Re: Datawarehouse->(ETL)->Datamart; Where should ETL be physically located?
From: Carlos <miotromailcarlos_at_netscape.net>
Date: Fri, 19 Mar 2010 10:51:30 -0700 (PDT)
Message-ID: <f86b3f6c-eb8d-4eda-b528-7b9b7e44b548_at_y17g2000yqd.googlegroups.com>
On 19 mar, 10:32, guser78 <qazmlp1..._at_rediffmail.com> wrote:
> On Mar 18, 8:03 pm, Carlos <miotromailcar..._at_netscape.net> wrote:
>
> > On Mar 18, 11:29 am, guser78 <qazmlp1..._at_rediffmail.com> wrote:
>
> > > In the 'Dependent Data mart' case, there is an ETL layer to extract
> > > the data from the central Data Warehouse to adapt and Load onto the
> > > 'Data mart'.
>
> > > Assume that the 'Data warehouse' and 'Data mart' run on separate
> > > physical servers. In such a case, where does the ETL layer (which is
> > > used to load onto Datamart from Datawarehouse) fit better? Should it
> > > be part of 'Data warehouse' box, or the 'Datamart' box? Why?
>
> > "Should it be part of 'Data warehouse' box, or the 'Datamart' box?"
>
> > Neither of them. It should be on a separate ETL server with the ETL
> > tools and processes installed on it (and a lot of free HD space too).
>
> I agree with that. But, what do you do when you have the option of
> having only 2 physical servers (1->DWH, 1->Datamart, ETL:where?)?
Date: Fri, 19 Mar 2010 10:51:30 -0700 (PDT)
Message-ID: <f86b3f6c-eb8d-4eda-b528-7b9b7e44b548_at_y17g2000yqd.googlegroups.com>
On 19 mar, 10:32, guser78 <qazmlp1..._at_rediffmail.com> wrote:
> On Mar 18, 8:03 pm, Carlos <miotromailcar..._at_netscape.net> wrote:
>
> > On Mar 18, 11:29 am, guser78 <qazmlp1..._at_rediffmail.com> wrote:
>
> > > In the 'Dependent Data mart' case, there is an ETL layer to extract
> > > the data from the central Data Warehouse to adapt and Load onto the
> > > 'Data mart'.
>
> > > Assume that the 'Data warehouse' and 'Data mart' run on separate
> > > physical servers. In such a case, where does the ETL layer (which is
> > > used to load onto Datamart from Datawarehouse) fit better? Should it
> > > be part of 'Data warehouse' box, or the 'Datamart' box? Why?
>
> > "Should it be part of 'Data warehouse' box, or the 'Datamart' box?"
>
> > Neither of them. It should be on a separate ETL server with the ETL
> > tools and processes installed on it (and a lot of free HD space too).
>
> I agree with that. But, what do you do when you have the option of
> having only 2 physical servers (1->DWH, 1->Datamart, ETL:where?)?
"I agree with that. But, what do you do when you have the option of" having only 2 physical servers (1->DWH, 1->Datamart, ETL:where?)?"
At a first glance, on the DataMart machine, but it depends. I would resume it as "...where the harm is smaller" If you put the ETL on the Datamart machine and the performance decreases dramatically, and you put the ETL on the DW machine and it can cope with the overhead (DW performance, general performance, filesystems, etc) then put it there.
But I would avoid to put it on the DW machine (been there, done that...)
HTH. Cheers.
Carlos. Received on Fri Mar 19 2010 - 12:51:30 CDT