[rrg] [GROW] Operational experience with cache based mapping ID

[PAPADIMITRIOU Dimitri]

bill, all,

a couple of comments here below

- on the scope: one may to limit to pull systems but the acquisition
techniques should also imho consider hybrid push-pull system

- on the structure/content:

  . section 2: i agree that a section should be included that provides
for an overview of the pull-system(s) and this before detailing
operational experience. but section 2 and 3 shall be reverted as metrics
are common (a section 3.1 on the metrics i detailed should be
considered)

  . section 2: segmentation of successful/unsuccessful is a matter of
choice but i would not qualify them as successful/unsuccessful at that
stage but keep analytical evaluation in section 4 

  . section 4: deals with characteristics and observations from the
field but there is a need to tackle the actual performance of these
systems. it is also worthwhile detailing performance wrt to the
different replacement algorithms (LRU,LFU,LRR)

thanks,
-dimitri.
ps: i missed your e-mail you sent on Dec.3rd

> -----Original Message-----
> From: rrg-bounces at irtf.org [mailto:rrg-bounces at irtf.org] On 
> Behalf Of William Herrin
> Sent: Friday, December 19, 2008 4:26 AM
> To: Peter Schoenmaker
> Cc: grow at ietf.org; Routing Research Group Mailing List
> Subject: Re: [rrg] [GROW] Operational experience with cache 
> based mapping ID
> 
> On Thu, Dec 18, 2008 at 6:46 PM, Peter Schoenmaker 
> <pds at lugs.com> wrote:
> > It seems there is interest in this topic.  But to move 
> forward we need
> > volunteer(s) to act as the editor.
> >
> > Do we have more specifics about the protocols to be 
> studied, and what the
> > paper would contain?
> 
> Hi Peter,
> 
> Apologies for my delinquency on this.
> 
> First, based on the feedback I've heard, I suggest keeping the
> discussion about operational experience with pull-cache mapping
> systems on GROW. GROW's traffic is pretty light right now. RRG has
> gotten active as of late and I suspect there are a number of folks on
> GROW who are interested in this subject but don't want to wade through
> all the RRG messages.
> 
> Folks on RRG who want to subscribe to GROW to follow and participate
> in this discussion can do so at:
> https://www1.ietf.org/mailman/listinfo/grow . I will cut the RRG list
> from future messages on this topic.
> 
> 
> 
> For the document, I was thinking along the lines of:
> 
> 1. Introduction: We're looking to learn the difference between
> successful and unsuccessful pull-cache mapping systems so that as we
> consider using pull-cache for Internet routing we can avoid proposals
> with the characteristics of an unsuccessful pull-cache map.
> 
> 2. Interesting pull-cache systems seen in operations (brief summaries)
> 
> 2a. At least two successful pull-cache systems. DNS and CPU 
> ram cache perhaps?
> 
> 2b. At least four superseded (unsuccessful) pull-cache systems. Cisco
> fast switching (cache single IP address destinations in a hash table).
> What else?
> 
> 3. Metrics for each system
> 
> 3a. Estimated upper bound of entries in the system (DNS has billions
> of entries. A CPU cache can handle millions of ram pages).
> 
> 3b. Estimated upper bound of active entries in typical cache. (a few
> hundred thousand names at once in a DNS cache. Around hundred thousand
> pages in a CPU ram cache).
> 
> 3c. Typical entries served per second (a few thousand with DNS. A
> couple billion with the CPU cache)
> 
> 3d. Typical delays on a cache miss (a couple hundred ns with the CPU
> cache. 50 to 2000 ms with the DNS)
> 
> 3e. Replacement algorithm
> 
> 3f. Etc.
> 
> 4. Commonalities
> 
> 4a. Characteristics common to both successful and unsuccessful
> pull-cache systems
> 
> 4b. Characteristics observed primarily in unsuccessful 
> pull-cache systems
> 
> 4c. Characteristics observed primarily in successful 
> pull-cache systems
> 
> 5. Conclusions: what to look for and what to avoid in proposed
> pull-cache systems for routing.
> 
> 
> Thoughts? Any suggestions for what interesting but failed pull-cache
> information systems we can glean insight from?
> 
> Regards,
> Bill Herrin
> 
> 
> p.s. Dimitri, did you receive my 12/2 response to your 11/29 email?
> Wondering if we have a rogue spam filter somewhere between us.
> 
> -- 
> William D. Herrin ................ herrin at dirtside.com  bill at herrin.us
> 3005 Crane Dr. ...................... Web: <http://bill.herrin.us/>
> Falls Church, VA 22042-3004
> _______________________________________________
> rrg mailing list
> rrg at irtf.org
> https://www.irtf.org/mailman/listinfo/rrg
>