Hi, I am setting up a server that is to be used as a web/sql server. It
has a hardware raid controller, and 6 disks in the array. 2 of the
disks are used as a mirrored RAID 1 volume for the OS (Windows Server
2003).

Now the question is how I should configure the last 4 disks (4 x 300
Gigabyte).

We need to have redundancy, so I have suggested that we set up 2 RAID 1
volumes:
- a volume for database files
- a volume for log and backup files

The person who is responsible for the servers is questioning if this is
the best setup, and has suggested we make 1 RAID 10 volume of 600GB. I
believe we need to have logfiles and database files on separate disks,
so I am disagreeing with this setup.

The server will have about 10 separate databases. Of these, only 2-3
will be updated with data continuously (causing logfile writes). Other
ones will be updated by nightly jobs.

The server is also hosting Analysis Services, and there is one
datawarehouse running. We also need to place these files somewhere.

So given the 4 disks, what is the best configuration?

Sincerely
Svein Terje Gaup

Svein Terje Gaup wrote:
There is no single answer to this question. Spanning a file system over
as many disks as possible is a good thing, especially for random
read/write. So RAID-10 is a good choice.

However with 4 disks you don't get much spanning, you basically are
using a more reliable RAID-0. RAID-10 is similar to RAID-0, but instead
of having your striping done on single disks, it is done on two-disks
sets (mirrors). (Unless you actually have RAID 1/0, which is the
opposite).

This being said, you would gain nothing from storing a transaction log
on a RAID-10 array, since the log is mostly a sequential write. RAID-1
is generally a better idea for logs.

All in all, either you go with a 4-disks RAID-10 or two 2-disks RAID-1.
You *might* get a slightly better performance in the RAID-10 scenario
since your I/O operations on the data will occur on 2 disks (mirrored)
instead of one. But your data and logs will use the same disks, so
there is more contention; it could be better to separate the main data
files and the tempdb+logs. I don't think there is a clear performance
advantage for one setup or the other, unless you can put in more disks
(in which case the RAID-10 is a winner).

One other point: the 4-disks set could be easier to manage, since a
600GB partition is more flexible than two 300GB partitions. As for the
recovery, you have the same level of protection and the same speed of
rebuild, so it's not a big factor. As long as you don't go with RAID-5
(which is not optimal for write operations and has an awful rebuild
speed).

Conclusion: both solutions are fine.

Oh yeah, one more thing. When you build RAID arrays, you should always
pick smaller disks, and have more of them. You get a better performance
since each disk will hold less data (less seek); also having more disks
allows you to build a bigger array, and striping over more disks will
offer a better performance. And the rebuild time is shorter as well.

Regards,
lucm