RAID - in a server

[atomiser]

mornin all.

im after a bit of advice for common practice really in server systems and the implementation of RAID.

i've found quite a bit of info about the types of RAID systems available but this hasn't really helped me make an informed decision about what to go for.

what im thinkin of doing is having 2 servers. each server will have 4 hard drives.

2 in a RAID array which will hold the operating system
2 in another, seperate, RAID array which will hold the 'user data'

the 2nd server will be a mirror image of the 1st server...(hopefully) through clustering.

whilst improvements in the read/write speed of the drives would be nice, its data resilience and integrity i am after.

would someone be able to point me in the correct direction?

many thanks in advance.

[donny_paycheck]

Using two drives per array will limit your choices to either RAID 0 or RAID 1. Check the sticky in this forum for info about the respective advantages and disadvantages of each. I'll briefly summarize here.

RAID 0 is for increased data throughput, but since it splits your data across two or more disks, all of your data is toast if one of the array drives augers in. RAID 1 mirrors your data across two or more disks, meaning that a RAID 1 array can only be as large as its smallest single disk. Since fault tolerance is your utmost priority that leaves RAID 1 as your only option.

My only additional suggestion is that you possibly re-think how you will allocate your drives. RAID 0 isn't what you are looking for and RAID 1 will waste you a lot of valueable storage space. Being that you are prepared to spend the money on two entire servers, my suggestion is that you go the extra mile for a quality RAID controller and implement a RAID 5. With 3 or more drives in one array you can stripe them in RAID 5 for fault tolerance with reasonable performance, as well as a much more efficient use of space. For servers, RAID 5 is pretty much the standard.

Also:

http://webpages.charter.net/rylandpage/estradawelcome.jpg

[Xaotic]

Mirroring the servers isn't strictly necessary, unless the data needs five nines access. Clustering off separate arrays can give very high overhead, even with background replication using a backbone assuming usage where data is frequently changed. Having good easily restorable backups and if necessary hardware spares is usually good enough for most businesses. Several questions would give a better idea of the proposed structure and usage needs.

First, which interface do you plan on using? IDE tends to be less expensive, a bit less secure data storage(due to lack of CRCs on data writes) and generally has less versatile controllers(though there are some exceptions. SCSI will be largely the reverse and allows for many more devices per channel.

Second, what are the uses of the servers and how heavily will they be loaded? For example low usage file servers require very little in the way of resources, while client/server application and database servers can require tremendous levels of hardware and optimization. This will also effect backbone replication constraints if using nonclustered storage arrays.

Third, what budget are you looking at? The budget will constrain hardware choices and pose limits on the systems themselves. Admittedly, the best way, for both speed and reliability, would probably be a fibre channel redundant SAN with redunancy throughout, but to properly implement a system on this magnitude is exceedingly expensive. Another choice would be a shared SCSI expansion that supports clustered storage and RAID-1 internal OS arrays on the servers, with hot spares on all arrays. Less expensive, would be RAID 0+1, or a similar level, on internal arrays within the servers with hot spares on each server.

Finally, what is your backup and disater recovery strategy? RAID is not a back up and cannot be considered as secure storage. It's still subject to physical and software losses and a backup will save your tail. I've seen arrays crash nonrecoverably within OS that the controller can still read. If both of the servers were in the same location a physical disaster could easily destroy both, including lightning strikes. For this reason make sure there is offsite storage of backups as well.