Re: Hardware config for Data Warehousing ??

In Article<335BCCAB.872_at_bellatlantic.net>, <marisett_at_bellatlantic.net> writes:  

>
> We are in the process of evaluating hardware and software for a medium
> sized (between 20-40 Gb) data warehouse. We have decided to use Oracle
> 7.3.2 as the back-end and Gentia (MDD) as the front end. What I am
> looking for is the following:
>
> (1) What UNIX platform is most suited for our needs? The mgt. has
> decided on an HP platform. So do you know of any specific model that
> would suit Data Warehouse applications.

It would depend on the number of users that are going to access the warehouse, the rate of growth you are expecting, the amount of data to be loaded into the DW and the level of summaries you will use. Furthermore the Oracle Parallel Query option only starts working with 3 or more CPU's. The K'series is probably your best bet - both from a raw power viewpoint as well as from an expandability viewpoint. For some test look at www.tpc.org for the TPC-D benchmark results, that should give you an idea of the direction the different vendors are going.

> (2) What type of disk configuration is suitable - RAID-0/3/5? I have a
> vendor who keeps insisting that RAID-0 is optimal for DW applications.
> If RAID-5 which vendor is good? (EMC, etc.)

Raid 0 - striping is probably the fastest. Caution: Take into consideration the availabilty expectations within your company. Once the DW gets accepted and used, the users will hang you if their information is unavailable so look at a striping/mirroring (RAID 0+1) strategy. If you start mirroring different options becomes available like using one half of the mirror for the loading of data while the other half stays available to user queries.

> Does anyone have experience with RAID-S configuration that EMC tauts?

RAID-S - is the EMC equivalent to RAID-5. Unfortunately you can only have Mirrored or RAID-S on EMC. You can use Unix utilities to mix and match though eg. stripe on Unix level - mirror on EMC level. Depending on the rate of growth you are expecting EMC would probably be an overkill for a 20 GB datawarehouse.

> (3) How many processors would be optimal from your experience? I know
> this applicationa and usage dependent but I would appreciate any
> pointers to research more (like web pages, white papers, etc.).
>

See 1)

> Also any pointers white papers or web pages on any of the above
> questions would be highly appreciated.

Try Larry Greenfield's page: http://pwp.starnetinc.com/larryg - its a Mine of information :).

Cheers
Lourens
lous_at_hotmail.com Received on Tue Apr 22 1997 - 00:00:00 CDT