidth; just
sloppy networking.
I can either build a big import filter, dropping routes
offered to me
at the IX that are subnets of routes advertised to me by my
transit
customers (doesn't scale); or just audit customer routes
versus peer
routes periodically, looking for "bandwidth
stealers". It sounds like
that is the usual approach.
-Bradley
|
|
| Re: more-specifics via IX |
  United States |
2007-10-18 12:57:17 |
On 17 Oct 2007, at 20:55, Bradley Urberg Carlson wrote:
>
> Thanks for the suggestions.
>
> On Oct 17, 2007, at 6:06 PM, Stephen Wilcox wrote:
>> well.. the problem of course is that you pull in
the traffic from
>> the aggregate transit prefix which costs you $$$
but then you
>> offload it to the customer via a peering link for
which you are
>> not being paid
>
> A bigger problem is that my IX peer pays less to my
customer for
> transit. If my customer notices that transit traffic
has been
> going around him, he may be grumpy. I prefer happy
customers.
Okay but:
1. Your customer/customer's customer is the one doing the
broken
routing here not you.. if he wants to be grumpy you should
point him
in the direction of the guy who is announcing the bad routes
in the
first place!
2. If I'm following this, your peer pays your customer? So
you are
peering with your customer's customer? If that was me I
would either
depeer them or tell them that you have an issue and need it
resolving
urgently or you my depeer them.
You're not the bad guy here ;)
>
>> its a pain but you cant stop the customer from
doing it.. you can
>> however filter your customers prefix at the IX (an
ASN filter
>> would be easiest)
>
> In this case, the IX peer had let their transit
provider (my
> customer) source the routes, then later advertised
their own routes
> at the IX using their own ASN (so inconsistent
source-as, and my as-
> path filter missed them). I don't think they were
trying to steal
> bandwidth; just sloppy networking.
wow, i think i need a diagram!! :P
i don't like sloppy networking, i would depeer anyone who i
find is
not up to my standards on what makes a 'peer'. this doesnt
happen
very often but if we want to educate people you can try
talking and
if that fails take action.
>
> I can either build a big import filter, dropping routes
offered to
> me at the IX that are subnets of routes advertised to
me by my
> transit customers (doesn't scale); or just audit
customer routes
> versus peer routes periodically, looking for
"bandwidth stealers".
> It sounds like that is the usual approach.
not really, its pretty unusual. now that i understand the
picture
better tho i think you dont want to be filtering.. 90% of
people
won't peer with downstreams to avoid this kind of issue..
either you
need to do that too or you need to make them fix it (if your
peering
is valuable to them they will do it)
don't forget they are getting a free lunch here, and that is
unacceptable. if they are intentionally stealing your
bandwidth then
that is a major problem, if its an accident then you really
should
take action and insist they fix it. immediately and
temporarily
dropping the peering would be a good option
Steve
|
|
| Re: more-specifics via IX |
  United States |
2007-10-18 16:03:36 |
Stephen Wilcox wrote:
>
>
> On 17 Oct 2007, at 20:55, Bradley Urberg Carlson
wrote:
>
>>
>> Thanks for the suggestions.
>>
>> On Oct 17, 2007, at 6:06 PM, Stephen Wilcox wrote:
>>> well.. the problem of course is that you pull
in the traffic from the
>>> aggregate transit prefix which costs you $$$
but then you offload it
>>> to the customer via a peering link for which
you are not being paid
>>
>> A bigger problem is that my IX peer pays less to my
customer for
>> transit. If my customer notices that transit
traffic has been going
>> around him, he may be grumpy. I prefer happy
customers.
>
> Okay but:
>
> 1. Your customer/customer's customer is the one doing
the broken routing
> here not you.. if he wants to be grumpy you should
point him in the
> direction of the guy who is announcing the bad routes
in the first place!
s/broken// and s/bad//
'broken' and 'bad' are all a matter of perspective here.
>
> 2. If I'm following this, your peer pays your customer?
So you are
> peering with your customer's customer? If that was me I
would either
> depeer them or tell them that you have an issue and
need it resolving
> urgently or you my depeer them.
It's an MLPA policy based exchange (and probably just using
a central
route-server) not bi-lateral peering. De-peering isn't
possible here.
It's not an excuse for lack of filters, but as the OP
pointed out, the
filters weren't expecting the routes from their customer's
customer.
-david
|
|
[1-11]
|
|