[cap-talk] impossibility of solving the "coordinated
attack problem" ("generals problem")
devbox at selnet.org
Thu Feb 2 19:07:31 EST 2006
On 02/02/2006, at 15.52, Jed at Webstart wrote:
>At 02:36 PM 2/2/2006, Valerio Bellizzomi wrote:
>>On 02/02/2006, at 14.14, Jed at Webstart wrote:
>>...What it really means is
>> >that it is NEVER possible to come to agreement, to do the
>> >"coordination" and to have both entities agree. I think perhaps
>> >an algorithmic approach is the best way to clarify the fundamental
>> >problem. You really have to put yourself in the position of the
>> >boss or the lieutenant. Let's try laying out possible algorithms:
>>At some point someone makes a decision of trust: he trusts that the
>>messenger will arrive, and so he executes the action anyway.
>I don't think there is an "alternatively".
I am exploring possibilities,
Alternative may be a full-duplex communication with ack: Both sides send
their messenger simultaneaously (by means of a previously agreed schedule),
and each side on receipt of the messenger sends back the same messenger
with their ack. Now the messengers meet in middle route, exchange their ack
and return. Each side receives back the messenger of the other side with
the ack from the other side. Now it is known to both sides that each one
has received the message and sent back an ack, it is also known that the
messengers have met in middle route, and it took only one round-trip and a
half for both to do their job. With the next message each messenger return
to their proper side with a full report of ack on the previous message.
Two messages make one message coordinated.
>>if the communication is simultaneous in both directions,
>It is, or at least it can be, of course, but the decision is made on
>one side only.
>>it just takes one round-trip in both directions.
>To do what? Not come to agreement/coordination.
>>In this case it is a question
>>of timing. If one messenger arrives out of time, the algorithm must
>I don't understand what you're getting at above. In any case
>the last one to send a message is in an ambiguous situation
>and can't be confident/sure that the attack is "coordinated".
>That last message might not arrive. There is no way out
>of this "paradox"/problem.
>cap-talk mailing list
>cap-talk at mail.eros-os.org
More information about the cap-talk