Portal Home > Knowledgebase > Articles Database > Different drive sizes
Different drive sizes
| Posted by bbrog, 01-11-2011, 06:23 PM |
| Hi, was just wondering how we can setup a raid5 with different disk sizes
ex. 250, 500, 1500GB disks. (this should equal 1499GB of usable space)
So how can we set this up using software raid? Can we partition the drives or something rather then the raid taking the smallest drive?
Also will this still allow fault tolerance or no?
|
| Posted by brentpresley, 01-11-2011, 07:18 PM |
| First - NEVER use software RAID for RAID 5 or 6. Let me repeat that just to be clear - NEVER.
I'm not trying to be harsh or anything, but to really sink in how bad of an idea it is. RAID 5 and 6 use parity striping to store a copy of the data across the drives so that any one drive can be recreated from the stripe in the event of a failure. Computing and writing that data stripe takes a good amount of dedicated hardware (i.e. the controller chip on the RAID card). Doing this is software slows down the entire process. How slow? When I tested this out on 6 x 1TB drives about 2 years ago the write speeds were in the neighborhood of 20-30MB/s. Yes, much slower than just a single drive alone would be. On a dedicated card, it would be 3-7 times faster than that, but still nothing like what you see in RAID 1 or RAID 10.
So, now to the second point. Trying to mix drives of varied sizes in RAID is also asking for a lot of headaches. I don't even recommend trying drives of different manufacturers if you want it to be reliable. And many RAID controllers just won't let you do anything more than RAID 0 out of mixed drives.
Hope that helps. Probably not the answer you were looking for.
|
| Posted by OLM | DavidG, 01-11-2011, 07:24 PM |
| I agree completely with what brentpresley stated about using software RAID5/6. If you want to use these RAID levels, look into the Adaptec 5405/5805 hardware RAID controllers, along with enterprise or RAID edition hard drives. Otherwise, stick with software RAID1. Mixing drives is also a bad idea. Try to always match the exact model numbers for reliability and performance.
|
| Posted by bbrog, 01-12-2011, 12:41 AM |
| Hi, thanks a lot for the answers!
One more thing how come its bad to use different size drives and models? Is it only because of performance issues or something else also?
|
| Posted by chaseideas, 01-12-2011, 12:50 AM |
| I'd have to agree with the advice that software raid 5/6 is a really bad idea, but would overall disagree with the statement about using different drive manufacturers (definitely use the same size drives though). The same speed drives, and generally well-supported models, are a boon to a reliable RAID setup in the traditional sense that you have more redundancy with different drive batches and whatnot, although most of my arrays are from the same drive type and manufacturer.. so, just saying.
|
| Posted by funkywizard, 01-12-2011, 05:59 AM |
| I wouldn't say raid 5 / 6 is a total bad idea on software. I was doing some copies from a single disk to a 3 drive software raid 5, and running iostat, it looked like the single drive was maxed out, but the raid 5 disks were not. As well, under top, the cpu use to do the parity was only about 3% of one cpu core, as you can tell since linux will tell you how much cpu is being used by mdstat-raid5. Certainly a good hardware raid controller is going to perform better, but it's not like you should *never* do software raid 5, that advice is just silly. Certainly for a write heavy environment, it's not what you want, but if you're just putting a bunch of files on there that will mostly only be read, it's not totally inappropriate.
|
| Posted by TheMagicMan, 01-12-2011, 06:36 AM |
| Here's some actual benchmarks...
http://www.linux.com/news/hardware/s...-software-raid
It looks like raid 5 is fine on software. They found only a 15% speed increase in using hardware over software. (Well depending on what benchmark you look at).
|
| Posted by brentpresley, 01-12-2011, 08:52 AM |
| Not sure where you are getting "fine" out of those numbers, but they look horrible to me.
Remember, you are seeing ONE DRIVE with higher write speeds than the RAID 5/6 array.
This guy ran across EXACTLY what you would expect:
http://www.linuxquestions.org/questi...rmance-844663/
He was getting 480MB/s read speeds, but when he had to rebuild the array from a defunct drive, he got 60MB/s write speeds. 60MB/s is about 1/2 the speed of a good SINGLE drive these days.
|
Add to Favourites 
Print this Article
Also Read
Shared Hosting
Shared Hosting
