[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [dist-obj] Seven Fallacies Reference



> Not really.  The two-phase/three-phase commit protocol requires an
> assumption about "benign" characteristics of underlying request-reply
> behavior.
> 
> Let's take two-phase commit for instance for the scenario. (3-Phase is
> similar, only marginally improve the reliability)
> The robot acked, the dispatcher got it, and said "commit", the robot didn't
> get the commit, 

>then the dispatcher thought it was done, cross out the task,

why would this happen ? Transactions theory is based on the assumption of a 
reliable network -- and of course transactions would only be used on a 
reliable network for tasks that need to be done reliably ! I do not understand 
why would someone run a transaction system on top of lossy wireless links ?

I do not see how link failures in mobile networks are any different from link 
failures in traditional wired distributed systems ?

> Multi-phase commit protocols do not solve the problem,  fundamentally
> because of the impossibility of consensus when arbitrary link failures are
> present.

Yes, in a system where links are unreliable, the hosts are unreliable and 
there is no synchronization service ---- is that what a transaction based 
mobile system would look like ? I don't think so ...

-U



==========================================================================
To manage your subscription, mailto:dist-obj-help@distributedcoalition.org 
Archives, FAQ, etc.     http://www.distributedcoalition.org/mailing_lists/