On Aug 24, Michael Parkin modulated:
Hi Karl,
Sorry, I have a couple of quick questions/observations... Apologies in advance if I have misunderstood anything in your email as sometimes happens.
On 24 Aug 2007, at 11:54, Karl Czajkowski wrote:
<snip/>
... However, due to the non-binding nature of these signals and the assumption of bounded rationality (we can ignore the possibility of bluffing and other psychological strategies), I am not sure these differences are significant. For privacy, one could imagine secure advertisement channels (with authorization) to further blur the boundaries.
Are you saying that 'bluffing and other psychological strategies' can be ignored in the negotiation protocol that is used or in the negotiation strategy of each participant? For me, these two aspects of reaching agreement are orthogonal, just like they are in (for example) contract law.
I am saying that we were not considering such strategies as relevant and therefore did not care to have multi-round negotiation to try to support them. It was assumed that someone who did care about these might propose further specifications that might reuse some of the agreement data models from WS-Agreement.
So far you have the messages: offer, advert, cancel, accept and reject... all of which are clear to me.
But can you explain what the 'offers claiming agreement' step (3a) is? I'm sorry, I don't understand what this means - is it an offer, accept or another message being sent from the initiator to the responder?
We have always assumed a notion of chaining agreements, e.g. part of the context of an agreement could be another agreement. In the fully agreement-based advance reservation case, the idiom is that you first get an 'advance reservation' agreement and you then follow it up with one or more 'claiming' agreements. The claiming agreement binds the reservation of resource capabilities to a particular use. There are other usage scenarios where the claiming/binding occurs in other domain-specific protocols instead of via claiming agreements. So it is another offer, with agreement terms which reference the existing reservation agreement. It is implicitly understood that the reservation agreement gives you the "right" to make these claims, subject to the finite capacity limits of the reservation. In essence, the entire reserve+claim process is the 2PC protocol for the actual domain specific agreement on the consumption of resources. You might note, this does imply a general form of multi-round negotiation, if you were to construct domain-specific terms which could refine or augment an existing agreement with more information. Once you had an agreement which could claim on an existing agreement and add more agreeable terms, you could converge towards the point of maximum/optimal agreement, starting with more conservative terms... however, no effort has been put into formalizing such term languages, outside the conventional advance reservation models we inherited into this work.
<snip/>
p.s. I will add that the fundamental challenge I see for GRAAP-WG and WS-Agreement is that everyone approaches it with the same natural (but incorrect) intuitions about trust, signalling, and consensus.
Can you explain some more about what you think these shared (but incorrect) intuitions are?
My intuition is to a) trust no-one, b) assume messages ('signals') can be lost, duplicated and re-ordered and c) assume that consensus may or not be reached - it depends :-). Are these the same as yours and/or GRAAP-WG's?
You're ahead of the game, I'd say. :-) The typical intuition people have is that consensus can be reached easily, discounting the byzantine faults of the Internet environment, and having a naive understanding of what agreement and commitment really mean. I have lost track of the number of times people have wanted to jump into the most elaborate negotiation strategies they've ever witnessed in real life, without considering: 1. The messaging (un)reliability model you mention. 2. The difference between intelligent and machine agents in terms of their ability to reason and game one another. 3. The difference between contracts signed by legally responsible entities and the automatic decisions of machines. 4. The realities of over-subscription and risk balancing in realistic economic environments (e.g. dealing in real numbered costs and benefits instead of hard absolutes). 5. The desired/expected/acceptable behavior for a large-scale system with many of their proposed agents functioning simultaneously. What I see is that people who focus on this area intensely will all eventually start to see the same general solutions, but these are not very compatible with what the naive user wants. So real deployment faces the challenge of users who do not appreciate the value of these systems nor the theoretical impossibility of what they desire.
Michael.
karl -- Karl Czajkowski Software Architect Univa Corporation 1001 Warrenville Road, Suite 550 Lisle, IL 60532 karlcz@univa.com www.univa.com ________________________________________________________________ www.univa.com. The Leaders of Open Source Cluster and Grid Software