On 14/11/2012 22:18, Christopher Morrow wrote:
>   1) What is a 'route leak' (perhaps the above draft identifies one
> examplar to be used in that definition)

dunno really.  Some foo which involves prefixes being sent to places where
they oughtn't have been sent to.  This is possibly more of a legal /
wordsmithing problem than anything else.

>   2) Are 'route leaks' a problem that Operations folks care about

speaking with my IXP operator hat on, hell yeah.  INEX operates a route
server cluster which provides service to ~70% of our members.  If we were
to operate an unprotected system and had a prefix leak from a client, it
would destroy confidence in the system and people would depeer.
Connectivity would be directly harmed.  In my experience, it takes very
little for people to lose trust in a system like this.

We use strict prefix filtering based on IRRDB data with a vengeance, but
ended up also having to implement maxprefixes after someone attached the
INEX route server as-set macro to their own as set macro (mistake #1), then
announced all the prefixes they received back to INEX (mistake #2).  As it
happened, their prefixes were depreferred because of the asn path length,
but if they hadn't been depreferred, it could have taken down the entire
system.  Ouch.

Anyway, point being: yes: route leaks matter, yes: at least some operations
people care, and belt + braces is sensible.

>   3) Should IDR (or the IETF proper) address 'route leaks' with some
> form(s) of fix action.

mmmm, difficult one.

So the current thinking seems to be heading in the direction of RPKI which
has a lot of interesting technical characteristics, probably the most
interesting of which is the max prefix length.  One interesting
characteristic of RPKI is that in order for it to have any real function on
a live network, it will be necessary to completely drop rpki-invalid
prefixes.  Merely depreferring them is a bit useless.

Problem is, the current implementations of rpki only deal with origin
validation, and not as path validation.  Work on this has begun in terms of
a list of specifications (draft-ietf-sidr-bgpsec-reqs), but hasn't
progressed much beyond that as far as I can tell.  I wonder if some of the
sine-qua-nons in this document are mutually exclusive.

RPKI will only be fully functional from the point of view of preventing
leaks (whatever leaks are) when we have full as path validation.

Being honest, RPKI concerns me from two points of view.  From a technical
point of view, BGP is a fully distributed system which forms the basis of
global network reachability, yet RPKI seeks to impose a top-down control /
validation mechanism for it, albeit based on many potential roots.  I
cannot seem to stop being concerned that this is an architecturally bad
idea from a technical point of view, even though there are many safeguards,
including:

- recommendations for using local caches
- lazy validation
- recommendations for validation only at ASN boundaries
- recommendations in draft-ietf-sidr-origin-ops

... even still, there is a double architectural inversion going on here,
where a full-distributed, very low level protocol is being made to depend
on a top-down, much higher level protocol.

>From a non-technical point of view I also have serious concerns about RPKI
in terms of it being used as a hammer that politicians / bureaucrats / LEAs
/ judges / etc are (and that's "are" as in present tense) extremely
interested in as a potential future means of blocking content / services
which they happen not to like.  Although GROW is not the place for this
particular discussion, this aspect of RPKI concerns me to the extent that I
am disturbed about the wisdom of developing the technology in the first
place:  we're building an on/off button for prefixes throughout the
Internet.  We need to be careful what we wish for because one day, we might
get it.

Nick

_______________________________________________
GROW mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/grow

Reply via email to