Re: To RAID or not to RAID ??

Sudheer Marisetti wrote:

> Hi,
>
> I know there were many discussions and threads on this topic. My
> apologies for reigniting the topic. I am looking for people who have
>
> used RAID for building data warehouse.
>
> We are designing and building a data warehouse about the size of
> 40-60
> GB on Oracle 7.3.2. Based on my research on this topic I believe
> RAID-5
> is most suitable in this case. But the hardware vendor providing the
>
> machine has installed RAID-1 (disk mirroring) and vehemently opposes
>
> changing this to RAID-5 on the basis that their experience tells
> them
> that RAID-5 gives a very poor performance for any database
> applications,
> including data warehousing. Also I received recommendation on using
> a
> combination of RAID-0+1 (stripping and mirroring).
>
> I would really appreciate if any of you could share your experiences
>
> with RAID-1 or RAID-5 or RAID-0+1 while building or using data
> warehouses or any database. Here is the set up we have:
>
> - HP9000/K260 Server
> - PA 8000 180 MHz processor (2 of them)
> - HP-UX 10.2
> - 384 MB memory
> - Database: Oracle 7.3.2

 Well, RAID 5 - when implemented in *hardware* (don't even think of using software based RAID or striping, IMHO) - will *potentially* give you better read performance at the expense of write performance. You will also get a degree of fault-tolerance for far lower cost than complete mirrored drive sets.

However, as you are building a database, and not just a general purpose file server, you are going to know a lot more about your data and its dynamics than any RAID 5 controller would ever know. Thus you want to be able to partition your data, indexes, binaries, etc. across drives of *your* choice. You can then tune and do some analysis, then partition some more until you get the best balance. This would thus lead you more towards a mirrored solution, assuming you want the fault-tolerance.

A combination of striping your data in localised sets can give you even better performance, so RAID 0+1 seems to be the preferred solution for most data warehouses - and this is in fact what Oracle recommend on their VLDB database courses. This also opens up the option of breaking parts of a RAID set (esp. if you have 3-way (or more) mirroring) to perform backups.

At the end of the day, the choice is yours:

• How much money have you got to spend?
• How much fault-tolerance do you require?
• What are your performance requirements?
• How flexible does the system need to be?
• What options are available for your server?
• How much expansion do you need (and at what cost)?
• etc., etc., etc.

There are a couple of Oracle books, Oracle Warehousing and Backup and Recovery Manual (both Oracle press) which overview the subject of RAID. You might want to use these as a start, before consulting Oracle (or some other independent consultancy) to give you a more accurate assessment of your system and the options open to you.

I have systems running both RAID 5 and some running RAID 0+1, and they all suit their own individual goals. I also have systems running niether type of RAID.

You really will have to look very closely at your own personal requirements, but I do think you should look beyond just RAID 5.

Hope the above helps.

Steve Phelan. Received on Thu Apr 24 1997 - 00:00:00 CDT