[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [Idr] IDR WG Minutes
Rahul:
I've included this change in the updated minutes. Please confirm
I've got it correctly.
Sue
==================================
IDR WG notes 2005.08.04 -(Dave Ward)
-------------------------
Agenda:
Agenda:
1. IDR Documents status update 5 mins
2. Tony Li
draft-li-as-hopcount-01.txt 30 mins
3. Zhang Renhai
draft-zhang-idr-bgp-entire-routes-reflect-01.txt 15 mins
4.
draft-walton-bgp-add-paths-03.txt
draft-walton-bgp-route-oscillation-stop-01.txt
20 mins
5. Nicolas Dubois
draft-dubois-bgp-pm-reqs-01.txt 15 mins
6. Sue Hares
draft-hares-bose-dynamic_as-01.txt 10 mins
7. Chandra Appanna
draft-ietf-idr-bgp-multisession-01.txt 10 mins
8. Gargi
draft-kapoor-nalawade-idr-bgp-ssa-02.txt
draft-nalawade-kapoor-tunnel-safi-03.txt. 15 mins
==================================================
Doc status
Danny McPherson to finish Confed implementation report in one month
Geoff Houston to complete 4-octet AS number
New MIB editor see draft
see ppt
-------------------------
Li AS Hop count Attribute
see ppt for info
Essence: Add TTL to each prefix, decrement at AS boundary Drop route
when TTL hits zero Transitive attribute
Alt: Distribution Lists
List contains ASs that can rx and cannot exclusion
NOTE: NO_EXPORT and AS_HOPCOUNT can be combined
Houston: is NO_EXPORT dropped or continued when combined?
NOTE: Can scope 'anycast'
Houston: when aspath prepending, is that expressed in the HOP_COUNT?
Li: Decremented by one always
Hares: Two more ASPATHS per route? How does scaling work?
Li: In the worst case, when setting up a list you can have an arbitrary
number in incl or excl list?
Hares: Is order semantics true but,
Rekhter: An ASPATH cannot carry all AS in the internet as not enough
space
Li: An avg ASPATH has ~4 AS' in it so, we are going to have on avg
roughly two X
Hares: Why do you say proxy usage is diffficult in Dist List case?
TTL is lighter but, why choose between two? Might have both.
Li: From an implementors POV, it is easier to do one and not two
Rekhter: The increase in number of AS' per route is not a practical
issue
Houston: At the edge, it may work but, topology is dense and we nee
Jason Schiller (UUNET): Scope - useful in subconfeds ... need a
separate counter?
Li: Want to see draft first. Technically what is required is a separate
attribute but, look almost identical
Houston: When comparing the two mechanisms need to look at "what is
going on." Distribution lists require complete knowledge of topology.
Problem is that you cannot insure that they do not leak. HOPCOUNT does
not have 'knowledge' problem but, focuses on accuracy.
Hares: Inclusion requires enumeration but exclusion does not.
Houston: HOPCOUNT allows more specifics w/ a boundary. Advocates
HOPCOUNT
Roque: No productive compare the two. Want to select upstream provider
w/ targeted ext comms. HOPCOUNT might prevent this technique. Most
common TE technique; customers preferring transit AS'
per prefix. Not applicable for transit exit scenario. Two issues and
solutions are orthogonal but, neither solution is devalued.
Chris Morrow (UUNET): proxy means "I can add" and/or "I can delete"
Li: Yes and not violate
Chris; Will it get used if provider ignores or removes
Rekhter: Who is entity that benefits from this? Which SP.
Li: Table size reduction is goal. One SP adds does the work but, gets no
benefits. Altruism or hope of it is a good reason.
Houston: Zhang drafts in GROW show that altruism is bunk instead
self-interest. Most UPDATEs come from lower tier points. Most
computation happens at lower level of peering .. therefore, small folks
get most thrash. If "I" do this, I help my problem ...
therefore altruism is there for the folks that use it and tag it.
David Ward: Affect Hop count?
Li: Not significantly as it would be longer prefixes.
Enke: Can we inject an AS numer as 'signature' of who adds attribute?
May be added to draft. No comments
-------------------------
Zhang Renhai Entire Route Reflect
Comments held until after Walton's draft see ppt for info
-------------------------
Walton Add Paths
Alvaro Retana (Cisco) presenter.
see ppt for info
NOTE: RFC 3107 must be updated
-------------------------
Walton Oscillation Stop
Alvaro Retana (Cisco) presenter
info in same ppt as "Add Paths"
Swallow: What is frequency of oscillation?
Retana: dependent on update timer setting (MIN ADVERT INTERVAL)
Combined questions for both draft authors:
Danny (unaffiliated) - We are increasing and decreasing table size.
Walton's draft is more general and flexible then Zhang's.
Retana: Most of increase of table size is internal to an AS (almost all)
zhang: I don't understand question.
Zhang: Walton's draft is more general. Because you add attribute, do you
think that efficiency of packing will be affected?
Retana: NLRI will be bigger but, no impact on packing.
Gargi (Cisco): Zhang's draft - must specify that you explicitly WD when
a nexthop changes or how to announce new NH? How do you delete the
first NH?
Zhang: If NH changes, just add it. To solve the problem there will taken
offline to email.
Paul ??: If in ECMP case, what AS sequence? How to pass multiple paths
to a legacy speaker?
Retana: Cap advert defines who can learn add_path. Other applications
like EBGP ECMP and propagating multiple paths will have to be addressed.
Retana: Constrained scope.
Roque: Although you say impact of deployment is localized. We need a
better understanding of what impact will be. For each route prefix, path
attrs in Adj RIB Out. Generally announcement, is only when Adj-RIB-Out
changes. What is expectation of propagating changes when inbound updates
would cause outbound content. What are consequences?
Size of Adj-RIB-Out? What is key of Adj-RIB-Out? etc.
Retana: Spec says send on path. Change is that we would have tracking of
two or more paths.
Rekhter: We can have as many ad hoc solutions as we want or a more
general solution. We as a group should decide on a more general
solution. We should look at potential new functionality once we have
more than one path being advertised.
Suggestion is that the authors of three proposals should find common
solution to all problems that WG would like to address. A mailing list
will be set up to discuss this problem and solution. On the IDR mailing
list please identify problems.
[3 proposals are:
draft-bhatia-ecmp-routes-in-bgp-01.txt
draft-zhang-idr-bgp-entire-routes-reflect-00.txt
draft-walton-bgp-add-paths-03.txt]
Rekhter: Do we need requirements document?
Ward: No, absolutely not.
-------------------------
Dubois PM Reqmts
see ppt for info
Rekhter: Send note to list on whether or not we should take as a WG doc.
Dubois: Earlier solution was not meeting requirements and there are many
possible solutions. We need a solution that fits the requirements; we
are not encouraging one solution over another:
Ted Seely (Sprint): Why not weight traffic away from peer or interface?
There are other techniques that don't require change to protocol.
Ruediger Volk: I lose traffic when I do this technique.
Ruediger: The convergence problem I see is due to the previous issue of
config change.
Jason Schiller (UUNET): Is there a hard requirement that this does not
require a config change ... taken to the list.
-------------------------
Hares Dynamic Confed AS
Not requested to be WG item.
see ppt for info
Roque: I would assume that you would want to do restoration in a
seamless way. What is motivation to make failure restoration a visible
outside the confederation?
SKH: Goal is to not drop the peering session. Therefore, the purpose is
policy rerouting wants to 'hang by itself.' If I drop out of the AS
confed, I would have had to drop the session.
Roque: Why not just tunnel back?
SKH: No IGP connectivity and tunnels ran into problems.
SKH: will send scenarios to list
Chandra Appanna (Cisco): I think you need a few more failsafes.
Resend is a MUST, and some mechanisms is nec. that moved did not
suddenly change. Need to resend routes due to policy change potential.
SKH: We have added failsafe. We delayed resend, it would be abnormal to
send all routes if nothing changed. Will send scenario in email.
Gargi (Cisco): Are there any mechanisms in mind for NO resend if ones
moves from one confed to another?
-------------------------
Multisession update - Chandra Appanna
See ppt for info
Added flexibility for any capability code to cause/create multiple
sessions
No comments
-------------------------
Kapoor SSA - Gargi, Scott Wainner
see ppt for info
-------------------------
Kapoor Tunnel SAFI
combined w/ SSA preso, same ppt
Gargi did technical side and changes to docs:
more TLVs to express encap:
MPLS, IPsec, GRE in IPsec, L2TPv3 in IPsec
added application scenarios: MPLS over IP tunnels
Scott presented motivations
Combined comments:
SKH: Can you explain status of this mechanism w/in each of these groups
(on referenced drafts)? What is approved, required, etc. We need to see
something from those chairs that the ref'ed drafts are "blessed."
Scott: Walked through
Gargi: We have chicken and egg to get this done. Encoding first or
formats first?
Intro work on L3VPN list.
Rekhter: Very reasonable model where must have both WGs review dependent
docs.
Bill Fenner: Must work w/ other WG for encoding.
Rekhter: Two possible models - first: semantics or encoding need to be
done in L3VPN
second: L2/L3VPN WGs tell us semantics and IDR produces codepoints
Rahul: There were two problems in the past: 1) Motivation wasn't clear
2) Authors of draft-raggarwa-ppvpn-tunnel-encap and authors of the
drafts presented at this IDR meeting were unable to come up with a joint
proposal. I suggested that since motivations are clearer now, this work
should be discussed again in the L3VPN WG.
Roque: Disagree w/ motivations. IGP can provide reachability. I think
you use BGP as signaling protocol which is not a problem. But,
presentation states that BGP needs to know this information. Please
clarify the layering split.
Rekhter: Can this move to L3VPN
SOLUTION: Joint Last Call in IDR and L3VPN.
-------------------------
SAFI Space Partition
In reply to Jeff Hass email on list.
Fenner to send notes on how to alloc SAFI space.
Proposal: take private use space and turn it into reserved and perhaps
16 or so as private use. Reserved is reserved so we can decide later on
FCFS or otherwise managed.
Are folks using private space? Do we need to skip some?
Rekhter: How long will it take to use up first come first served?
Fenner: We should be able to straighten this out quickly and IANA is
keeping up. 30 days deadline.
Danny McPherson: Needs to be uniform across all WGs and Areas. Make it
IETF aligned.
Roque: I do not want to see vendor specific space go away altogether.
There is much more usage than one case alluded to. Today is only
practical alternative, cannot be locked out. Can it be shown to work
before we assume 30 days?
Fenner: There are publications on these agreements and promises that it
will be done in 30 days. FCFS is you get it now .... no draft, no wait.
Chandra; Cannot get rid of vendor space. AF/SAFI has lost meaning and
is a way to differentiate addrs between two peers. Just need opaque
space that we can use between two peers.
Gargi: Must have vendor space perhaps reduced is ok. Second, IANA is
faster but, it seems to required Routing ADs having lunch w/ IANA.
Rekhter: IANA must have strict 30 days max as rule.
Fenner: We are working on this and the stats show that 30 days is being
met.
_______________________________________________
Idr mailing list
Idr at ietf.org
https://www1.ietf.org/mailman/listinfo/idr