[Spread-users] Unique message id
yishengsam at yahoo.ca
Mon Aug 12 19:17:43 EDT 2002
yes, I see..., what do you mean "specify the service",
is that to let the application to take care of the
message id when sending messages in this service(as
Jonathan has mentioned)?
The primitive I am interested in is a service that can
let the message to be delivered before the order is
determined, after the definite order is determined, a
correction message only contains the message id will
be delivered to the application for order correction.
With this service, the execution of the group
communication layer can overlap with the application
layer, thus hiding the cost of group communication
behind the application execution, and thereby
achieving improvement without sacrificing performance.
Since the application has to compare the tentative
order and the definite order, a message id is needed
in this service.
For detail description, please refer "Processing
Transactions over Optimistic Atomic Broadcast
Protocols", written by Prof. Betinna Kemme.
Thank you very much for your advice.
--- Yair Amir <yairamir at cnds.jhu.edu> wrote:
> The internal sequence numbers are unique within each
> daemon view in
> the current protocols and their current
> However, you have to remember that this may not have
> any effect on the
> group membership. Therefore, if you use these
> internal counters,
> the daemon membership can change without your group
> changing (for example, a new daemon merged into the
> daemon membership
> but this daemon currently has no processes connected
> with it).
> In this case, your group will NOT see a membership
> event but the internal
> counters will be reset to 1.
> Therefore I think you cannot guarantee that this
> kind of a solution
> will be correct.
> If you could specify the service you are interested
> in then
> there may be a correct (and cleaner) way to
> implement it.
> :) Yair.
--- Jonathan Stanton <jonathan at cnds.jhu.edu> wrote:
> I'd be very interested in what the semantics of the
> broadcast primitive
> you are adding are. Usuallly it is easier to add
> higher level semantics
> on top of Spread's messaging layer, so I'm
> interested in what messaging
> service you found lacking.
> Thanks, and although the sequence numbers might work
> for you now, they
> really are an implementation detail and could change
> in later versions
> (althout not in the immediate future).
> Sam> I am trying to add a special broadcast
> primitive into
> Sam> spread, if the internal sequence numbers are
> Sam> within each 'view' of the membership, they will
> Sam> fine to fit in the things that I am working on,
> Sam> you :).
> Sam> yisheng
> Sam> --- Jonathan Stanton <jonathan at cnds.jhu.edu>
> >> Hi,
> >> The sequence numbers generated by Spread to order
> >> messages are not
> >> accessible to applications, so I'm not sure how
> >> were going to use
> >> them. In actuality the internal sequence numbers
> >> unique within each
> >> 'view' of the membership, but as soon as a daemon
> >> crashes or restarts or
> >> a network partition occurs or recoveres, the
> >> sequence numbers will start
> >> over at 1, so I don't think they will work for
> >> The best way is to just tag your application
> >> messages with a unique
> >> message id such as a tuple of local application
> >> sequence number and
> >> spread connection private name (which is
> >> to be unique until
> >> your application disconnects) or some other
> >> attached to each
> >> instance of your application.
> >> Jonathan
> >> On Fri, Aug 09, 2002 at 02:50:23PM -0400, Sam Liu
> >> wrote:
> >> > Hi,
> >> >
> >> > I would need to attach a message id to each
> >> message
> >> > delivered to the applications. I am considering
> >> use
> >> > the seq no of the last packet of the message as
> >> the
> >> > message id. Is the seq no of the packet
> >> > unique in a specific configuration? Thanks.
> >> >
> >> >
> >> > yisheng
> >> >
> >> > McGill U
> >> > Canada
> >> --
> >> Jonathan R. Stanton jonathan at cs.jhu.edu
> >> Dept. of Computer Science
> >> Johns Hopkins University
> >> _______________________________________________
> >> Spread-users mailing list
> >> Spread-users at lists.spread.org
> Sam> Post your ad for free now!
> Sam> _______________________________________________
> Sam> Spread-users mailing list
> Sam> Spread-users at lists.spread.org
> Spread-users mailing list
> Spread-users at lists.spread.org
Post your ad for free now! http://personals.yahoo.ca
More information about the Spread-users