RE: Is State-based LDUP needed?

Steven,

We're in agreement then.

Regards,
Tim Hahn

Internet: [email protected]
Internal: Timothy Hahn/Endicott/IBM@IBMUS or IBMUSM00(HAHNT)
phone: 607.752.6388     tie-line: 8/852.6388
fax: 607.752.3681






"Steven Legg" <[email protected]>
01/20/2002 06:25 PM
Please respond to steven.legg

 
        To:     Timothy Hahn/Endicott/IBM@IBMUS
        cc:     <[email protected]>
        Subject:        RE: Is State-based LDUP needed?

 


Hi Tim,

Timothy Hahn wrote:
> Steven,
>
> I followed you all the way until the end of your note.
>
> I am of the opinion that updates sent during a replication session be 
sent
> in CSN order.  While this may cause "up-front" work for "state-based"
> implementations, I believe that it allows consumer servers to make the
most
> progress possible - even if the communications link is lost before the
> replication session can complete.  URP will take care of the cases where
> the consumer sees a CSN in an update that is "older" when sent by a
> different path through the network.
>
> I couldn't tell in your last two paragraphs what you were asserting be
> done with respect to update send (and thus expected receive) order.

I wasn't so much making a specific proposal as pointing out the technical
issues. Given that there will be few or no state-based implementations
then I am of the opinion that we should make life easier all round by
requiring that, with respect to each replica ID, updates must be sent
in CSN order.

We don't want to require that updates be sent in global CSN order since
that would require even the log-based implementations to sort updates
before sending them.

Regards,
Steven