Very defensive stance towards an opinion...
Originally Posted by bossman
My point though was that anything held on that raid, if the array dies, you lose the lot there and then, meaning you have to go through the process of replacing the dead drive, rebuilding the array before you can think of recovering the lost data. Just one extra drive and RAID5 and you (hopefully) never have to worry about all the time you'd lose from doing the rebuild etc. It's the cost of the single extra drive vs the cost of the amount of time it takes you to rebuild and restore.
You're then talking about getting another SAN to mirror followed by talking about not having enough money to spend on the system? One drive = failover if an array dies, a FAR more likely occourance than the entire SAN dieing. (not that I think it's a bad idea to have two SANs mirroring each other, just questioning your logic over lack of funds and failover) Also may be worth pointing out that RAID5 with 3 disks vs RAID0 with 2 disks give you identical speed benefits for reads, you just lose the write benefit of RAID0.
I completely agree with mrbios, no offense intended here but the main negative with Raid 0 is it has absolutely no redundancy, so when a disk fails your webserver/media streaming etc will be down while you replace the disk, rebuild the array and restore a backup.
For the cost of one extra disk you can negate that downtime with a 3 disk raid 5, and since you mentioned that you use raid 0 for applications where there is no write, and high performance read is required, raid 5 is essentially raid 0 with parity so although write will be slower, with a highly read-biased environment, a 3 disk raid 5 on a decent controller will outperform a 2 disk raid 0 anyway (not equal it as mrbios stated, because the parity is distributed across all disks and so is the data, unlike raid 3/4). Adding spindles to any raid is the primary way of increasing performance.
Also, a 2 disk raid 0 has approx double the chance of failing in a given period of time than a single disk.
Basically you should use Raid 5 or 6 for predominantly read based applications such as web servers and streaming servers or file severs in general, and raid 1 or 10 for applications where write performance is important, such as databases. Obviously you can use raid 50/60 etc for added redundancy/performance but as a basic guideline I cant see any justification for ever using Raid 0 in a production environment.
Even if you can't afford a 3rd disk, on a decent controller, a Raid 1 array would perform better read performance than a single disk, since both disks can perform seperate reads simultaneously.
I think the concern is that RAID0 isn't technically RAID as it fails the 'redundant' part. Most people who want speed use RAID10 (or 0+1) as that gives you the speed of RAID0 with the redundancy of RAID1 (give or take, I know it's not quite that simple as there are overheads). If you've got failover then fair enough, but doesn't having to failover your whole disk pool due to a single disk failure seem a bit excessive?
Originally Posted by bossman
To each their own though, and if it's working for you then I'm certainly not criticising. :)