Re: 9i streams vs triggers

"servant" <mjohns1_at_uhc.com> wrote in message news:avn1qs$nm2$1_at_gabriel.uhc.com...
> You may be right. I knew Oracle was different, but it looks like I need
to
> get up to speed.
>
> BTW, I searched for "MULTIVERSIONING." and got no hits. I'll keep
looking.
>

Yeah, "multiversioning" is not a term that's frequently used. Look for the term like "read consistency" instead.

IMHO, the phrase 'multiversioning' should be reserved for 9i's Workspace Management feature, where a table really can have multiple versions of the same row physically stored in the same table. But I suspect I'm just whistling in the wind here.

Read-consistency means that Oracle knocks up a copy of a block of data and rolls it back to a prior time so that you see what the data *used* to look like before other people started making changes to it. It's this mechanism that means that my writes at 10:03am don't block your read which started at 10.00am and will chug along until 10:30am. When your read at 10:06am encounters the block I changed at 10:03, it simply copies the block and rolls it back so it looks like it did at 10:00. The copy is known as a 'read consistent image' of the data, and is never flushed to disk -it's purely a memory thing.

Regards
HJR
>
>
> "DA Morgan" <damorgan_at_exesolutions.com> wrote in message
> news:3E1EFA12.C88758AE_at_exesolutions.com...
> > servant wrote:
> >
> > > Sybrand,
> > >
> > > I am having trouble understanding how I can design my database to have
> data
> > > that is updated and selected simultaneously without contention. The
> update
> > > will lock the row and block the select and vice versa, right? Can you
> > > expand . . .?
> > >
> > > What I meant by replication is logical hot standby so I think we agree
> > > there. That solves the contention issue but does not address the
> complexity
> > > issue. I agree that this is not a pretty solution, but I need a way
to
> take
> > > the data from one data model, massage it with existing code (so I
don't
> have
> > > to develop/maintain it in two places), and write it to another
> "reporting"
> > > database at near real time. Those requirements don't leave me with
any
> > > pretty solutions that I can think of. I am no guru, though, so I
> certainly
> > > welcome any suggestions.
> > >
> > > RAC, as I understand it, will give me scalability which, again, solves
> the
> > > performance/contention issue but does not address the complexity
issue.
> > >
> > > You mentioned "Advanced Replication". I need to find out what that is
> (I am
> > > new to Oracle but learning fast). If it can call Java methods it just
> might
> > > work. Can you offer opinions on this versus triggers or streams?
> > >
> > > I'm off to tahiti (.oracle.com) . . .
> > >
> > > "Sybrand Bakker" <gooiditweg_at_nospam.demon.nl> wrote in message
> > > news:77pr1v42q0kef48ad29cacqjvsinbqp6vl_at_4ax.com...
> > > > On Thu, 9 Jan 2003 13:06:43 -0600, "servant" <mjohns1_at_uhc.com>
wrote:
> > > >
> > > > >I have been tasked with developing a reporting architecture for an
> > > > >application that has been developed over the last few months. I
have
> > > been
> > > > >told there are 2 main concerns: performance (contention) and
> complexity
> > > of
> > > > >the data model. I am thinking of using Oracle's replication to
> maintain
> > > a
> > > > >(near) real-time replica. That (basically) resolves the contention
> > > issue.
> > > > >The complexity issue is more difficult.
> > > > >
> > > > >I am thinking about using 1) streams against the redo logs or 2)
> triggers
> > > to
> > > > >pull the data from the replica, apply business logic to it, and
> populate
> > > a
> > > > >"logical" data model on which reports can be generated. I would
like
> to
> > > use
> > > > >the same abstraction layer to go from the physical to the logical
> data
> > > model
> > > > >as the application interface uses (Java objects with EJBs). In
other
> > > words,
> > > > >I need the code inside the streams or triggers to call Java methods
> (ie:
> > > > >loadCustomerData(cust_id)) which would get the data and apply any
> > > business
> > > > >logic and then populate the "logical" database.
> > > > >
> > > > >In essence, I think I am being asked to create a data warehouse,
but
> I
> > > > >haven't convinced myself of that yet.
> > > > >
> > > > >Anyway, here is my question: Does anyone have an opinion on which
> method
> > > > >(streams or triggers) would tax my replica database more--from a
CPU,
> > > > >network, and/or contention perspective? Any opinions from other
> > > > >perspectives which option is better (maintenance, etc).
> > > > >
> > > > >
> > > >
> > > > First of all I would like to question the need of a replica database
> > > > for reporting purposes. As for contention, this is simply not an
> > > > issue, if the application has been designed properly. Your post has
a
> > > > distinct smell of finding a 'hardware solution' read 'workaround'
for
> > > > what is in reality a software design problem.
> > > > If you really think you need a hardware solution you could just as
> > > > well consider RAC, to have multiple servers running on the same
> > > > instance.
> > > > As for your option 1) you seem to be trying to re-invent the standby
> > > > database technology . One simple advice: don't. You would never get
it
> > > > work, and provided your organisation has still a bag of $$ floating
> > > > arond: get Oracle Enterprise Edition, set up a standby database
which
> > > > will use the redolog to keep the standby database up-to-date and
start
> > > > living
> > > > As for your option 2) this seems like yet another solution which is
> > > > going to be designed in a bike-shed. Replication can be accomplished
> > > > by setting up the Advanced Replication option, and designing
> > > > 'home-grown' solutions will definitely make you bald sooner and cost
> > > > more.
> > > >
> > > > Regards
> > > >
> > > >
> > > > Sybrand Bakker, Senior Oracle DBA
> > > >
> > > > To reply remove -verwijderdit from my e-mail address
> >
> > When you write: "I am having trouble understanding how I can design my
> database
> > to have data
> > that is updated and selected simultaneously without contention." it
leads
> me to
> > think that you are
> > completely unaware of Oracle architecture.
> >
> > In Oracle reads don't block writes and writes don't block reads. If you
> are
> > designing, or your
> > architect is designing based on knowledge of some other RDBMS it is time
> to
> > stop work on head
> > for the Oracle concepts and architecture documentation.
> >
> > Daniel Morgan
> >
>
>
Received on Fri Jan 10 2003 - 15:24:03 CST