Re: idea for a geographically distributed database: how best

O codeWarrior έγραψε στις Nov 17, 2005 :

> Sounds like a mis-guided approach to me...
>
> You might want to consider using latitude and longitude or zip codes or
> taking more of a traditional "GIS" approach rather than duplicating data
> across redundant databases.
>
> Another issue is that you end up having to query every database to find
> proximity... Suppose you have 500 "cells" ? You now have to invoke some sort
> of RPC mechanism on 499 other RDBMS, manage 500 user names and passwords,
> retrieve potentially 500 recordsets, merge the data, etc...
>
> Your problems will continue to get more and more complex... You are better
> off with a single RDBMS and a single schema...

Andy, i agree with what codeWarrior says.
But if you are interested in replication, dbmirror is very elegant
(altho not as trendy) simple, and highly customizable replication
solution.

I have heavily modified dbmirror to even support
Asynchronous Conditional row grained
Foreign key dependency Traversal oriented
Lazy Replication!!!
(which ofcourse nobody wants :)

>
>
>
>
> "Andy Ballingall" <andy(at)areyoulocal(dot)co(dot)uk> wrote in message
> news:ECOWS04Mp8nkfZyufzT00010fe4(at)smtp-out4(dot)blueyonder(dot)co(dot)uk(dot)(dot)(dot)
> > Hello,
> >
> > I've got a database for a website which is a variant of the 'show stuff
> > near
> > to me' sort of thing.
> >
> > Rather than host this database on a single server, I have a scheme in mind
> > to break the database up geographically so that each one can run
> > comfortably
> > on a small server, but I'm not sure about the best way of implementing it.
> >
> > Here's the scheme:
> >
> > --------------------------------
> > Imagine that the country is split into an array of square cells.
> > Each cell contains a database that stores information about people who
> > live
> > in the area covered by the cell.
> >
> > There's one problem with this scheme. What happens if you live near the
> > edge
> > of a cell?
> >
> > My solution is that any inserted data which lies near to the edge of cell
> > A
> > is *also* inserted in the database of the relevant neighbouring cell -
> > let's
> > say cell B.
> >
> > Thus, if someone lives in cell B, but close to the border with cell A,
> > they'll see the data that is geographically close to
> > them, even if it lies in cell A.
> >
> > --------------------------------
> >
> > Is this a common pattern?
> >
> > I could, of course, simply find every insert, update and delete in the
> > application and alter the code to explicitly update all the relevant
> > databases, but is there a more elegant way of simply saying: "Do this
> > transaction on both Database A and Database B" monotonically?
> >
> > I've had a look at some replication solutions, but they all seem to
> > involve
> > replicating an entire database. The advantage of my scheme is that if I
> > can
> > distribute my application over large numbers of small servers, I'll end up
> > with more bangs for the buck, and it'll be much easier to manage growth by
> > managing the number of servers, and number of cells hosted on each server.
> >
> > Thanks for any suggestions!
> > Andy Ballingall
> >
> >
> > ---------------------------(end of broadcast)---------------------------
> > TIP 3: Have you checked our extensive FAQ?
> >
> > http://www.postgresql.org/docs/faq
> >
>
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 9: In versions below 8.0, the planner will ignore your desire to
> choose an index scan if your joining column's datatypes do not
> match
>

--
-Achilleus