report from the alias BOF

[Sally Floyd <floyd@icir.org>]

I am resending my notes from the alias BOF below.
I just had a talk with Mark Handley about this,
and plan to send a longer email after I have read the current
version of the proposed charter.

- Sally

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

Access Link Intermediaries Assisting Services BOF (alias)

Disclaimer:  I did not read the mailing list.

Revised charter was posted to the mailing list yesterday.

Summary of charter changes:
* Definition of transport intermediaries.
* Enumerate interactions: discovery, request, control, and information reporting.
* Clarification that security association is optional.
* Expand survey of existing intermediaries, to future intermediaries.
* Define "in-scope transport intermediary services" after working group starts.
* Update deliverables and milestones.

Bernard: Trouble with Triggers.
Experience with triggers in DNA bof and in transport
draft-ietf-dhc-ipv4-dna-04.txt
"Hint" might be more clear.  Hints can be wrong or misleading.
Bad Hint, No Donut:  
  - SSID as a hint in 802.11
  - link down - it could be a momentary connectivity loss
  - link up - when is link usable for IP connectivity?
    You need a layer-3 proof of reachability to be sure that the link is up.
    Stories of problems with link up in 802.11.
Importance of Robustness:
  - Can the host verify the hint?  recover if the hint is wrong?
Questions:
  - Do we know if a hint is strong or weak?  Verification?  Recovery?
Strong vs. Weak Hints:
  - Probability of being misleading.

Discussion: 
* Charity begins at home:  not just blaming the link layer.
* Spencer: Appreciation.  There is an ugly layer-3 version of misleading hints,
  e.g., about firewalls and such.
* Problems with other link-level technologies?
* Steve Bellovin:  Use hint in a way that might improve your performance *if*
  the hint is right, but keep robustness.
* ICMP host/port unreachable messages.  They should be hints.

Kevin, Observations:
* The term "access links": this is an unclear term.
* Where is state maintained: 
  - security associations are stateful.
  - some triggers are stateless.
* How is signalling delivered:   
  - ICMP vs. ECN model (or direct vs. indirect signalling).
* Things to do with ICMP:
  - ICMP is a stateless, standard signalling protocol.
  - Not much protection against on-path adversaries.  Do we need to fix ICMP?
* Strawperson ICMP messages (just to think about, not advocating):
  - "Middlebox on path" message? 
  - "Radical link change" message?
  - "Wrong source address" message?

Jon P.:
Name?

Name suggested by chairs:
LIES: Link Intermediaries for Enhanced Services

Melinda:
Who initiates the dialog.  If network devices initiate the dialog, that
  might make it easy to be lied to.

Michael Richardson:  
What is the consensual model for ICMP messages from middleboxes?

Thomas Woo, Transport intermediary services:
An initial attempt at a survey of transport intermediary services.
* TCP PEP.
  - 3GIX (high speed wireless circuit data) dynamically allocates
    supplemental channels, based on queue backlog
  - HDR (high speed wireless packet data) uses channel-based
    scheduling.
  - TCP doesn't work well for channels with a highly variable
    bandwidth-delay product.
* One approach: regulate flow of acks back to the source.
* IPSec ESP tunnels hide TCP header from intermediary

* Discussion: 
  - What is being signalled here?
  - Melinda:  Why aren't you using transport-friendly ESP?
    Answer:  This is not excluded as an answer.
    Melinda:  Then why does this require trigtran/alias?
    ?:  Is this in scope for this meeting?  This isn't isolated to access links,
        this is a generic problem.
    Jon P.:  We are still trying to understand the scope of the WG,
      particular to not be restricted to access links.
    ?:  The description of the problem is in scope.  He thinks there exists
      solutions that don't require transport-friendly ESP.
    Answer:  This is an example, we need common solutions to a range of examples.
    JTW:  For finding generic examples, we don't want to create intermediaries
      where they aren't needed.

* Header compression.
* Common issues.

---------------------------------------------------------------
Documents:
Securely Enabling Intermediary-based Transport Services
http://www.ietf.org/internet-drafts/draft-blumenthal-intermediary-transport-01.txt