Mmm, indeed.

The meeting materials page records that minutes and agenda were uploaded 
successfully.  So I do not know why it is not appearing in the proceedings page.

Thanks for the alert.  I'll check with the secretariat as to what happened.

As for the pdf - the Acrobat reader is openly available without a license, as 
far as I know.  As noted, there are other openly available readers.  And it is 
one of the accepted formats for submitting minutes.  So publishing in pdf is 
within requirements.  

That of course doesn't help you all that much, so  I've constructed a .txt file 
which is appended.  Some formatting was lost, so I'd advise viewing the pdf if 
you can.

As for the jabber logs - there are jabber logs for many more days than the wg 
has scheduled meetings.  Access to the jabber room is open to all at any time.  
You can not conclude that there was a meeting from the existence of a jabber 
log for any day.

--Sandy
________________________________________
From: t.petch [[email protected]]
Sent: Thursday, May 17, 2012 4:28 AM
To: Murphy, Sandra; [email protected]
Subject: Re: [sidr] minutes of 30 Apr 2012 interim meeting

Mmmm

The proceedings page still has no link for Minutes or Agenda, 18 days
on.

The attachment to this, which makes the size of the e-mail trip a spam
trap for me, is in .pdf, which I do not find friendly, not having an
Acrobat licence.

The jabber log lists 11 (and rising) logs for May 2012 - is the meeting
still going on:-)

Mmm

Tom Petch

----- Original Message -----
From: "Murphy, Sandra" <[email protected]>
To: <[email protected]>
Sent: Friday, May 11, 2012 10:57 AM
Subject: [sidr] minutes of 30 Apr 2012 interim meeting


The minutes of the 30 Apr 2012 interim meeting have been uploaded to the
proceedings site.

The minutes can be found at the proceedings page for the meeting:

http://www.ietf.org/proceedings/interim/2012/04/30/sidr/proceedings.html

Pointers to the jabber log and the webex recording are in the minutes.

The proceedings page does not yet show the uploaded files, so I am
appending the minutes pdf here.

Please send comments, corrections or additions to the list.

One disappointing note.  There were many complaints on the jabber about
the distraction of the typing sounds in the audio.  No one mentioned a
constant echo in the audio.  This makes the recording very difficult to
listen to.  I trust (from the lack of complaint) that this was not heard
at the time by those on the conference call!  I do not know why it is in
the webex recording.

--Sandy


------------------------------------------------------------------------
--------


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


Webex recording: 
https://ietf.webex.com/ietf/lsr.php?AT=pb&SP=EC&rID=6714127&rKey=0ad5dd297cf4
b13e

Jabber Log: http://www.ietf.org/jabber/logs/sidr/2012-04-30.html


Meeting delayed due to technical difficulties – getting remote participation 
tools working 
together (teleconference call, webex session, webex integrated conference call, 
conference phone, feedback, etc.)

Chris Morrow and Sandy Murphy serving as chairs

Intro slides (note well, housekeeping tasks, agenda) viewed

(notes slide 2)
Basic Incremental Deployment Strategy
•       Origin Validation
–       Deploy code
–       Start doing some metrics
–       Start turning it on
•       Bgpsec
–       Deploy code
–       Start doing some metrics
–       Start checking – logging
–       Don’t need to worry about local originated route

(notes slide 3)
Operational Issues in Incremental Deployment
•       Bgpsec incremental deployment
–       Randy: if I am on this router, I want origin validation checked now on 
this 
router – don’t want to find out from neighbors that things are bad
–       Chris says want to check signatures to make sure the origin is 
validated 
and authentic
–       Chris:If you don’t have RRs, you have full mesh, simplest case
–       Randy: if a RR and its clients have no externals, then those RRs don’t 
need to speak bgpsec
–       John: to repeat: in general RRs have to be bgpsec aware because bgpsec 
attrbs are non-transitive. But only those RRs in a control path betw two 
ebgp borders need to have bgpsec enabled.
–       Chris: in the case of bgpsec capability negotiated – don’t send attrbs. 
 So if 
RR is not bgpsec aware will not receive attrbs and they don’t get to other 
side of AS.  So RRs need to be upgraded when they are in the control path 
between ebgp routers.
–       Randy: there are routers in your topology that will not need to be 
bgpsec 
aware for your AS to be bgpsec capable.  Some routers will be bgpsec 
capable but may not validate and may sign anyway. 
–       For locally originated routes – you only apply bgpsec attrbs 
(signatures) 
when you are sending the update over an ebgp session

(notes slide 4)
[Sidebar: Issues in Protocol Deployment]
•       Stack of to be discussed
–       Confeds
–       Add-path
–       AS aliases (replace-as in juniper speak)
–       Forward policy (eg route servers) –brief discussion pointed to need for 
firm definition (is route server example only case? Etc.)
–       Brian’s AH-HA to be explained later

(notes slide 5)
Operational Issues in Incremental Deployment (continued)
•       In bgpsec we only do signatures and validations of signatures at the 
borders
–       What about RRs doing best path selection
–       A:Edge router passes on results of its signature validation in some AS 
specific policy knob (community, localpref, etc)
–       Q: is there any requirement that RR consider signatures in best path 
selection
–       A: (Ruediger) RR should do its best path selection based on this local 
policy knob
–       A: (Jason) the local policy knob is not bgpsec attribute so any router 
can 
make a decision based on that.  Operational quirk – have to set up RR to 
make decisions on locally originated routes as if they were signed.
–       A: Randy: routes is marked somehow – and the mark is available to your 
policy.
–       A: John: you should mark the route as if it had been signature checked 
and 
marked (you trust yourself to have locally originated)
–       Q: Sandy: you have to be careful to make sure that there’s no strange 
way 
to make the AS_PATH null so an external path doesn’t look local (as an 
attack)
–       A: Brian: if you redistribute from bgp to igp and back to bgp, origin 
validation should catch that.
(notes slide 6)
•       Another inflection point:
–       Going from not doing bgpsec to doing bgpsec
–       Could change preference of best path 
•       Could make big changes in traffic movement in your AS, traffic 
swings
–       Randy: at layer 9, we are going to a more secure network, that’s a 
feature!
–       Chris: hopefully you do some modeling first
•       In case of 701, as-path wins, not peer
–       Brian: much bigger than that.  When change of policy of two bgpsec 
capable peers, to change to prefer signed routes, you may have difficulties 
making that an incremental change
–       Randy: you are changing your policy, you know there will be 
consequences, these are consequences you want to have happen.
–       Wes: need tools Randy: have tools now to model policy change
–       Brian: but there may not be a way to tell who goes first – chicken and 
egg
–       Chris: internally, you may choose to prefer signed, but process will be:
•       Deploy code
•       First mark routes with potential bgpsec result
•       Look and see what is right and what is wrong
•       Still believe this will be multi-year process (5? 7?)
•       Not a flag day.
–       Wes: in documentation of simple case or somewhere
•       Update code
•       Watch what goes by and see what changes – dump somewhere else
•       Right policy based on that, before you pull trigger on policy 
change
•       Don’t drop invalid initially, just downpref it.
•       Maybe this goes in a secops drop
(notes slide 7)
•       Randy: ops do these things every day, we have tools (not great), we do 
deployments, upgrade links, etc.
•       Brian: impact is minimized by making policy change late
•       Chris: every network will make a different choice.  And a deployment 
time frame 
of years.  But not a knife edge deployment – takes a couple years of
•       Chris/John: do we need an RFC, or a presentation, John: but confeds may 
need 
protocol change, Chris/John: some common deployment+network models
•       Warren: isn’t this just another knob in policy?  But then there are 
presentations 
about how you tune meds, so maybe.
•       Chris: don’t think this is RFC, maybe vendor docs.
•       Brian: think RFC is right documentation model  just simple case with 
RRs.  
(Haven’t proved it is scalable if you can’t show you’ve considered that.)  
Brian 
volunteers to write up simple case with RRs.
•       Ruediger: remember 4byte AS attribute – there was no deployment draft.  
Chris: 
to Ruediger – are you saying that went well and is proof we don’t need a draft, 
or 
that is a warning that we DO need one.
•       Randy: can’t write docs for all the possible things it would be good to 
know.  
Where do you draw a line.  Need a decision of what MUST go in.  Heather: docs 
can’t hurt – what is the down side?
(notes slide 8)
After Break – Next Topics?: Forward Policy Signing
•       John: votes to discuss things that change protocol spec.  Eg confeds
•       Brian begins: need some way for signer sending update to express what 
it was 
authorizing the recipient to do.  eg.: authorizing route server to remove 
itself from 
as-path
–       If the recipient might do something other than just add its as to the 
as-path, 
there should be somewhere that this authorization should be expressed.
•       John: interesting, but old bgpsec goal was to protect what is currently 
in bgp 
without changing the spec’d model. You are talking about changing the bgp 
spec’d behavior.  Might be a non-trivial change.
•       Rob: you might want to think about whether this needs to be in the 
bgpsec 
protocol – maybe a new rpki object?
•       Warren: you have one suggestion – but there are lots of others.  
Carrying policy 
around
•       Randy: not carrying policy – carrying a policy of what YOU can do – a 
very deep 
hole
•       John: is there good semantics in what we have now?  Is there room for 
later added 
semantics?  Those should be the decision points
•       Brian: but bgpsec is authorizing the path validation that comes to you
•       Randy: but we forward on destination, we accept as-path and propagate
•       Brian: but we are validating the path, right?
•       Randy/John: yes, path we got,  you are talking about path forward.
(notes slide 9)
Route Server
•       Y does not know that RS is really a route server
•       X and Y should authorize and check the role of the route server
•       Randy: Y can validate that someone doing pcount=0 is a route server to 
it
•       How does Z know that Y was correct in what it did
                        X   ----->     RS   ---->  Y   --->   Z
•       Randy: X saying that RS is an authorized route server is forward policy 
restriction
•       And is a big hole
•       Randy: we are trying to protect the protocol from being violated, not 
the business 
relationship
•       John: kinda analgous to route leaks – you want Z to second guess if 
what Y did 
was in accordance with what X wanted it to do.   Given that we have the option 
for Y to check the use of pcount and right now Z must trust that.  Suppose you 
stuck something in RPKI – who would be making the authorization
•       Brian: all members of the route server would authorize route server
•       Jason: ebgp multi-hop and gre tunnels should be considered as well.
(notes slide 10)
Discussion of Confeds
•       Is there some way to indicate that the as-path is only for in the 
boundary of a 
confed?
•       Is there a way to indicate semantics not already covered in the bgpsec 
attrb?
•       Issue with confeds:
–       In current real world – you have some sub-as’s, adding themselves to 
the 
path as as-confed sequences, get removed at boundaries.
–       But no way to add these as-confed type things in the as-path.  Needed 
for 
loop detection.
–       Need some flag in sig attrb?  Mark each – throw away all of the confed-
marked sig attrb.
–       Warren: don’t you know all the as’s used inside the boundary? A: you 
don’t know the enumeration
–       Randy: spec says it can be arbitrary topologies, but deployed 
topologies 
are more simple – a core AS with some stub ASs attached to core.
–       Q:Sandy – external links come in where?
–       Randy: the simple case – only possible loops in this case would be 
solved 
by adding normal sig attrbs and then dropping at the boundaries.
–       Rob: think John says: mark confeds inside confed so there is an easy 
way 
to strip the confed sigs.  Randy says: inside confed, add sigs as usual and 
strip at border.
–       John: current bgp implementation strips all confed sequences and does 
not 
consider the as number
–       John: this is part of discussion of can we get rid of AS-PATH entirely
–       Randy: external AS neighbor of stub AS in the confed topolgy knows only 
the core AS number.  Externally only core AS number is seen.
–       Brian: we could do this using pcount=0, using some local AS

<<< picture at board was large circle labeled "coreAS" with smaller circles 
adjoining at 
the boundary being the stub ASs>>>

(notes slide 11)
•       John: by defn: you don’t use core AS as the public AS.  Randy: oh, no, 
not true.  
John: ooo, did not know that could happen.  Wes: that’s how we do it also.
•       Chris: external to ebgp router in subAS, cbgp from subAS to core, 
External 
forward signs to ebgp customer facign router, cbgps forward signs to core, cbgp 
by core to subAS, then subAS ebgp router wants to strip the first subAS number 
and does not know that subAS number. (Hard to do this in text).
•       John:  One way to tell what to strip: start at origin.  When you find a 
forward sign 
to your public AS, strip everything past that.
•       (External neighbor thinks it is connecting to the public AS.  Because 
some people 
use the public AS as the core AS, can’t use the core AS as the marker of the 
confed topology border.)
•       One thing that this hack does break: right now in the protocol, my AS 
can appear 
in the protocol more than once – the allow loops (allow ownAS in (in cisco), 
loops (in juniper))
•       Heather: lies are bad, so dropping lies are perfectly fine.
•       Warren: what about path poisoning?  A: path poisoning will not work in 
bgpsec.  
End story.
•       (Work at board emphasizing that customer external AS configs itself as 
peering 
with the public AS.  Wes wants to confirm with his network.)
(notes slide 12)
More Agenda Bashing
•       Some early departures want to influence the agenda ordering.
–       Confeds (cont’d)
–       Add-path
–       AS aliasing (replace-AS, local-AS, etc.)
–       Revisit the incremental intra-AS deployment – anything more there?
–       AS-PATH – in the protocol
•       Brian took five min to discuss a out-of-the box way to do different 
crypto (maybe 
shared secret based) – on the list.  (See Brian's AH-HA in notes slide 3)
(notes slide 13)
Discussion of Confeds (contd)
•       Confeds solution:
–       Use pcount=0 everywhere and strip all pcount=0 that have your AS in 
them.
•       Problem – you may be usign a sub-as to peer with a RS
•       Have an attribute that has an attrib that carried the confed nature of 
paths.  John 
proposes that that attrb is the as-path, but that attrb is not currently in the 
bgpsec 
protocol
–       Sriram: the as’s are in the bgpsec attrb A: but we are speaking about 
the 
real bgp as-path attrb
–       Warren: was not thinking of using as-path attrb, a brand new attrb.
•       John: Augment current bgpsec sig attrb with a flag field that says 
“confed”.
•       Rob: likes having a separate attrb, so people who want this pay the 
pain, not the 
whole world.  John: but bgp attrb space is limited, would rather not use 
another 
bit for this.  Small point, but.
•       Sirram: what distinguishes confeds in current bgp?  John: type code.  
We don’t 
have that in bgpsec protocol.  Maybe we can add that.  (randy: why have extra 
complexity – people will think up ways to use it – can't predict future)
•       John’s summation
–       1  rewind (start from origin to first signing mentioning public as – 
relies 
on external session going to public AS)
–       2 Flag in sig attr
–       3 Child-of-AS-path (some other attribute that carries “confed” 
characteristic
•       Warren: can we just use a marker in the sig attrb?
–       Trust model: ski length 0 , sig length 0, --- marker – don’t need to 
sign 
over it.
–       Rob: repurpose whole bgpsec sig attrb to carry this John: that’s what 
we 
bit into when we got rid of as-path
•       John adds
–       4 explicit “entered confed” marker
•       Me: is this a problem with someone entering the marker who should not 
do so?
–       A: entry point into confed knows that it is in a confed and knows it is 
the 
entry point
–       And can be required to check incoming data for entered confed marker – 
and throw error
•       John: likes marker, only thing better about flag is it fits existing 
structure
•       Warren: doesn't like flag because it gives 7 extra bits  Matt: 
additional bits would 
allow some coloring
•       Wes: additional bits in attrb – does that get signed?  A: yes.
•       Warren: that means all routers have to sign on entry and exit from 
confeds – 
incremental deployment issue
•       Randy: learning anything more? John: probably not.  Go off, think about 
it, 
discuss on email?
(notes slide 14)
Add-Paths
•       Issue: only sign your best path
–       But that breaks add-path
–       Sriram had text and will send it to the sidr list.
(notes slide15)
AS aliasing Replace AS Local AS
•       Wes: channeling Shane – what about AS merging?
•       And path poisoning
•       John: in my implementation
–       There’s something that lets me be any one of a set of ASs when I talk 
to 
an external peer
–       Strip private ASs – ok if they are all on the right side of the path
•       Warren: Don’t you have to rewind until you find the first that is 
not private and  not in your member list
•       Would that not work for confeds?
–       Replace arbitrary AS in the middle of the path – that’s never going to 
work, give it up
(notes slide 16)
Replace as – AS migration
•       AS 1 is connected to 2 and 3.  but 3 thinks 1 is 5
                                  2  -----     1     -----  3
–       One way is use pcount=0.
–       This exposes the fact that 1 exists to 3
–       Randy: what does 3 see from 1?  Jason: It might see 5 1 2 or 5 2, 
depends 
on the config.  From 3 to 2, 2 sees 3 1 or 3 5 1.
–       Going to have to be knob.
•       Documented?  (keeps coming up, so should be documented)  (Matt: doesn’t 
think 
it belongs in core spec, Geoff agrees on jabber, Randy agrees to go into bgpsec 
ops)
•       Remove private hack: what about forward signing problem if external 
peer signs 
to private
–       Wes: can you transit a private AS?
–       Randy: keys used to sign for router with private AS decend from a trust 
anchor that anyone wanting to validate the signatures must have that same 
private trust anchor (operationally quite messy)
–       Geoff on jabber: that would mean that if a private AS the rest of the 
Internet must have that private trust anchor
–       Randy: can’t transit a private AS in bgpsec
–       Wes: but what about external person peering with the private AS?
•       John: but we have 4 byte AS – why do we need private ASs
•       Ruediger: private AS as stub is OK – remove private AS and originate at 
upstream where private AS is stripped.
•       Wes: is it ok to make it impossible to transit a private AS – because 
you run into 
all these problems.
•       Brian: use public IP address for the router?  (but how to establish 
relationship 
between router IP addr and AS number)
(notes slide 17)
There Might Be 50 Ways to Leave Your Lover but 150 Ways to Mess Yourself Up in 
BGP
•       Ruediger: don’t add complexity for those 5000 AS numbers.  Do not 
propagate 
private ASs into the public internet.
•       Ruediger: some customer is sending 200 /32s with private ASs.
•       Geoff: I see about 10 of them.
•       Rob: but these are supposed to be used in trust boundaries and stripped 
outside – 
transiting is different and hard to define trust model.
•       Wes: if you have a reason you need to transit the AS that is using a 
private AS, 
use a public AS (maybe one that is shared for this purpose).
•       Chris: need to address this if there is a use case where it woulod be 
very painful to 
use any other mechanism.
•       Brian: people are using BGP but their engineering was done by 
consultants who 
have been gone – and want to add bgpsec.
•       So until we find the painful can’t-do-this-any-other-way case, we will 
move on.
•       Jason: when people are using private AS, do they sign? Always? Never? 
Sometimes?  
•       Brad NTT: some customer who share a public AS number.  [RFC2270 sort of 
ASN]  – (701 uses one of these things as a shared cutomer AS number) – 701 
could sign the prefix as [RFC2270 ASN] and 701.  The [RFC2270 ASN] could be 
stripped (Wes: for Sprint this does get stripped.)  If the IP addr belongs to 
the 
customer, the [RFC2270 ASN] is not stripped.  Allows one customer to sign an 
announcement originating another [RFC2270 AS] customer’s space (because each 
customer would allow [RFC 2270 AS] to originate their IP prefix)
•       Discussion of DNS analogy that I missed about primaries and secondaries 
and 
muddying the waters.
•       When there’s a routing issue because a /24 is flapping, you want it to 
be obvious 
your customer’s fault not yours and that is why you do not strip their use of 
RFC2270 AS
•       Brian: if the [RFC2270 AS] is doing the signing on a router that is not 
under the 
customer control
•       Can’t tell the AS just go get your own public AS, because they are not 
multi-
homed and do not qualify under RIR policy.
–       Just fix the RIR policy
(notes slide 18)
Path Poisoning
•       Just can not work
•       Deal!
(notes slide 19)
Intra-as deployment
•       Are we done?
•       Chris: we have a path forward
–        brian volunteered to write a simple RR case
–       And then things will go forward
•       Randy: do we understand the technology and are comfortable with it.
–       Simple example seems to work.
•       Brian but what about case where AS wants to sign the origin 
Randy: when injection occurs from another protocol into my bgp, 
there is no AS-PATH.  Inside my AS, I know it is mine, so I know 
what the AS number is, so I can add that first signature.
(notes slide 20)
Discussion of Signing over IBGP Sessions
•       John: what is use case for signing over ibgp session?  Don’t trust 
router?  
•       Brian: you are leasing routers, leasing bandwidth, people can put 
themselves in 
MITM position.  Some people may want to use bgpsec on ibgp sessions rather 
than transport on their sessions.  
•       Warren: how many people are using transport security on their ibgp 
sessions?  A: 
most of the big guys will do that!!!  
•       Rob: hard to say there may be a need for this someday so lets add this 
complexity 
now.  
•       Geoff: hard to address unless we understand threat model.  
•       Brian: was raised on list as comment on threat model document and 
“shouted 
down by chairs”.  
•       Ruediger: do you want the thousands of ibgp routes I carry internally 
in my 
network to have this protection?  
•       Ruediger: Brian, what you are interested in is protecting the other 
attrbs in the 
update, not the empty as-path.  
•       John: since I don’t understand the use case: (a) transport security 
will not solve 
problem? or (b) I don’t want to use transport security?  
•       Brian: I don’t want to use transport security.  Don’t want to have to 
deploy 
transport  - two security solutions at the same time.  
•       Sandy: but transport security protects against other vulnerabilities 
that bgpsec of 
ibgp updates would leave unprotected.  
•       Brian: understand that, presenting this as a incremental way to get 
bgpsec 
deployed.  
•       Rob: thinks security ADs would object to a solution that would leave 
such an 
obvious vulnerability uncovered.  
•       Ruediger: does your proposal apply only to originated routes?  Are 
other attrbs 
that are not used externally (if you don’t your own network) also to be signed. 
 
Interested when mitm can inject routes.  Think if you can’t trust your own 
internal 
network, someone can inject something bad
(notes slide 21)
Prepending Issues
•       <anonymous> reports from message from Jeff Haas.  In bgp today, 
upstreams can 
add prepends of my AS number (with or without my permission) and those 
prepends stay when upstream propagtes the update.
–       John: some implementations do loop detection on ibgp as well as ebgp, 
which means prepended looks like a loop.
•       <anonymous> prepend a private AS the needed number of times and 
translate to 
my public AS number 
•       (Brian reports a solution to his own concern about untrustworthy 
internal 
networks: use private AS internally to sign with, then strip on exit)
(notes slide 22)
Protecting Other Attributes
•       Brian: Ruediger brought up a point – is there any way to protect any 
attribute than 
AS_PATH
–       Randy: unsure of trust model of other attrb
–       Randy: two possibilities presently 
•       Business relationship (route leaks)
•       Signature expiration time
•       Geoff on jabber: Brian needs to go back to SteveKent’s analysis of what 
needs to 
be signed.
(notes slide 23)
Freshness
•       SteveBellovin: still concerned that we don’t have a way in a short time 
frame to 
roll a router signing key.
–       Randy: concern is security compromise?  Yes
–       Beaconing does not seem to work
•       One suggestion (Steve Kent’s ??) flood a “refresh your rpki” in bgp. 
Randy: 
inband?  The one that wants to do this is compromised router SteveB/John: or 
the 
AS/NOC
–       Beaconing has problem that there is incentive to beacon too often, this 
does not
•       A: does this accomplish this  B: does it work
•       Rob: how do you authenticate this?  A: what you want to do is rate 
limit this.  Q: 
how to sign  A: AS number-ish
•       <Geoff Huston> a broadcast message to everyone else that says "generate 
refresh 
queries to the repositories" is a great DOS attack
•       Warren: someone owns my router and issues enough of these to make rate 
limiter 
throw alarm and router’s key gets rolled.  (if noc gets compromised, game over) 
 
Wes: or employee leaves.  There are methods to manage this.
•       Brian: if you could flood info and identify alternate paths, could 
signal with route 
leak material to ensure people who need the new info get it.  Chris: not 
talking 
about BGP data, talking about RPKI data like CRLs.  Brian: using BGP to signal
(notes slide 24)
Freshness (Cont'd)
•       Back to freshness and stolen keys…
•       Chris: problem is that we have all data in same place and so same fetch 
process 
applies (poll from repository)
–       Can we have a separate communication mechanism for some objects?
–       Rob: there is a protocol for retrieving CRLs in real time, but we are 
trying 
to avoid it!
–       Chris: Danny asks why twitter can communicate in near real time
•       Brian: if 13 routers are compromised…  warren: if someone owns 13 
routers they 
could just turn them off.
•       steveB has left but he said he would be happier if there was a way to 
get the key 
change distributed faster.  Rob: me too, I just don’t know how to do it cheaply.
•       Chris: a way for repository to say “ok, ok, ok, new info”
•       Rob: poll or flood, take your pick
•       Chris: just trying to say that there are different classes of data and 
some have 
different freshness concerns.  Rob: but just talking about different protocols. 
 
Would like to talk to this with Tim and SteveB in the room.    Chris: besides 
different protocols can we concern ourselves with different requirements for 
different objects?  Need DATA.  Tim might be interested in some long term 
measurements.  Rob: Randy says we don’t have a long term baseline for how this 
would work in the future – we don’t know what the expected behavior would be.  
Should be.
•       Reports that rsync has some features that don’t work well in 
validators.  Rob: like 
maybe an atomic publication of a bunch of data.
•       Chris: freshness requirement – some time to live – cert expiration time 
gets 
reached.
•       Rob: DNSSEC has signature lifetimes AND ttls.  Should not be combined.  
Signatures are end-end, ttls are cache behavior.  Different animals.
(notes slide 25)
Repository Discussion
•       Chris: if I am gathering from the world but you won’t give me this one 
object so I 
keep using the object I have but then the sig expires and OOPS.  A: yep.
•       Rob: for any publication mechanism, you have to worry about how long it 
takes 
for data to get through the system.
•       SamWeiler: any real need to have a push of data?  (Actually a notify – 
go get new 
info).  IETF has an old protocol 2244 called ACAP and ask for notifies for 
anything that would meet that search criteria.  <<something about it does not 
scale>>  Brian: but you can’t block a notify  Rob: oh, yes you can.  SamWeiler: 
acap has authentication per user.
(notes slide 26)
Repository Discussion (Cont'd)
•       Chris: rqmts about repository – how many objects, how many retrievals, 
how 
often, etc.
–       Rob: George Michaelson in RIR group looking at this long ago – 10K-
100K of objects in the rpki.  So nowadays, maybe 100K-1M.  Originally 
retrievals once per day, now I poll once per hour, could poll several times 
a day (4-6hrs) but 1/day is a bit slow.
–       If you are doing a make before break, everyone needs to get the make 
before you do the break, so any slower than 1/day meant you deserved to 
lose.
–       ROAs are most frequent data change.  Manifests change when any data 
changes.  Crls change ..<missed>
•       Chris: we have 1M objects now, in future bounded by (multiple of the) 
size of 
routing system. Currently 405K+8K routes (v4+v6). Geoff reports his extreme 
outside view is upper bound 25% growth per year.
•       Churn is low, number of objects is large.
•       Geoff: v6 AS advertises 2 routes, v4 AS advertises 4 routes.
•       Warren: deaggreages my routes is an advantage, deaggregates my ROAs is 
not.  
Geoff: but you want to have every alternative covered.  Would pre-provision all 
5 
neighbors even though currently announcing to just 2.
(note slide 27)
Discussion of Repository Size
•       Chris: are there other things that I might want to certify that I might 
want to use 
the rpki system for?
–       Geoff: may want own system for each
–       Rob: there are router keys, crls, manifests, etc.
–       Chris: size of routing system times multiple of 2? For right now.  That 
seems to give lots of room and a safe upper bound.
–       Geoff: 600K in five years  Chris: so you will need 1.1-1.2K objects in 
five 
years.  If the change rate is 20% then you need to extranct 250K objects 
each interval.
–       Geoff: you are needing a new object, not a return to previously 
announced 
state.
–       Geoff: 10-12 new ASs a day, 30 new routes.
–       Rob: use new keys each time, so a return to previous state still looks 
the 
same.
–       Chris: need to understand the size and the speed of change.  Need to 
collect the data or extract from routing history
–       Geoff: troll through RV and RIS and presume it is all certified.
–       Chris: that is what Tim and (RIPE guy from Canada) are interested in 
this 
(Canada guy = andre took) and are trying to prepare for ietf vancouver.
–       Rob: if you are running this with rsync that is binary only and uses 
inodes, 
you might want  to be looking at database.
–       Chris: rsync works (Rob: has a lot of the nice properties) but having 
more 
protocols is OK.  (people troubled by having a large number).  Need 99% 
uptime.  Rob: no SLA for DNS servers.  Brian: there is for root servers.  
Chris: for RIRs, this might be like root servers, but for me, it is more like 
my local dns server
(note slide 28)
Repository Discussion (Cont'd)
•       Chris: stuff you retrieve needs to be high rate, stuff you serve may 
not need to be 
updated at lower rate.  Rob: probably more necessary to be available for people 
to 
retrieve than for me to be able to update quickly
•       Wes: this may change if we are dropping invalids.  I recall calls to 
noc that says 
“you must update your dns server immediately”.  I don’t think that it is valid 
to 
assume those calls won’t come in.  Rob: distinction that your customers will 
want 
a immediate reaction and the chance that the call will come in when you have 
the 
system down for maintenance.
•       Brian: you want your servers available to your neighbors than to the 
people 12 
zones away.  Rob/Chris: don’t buy that – global need for this data.  Everyone 
wants a copy of the worldwide system.
•       Warren: how much data is 1M objects?  Rob: 1K per object, max.  Warren: 
 round 
trip to himalyas is 10 sec.  Rob: worried about pounding primary servers into 
the 
ground.
•       Sriram: if we are doing BrianW/Roque’s proposal from last IETF for 
replay 
protection, you could increase churn significantly.
•       Brian: you could do it that frequently, but we weren’t proposing that 
you do it that 
quickly.  If you are worried about some particular event , you could increase 
rate.
•       RussH: during algorithm rollover, you need another factor of 2 in that 
– multiplier 
is 2 at the peak and builds up and falls off from there.
•       Chris: also think we need to capture the entire size of the repository
•       Rob: CRLS are potentially unbounded because people might have too long 
expiration times, which means nothing ever expires from CRL.  (story of very 
very very large CRLS because of long expiration times)



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

Reply via email to