IRTF Open Meeting @ IETF-85 Atlanta, GA, USA

IPR Policy
	added statement about timeliness in the IPR disclosure
	"period should be days or weeks, not months"

Agenda
	State of the IRTF ANRP Awards

    Srikanth Sundaresan, Walter de Donato, Nick Feamster, Renata
    Teixeira, Sam Crawford and Antonio Pescape. Broadband Internet
    Performance: A View From the Gateway. Proc. ACM SIGCOMM,August
    2011, Toronto, Canada.

    Peyman Kazemian, George Varghese and Nick McKeown. Header Space
    Analysis: Static Checking For Networks. Proc. USENIX Symposium on
    Networked Systems Design and Implementation (NSDI), April 2012,
    San Jose, CA, USA.

    4 RGs meeting this week
	NCRG, ICCRG, ICNRG, DTNRG
	
	one proposed RG meeting: SDNRG

    Review the DTNRG on Thu morning with IAB

    Likely SDNRG will be formed by next IETF

Status (as suggested by Lars)

    Active
    	ASRG, CFRG, DTNRG, ICCRG, ICNRG
    Little activity
	   NCRG, NMRG, RRG
    Closing
    	P2PRG, SAMRG

ANRP is handed out by IRTF with ISOC
	20 noms for 2012 IETF/IRTF selection committee 3 prize winners
	for 012:
		Alberto Dainotti (IETF84)
		Srikanth Sundaresan 
		Peyman Kazemian

	Nomination period for 2013 ends Nov 30 http://irtf.org/anrp
    For details, please see IRTF ANRP pages

Broadband Internet Performance: A View from the Gateway
Srikanth Sundaresan

    Q&A:

    Mat Ford (ISOC): Did you look at latency under load in downstream
    direction?

	A: Yes, but its less of an issue as downstream bandwidth is higher.
	
    Lars Eggert: Where did these routers come from - from ISPs?

	A: Yes. They know its a problem.  Not sure what they are doing
	about it.

    Piers O'Hanlon: Did you look at gateways that do prioritization/
    traffic separation?
    
	A: We havent seen that.  With powerboost we do see different
	queues, but not across applications. 
	
	Q: Did routers have remote mgmt configuration? [so they might 
	change the operations]
	
	A: Cable modems have that capability; not so sure about DSL.
	
	Q: But that could be an issue; might just be able to drop queues
	down.

    NIST: Upstream b/w is generally
    
    A: we are seeing cable performance pretty consistent in the US.
    We actually saw more variance in DSL; we speculate that
    this is because DSL is more overprovisioned.

    Al Morton: Could you put up 2nd or 3rd slide?  Comment from Google on
	500ms causes 20% decrease in GOOG search; is that delay a response time?
	
	A: Think they did this by adding the delay and measuring.
	People get impatient. There is an anecdote for AMZN.

    Matt Mathis (GOOG): I'm not into measuring user experience, but...
    We make a lot of $ off clicks.  Typical person clicks on
    an ad, so they have to load fast The marginal rate for a
    page to load is a lot of $.

Header Space Analysis: Static Checking for Networks
Peyman Kazemian (Stanford)

    Q&A:

    Mat Ford: Is your 'header' the whole packet?

	A: Yes it could be.

    Kevin Fall: Are you really comfortable with your use of the term
    'transfer function'?  What about the previous work in SIGCOMM 2007?
	
	A: Yes, we think that is ok A: we weren't aware of the 2007 paper
	until we did the work. There is some similarity but they were 
	focusing on a different problem

    Q: How do you distinguish between say ACL deny versus black hole
    type of situations?
    
	A: You take a snapshot of all the rules, you do the analysis
	(static) [the rest lost]

    Sharam: On spanning tree issue: The example is a simple
    misconfiguration.
    
	A: This isn't a protocol; it.s a framework for finding issues.
	
	Q: There are many ways of doing that - do we need something
	else?
	
	Q: Have you looked at scalability of this?  
	
	A: No, we didn't have access of provider networks
	
	Q: You don't have the info on others' networks?
	
	A: You'd monitor your own network
	
	Q: So your thing seems to work on smaller networks?
	
	A: It relies on having access to state of the network; can't do the whole
	thing.

    C Ram: You mentioned the inverse should exist, right?
    
    A: Yes.
    
    Q: Even if your net is static, you might have say random routing
	right? (e.g., for TE) How does this work?

	A: Two answers: I presented a deterministic approach in my analysis.
    Usually we simply invert the 'arrows' of the map you could
    expand the notion of transfer function with probabilities.

    Lars Eggert: Computational complexity - how does it affect the
    runtime?
    
	Q: So you have analysis of this?
	
	A: It tries to analyze things in the same equivalency class all at once
    It's proportional to O(r^2) where r is # of rules and O(d)
    where d is network diameter.


completion early