Re: [alto] General comment about draft-penno-alto-protocol-03.txt

[Obi Akonjang <[e-mail]>]

Hi Martin,

We do understand your concerns about the inclusion of the Oracle
approach in the currently merged ALTO proposal and would like to address
them the best we can.

First, though the approaches may differ, we're all trying to resolve a
common problem and do understand the need for a common protocol. We thus
decided to work with the other groups together to achieve this goal.

Second, the Oracle (likewise P4P) MUST NOT calculate all information
on-the-fly, since it uses and maintains a back-end database that gets
updated at pre-determined intervals or when adjusting to sudden
changes. 

Third, the Oracle's ranking approach does not need topology maps, which
immensely simplifies interactions between the server and potential
clients. This (IMHO) adds more weight to the protocol's heaviness.


It would be good if you could explain what you mean by the Oracle
turning the argument of caching upside down. We instead see lots of
potentials and common grounds between the Oracle approach and caches.


Hope this clarifies your concerns about the Oracle merge :-)

Kind regards,
Obi




On Mon, 2009-08-10 at 10:31 +0200, Martin Stiemerling wrote:
> Hi all,
> 
> I have the severe doubt that the integration of all existing ALTO protocol proposals into one protocol is a good idea at all, after spending some cycles on the P4P draft over the weekend.
> 
> The beauty of P4P seems to be (IMHO) that the server itself can calculate all information beforehand based on the PIDs. I.e., there is actually no calculation needed during the runtime, leaving the load on the server to serving the HTTP GET requests (we know how to handle this by today for a large request volume). 
> 
> However, there is a shift in protocol heaviness, by introducing the oracle approach to the protocol proposal. The oracle requires calculations during the runtime, i.e., the server needs to serve the requests and do extra cycles on calculating the guidance. The oracle also turns the argument of caching upside down, as the responses are not meaningful to caching anymore.
> 
> I would leave the P4P proposal without the oracle or is the goal to generate a protocol that can do all, but with too many caveats?
> 
> Thanks,
> 
>   Martin
> 
> [e-mail]
> 
> NEC Laboratories Europe - Network Research Division
> NEC Europe Limited | Registered Office: NEC House, 1 Victoria Road, London W3 6BL | Registered in England 2832014 
> 
> > -----Original Message-----
> > From: [e-mail] [mailto:[e-mail]] On Behalf Of
> > Y. Richard Yang
> > Sent: Tuesday, July 28, 2009 5:10 PM
> > To: Martin Stiemerling
> > Cc: alto
> > Subject: Re: [alto] General comment about draft-penno-alto-protocol-
> > 03.txt
> > 
> > Hi Martin,
> > 
> > I really need to say that this is an excellent, fundamental comment!
> > 
> > Let me start with your second comment ("Second, I couldn't really find
> > out what parts are from the various other incorporated protocols ...").
> > In this sense, we have succeeded in achieving seamless integration :-)
> >  From the start,  it is clear to multiple authors of multiple proposals
> > that there is not an inherent conflict of the many proposals. There is
> > much synergy, in terms of service model, among many proposals already
> > (I
> > remember saying this at last IETF :-). The presentation slides listed
> > key features of contributing proposals, and you can see large overlap.
> > During our merge, we made an effort to *not* do a non-overlapping
> > partition of ideas to different previous proposals.
> > As someone working on P4P, I would say that the current design has
> > improved substantially over original P4P both in terms of service model
> > and protocol. Designers of other contributing proposals sure can share
> > how they think the current design improves over their previous
> > proposals. Let me list some key *protocol/implementation* features that
> > substantially improved over the very original P4P design and thus show
> > the benefits. Of course, this list is my personal opinion and is by no
> > means complete:
> > 
> > - A general, flexible protocol framework (e.g., server capability query
> > as some kind of negotiation, generic endpoint property, reverse map)
> > for
> > extension
> > - XML protocol encoding for flexibility
> > - HTTP caching
> > - Possibility of batched distribution of ALTO info, for example, as a
> > file
> > 
> > As a general rule, I feel that protocol design should be flexible,
> > extensible, and consider implementation efficiency. Many good ideas of
> > contributing proposals are reflected in the current, merged design,
> > IMHO :-)
> > 
> > In terms of service model, you raised an excellent point as well. I see
> > two basic service models in terms of providing path rating service:
> > 
> > - Network location map (clustering) to (implicitly) indicate
> > coarse-grained proximity/preference (i.e., if same location, then it is
> > preferred); preferred by some ISPs such as China Telecom as a starting
> > point;
> > 
> > - Explicit rating of communication patterns: even here it can provide
> > info in two ways (batched cost map or ranking list).
> > 
> > It is amazing to see that the preceding service models integrated quite
> > cleanly in the current design without loss or substantial complexity.
> > So
> > I worry less about losing functionality, but more about providing
> > guidelines to ISPs on how to use the ALTO service in their scenarios.
> > Clearly a use-case document can be helpful.
> > 
> > Again, thank you so much for the excellent comment!
> > 
> > Richard
> > 
> > Martin Stiemerling wrote:
> > > Hi all,
> > >
> > > I have a general comment about draft-penno-alto-protocol-03.txt after
> > reading it and some more  comments:
> > >
> > > The draft tries to incorporates now all other in parallel existing
> > approaches (i.e., P4P, ALTO Info Export, Query/Response, ATTP, and
> > Proxidor) but without any major discussion about the usefulness of
> > this. All approaches have their pros and cons, and they are quite
> > different ways of tackling with ALTO. The discussion about the various
> > proposal just started and IMHO it was not clear which of the many does
> > actually address the general challenges of ALTO.
> > >
> > > This means (taking some as example):
> > > - P4P addresses the challenges out of the P4P project, which the
> > specific mechanics of Comcast and Pando. I do see the value of P4P, but
> > it is one case how you could do it, i.e., I'm not saying that this is
> > bad. I like the approach very much. However, I do not (yet) see the
> > evidence that P4P works for tracker-less P2P and in other deployments,
> > even though most of the people believe this.
> > >
> > > - Proxidor has a different view IMHO to ALTO, i.e., very operator
> > driven. By solely integrating this, there might be a loss of
> > functionality, e.g., Proxidor works also tracker-less p2p.
> > >
> > > - ATTP was a bit orthogonal to the other approaches and less to do
> > with ALTO (even though it is related).
> > >
> > >
> > > Second, I couldn't really find out what parts are from the various
> > other incorporated protocols, other than this looks as an evolved P4P
> > proposal without explicitly saying what the benefits of the merger from
> > the various protocols are.
> > >
> > > The protocol claims to "At the same time, it introduces additional
> > techniques  to address potential scalability and privacy issues."
> > (first paragraph, 2nd sentence of Section 2). However, I'm clueless
> > after reading what these techniques are and why they're not discussed
> > in the security section (privacy as term pops up in this sentence, and
> > nowhere else).
> > >
> > > Thanks,
> > >
> > >   Martin
> > >
> > >
> > > [e-mail]
> > >
> > > NEC Laboratories Europe - Network Research Division
> > > NEC Europe Limited | Registered Office: NEC House, 1 Victoria Road,
> > London W3 6BL | Registered in England 2832014
> > >
> > > _______________________________________________
> > > alto mailing list
> > > [e-mail]
> > > https://www.ietf.org/mailman/listinfo/alto
> > >
> > 
> > _______________________________________________
> > alto mailing list
> > [e-mail]
> > https://www.ietf.org/mailman/listinfo/alto
> _______________________________________________
> alto mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/alto