Shawn,
Don't worry, I asked that because I really didn't understand what you mean.
I thought maybe it could be a new idea.
If I understand you well (just because to read english is something of
non-trivial for me :) ) what you propose is right what I propose with LDD
tables. The 'partitioning' is proposed there (I just haven't used to speak
this term), and the "region field" is proposed, too (by name of LOCAL
column).
I intend to delivery data service using transparent means, i.e., for
end-user, he will have the "feeling" to be accessing a single server,
through any server in cluster he wants.
The only difference, in practice, would be the time response time:
A) If he try to access, in server 1, records "where region = 1" (that's what
really happen in more than 90% of cases), the data is delivered in a short
time (time to read local files).
B) If he try to access, in server 5, records "where region = 1"... well, the
response time will cost something like a network request from server 1 to
server 5.
But my idea, at first, is to manage one partition per server. Do you think
that's not enough?
Fabricio
2005/11/17, SGreen@stripped <SGreen@stripped>:
>
>
> Fabricio Mota <fabricio.mota@stripped> wrote on 11/17/2005 07:24:21 AM:
>
> > Shawn,
> > What do you mean about DDE_PARTITION?
> > Regards,
> > Fabricio
>
> You must forgive me for not THOROUGHLY reading your project spec yet. I
> have read through it twice and I think I understand the scope and purpose of
> your project but I could have missed some of the details. If you did not
> intend to support data partitioning between different servers of the same
> DDE_CLUSTER, I am sorry. It makes sense, though, for "regional" servers to
> partition their data based on some kind of 'region' field. That way if a
> user in a region asks their "local" server for information then that server
> is more likely to have local information on hand skipping the need to ask
> the rest of the cluster. If the data is partitioned using a 'region' field
> and you ask for data that is all from some other region, a peer-to-peer
> connection (region A's server directly queries region B's server) could
> occur minimizing the load on the central, coordinating server.
>
> The use of regional partitions is only an example as real data could be
> logically and physically partitioned any number of ways. I used the DDE_
> prefix to distinguish DDE_CLUSTER partitioning from NDB partitioning. The
> data in a DDE_CLUSTER does not need to be partitioned but it could be
> something to think about as another way of localizing the data.
>
> Shawn Green
> Database Administrator
> Unimin Corporation - Spruce Pine
>
>
--
Sem mais,
Fabricio Mota
Oda Mae Brown - Aprecie sem moderação.
http://www.odamaebrown.com.br
