Take the 2-minute tour ×
Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

With RAID 5 capability is lost if there are disks of different sizes? Same as RAID 0 and RAID 1. If something happens the speeds are different? 10K and 15K.

share|improve this question

5 Answers 5

The Dell Perc controllers are the only ones I have much experience with, and with these controllers you can use any combination of disks that you want. The smallest disk will constrain the capacity and the slowest disk will constrain the speed.

But this is an odd question. If you're paying all that money for a shiny new RAID controller why would you handicap it by using old disks? I have occasionally had to use an old disk when a disk failed and I didn't have a new replacement to hand. I wouldn't recommend it though.

JR

share|improve this answer

Yes, in a RAID5 array the smallest physical volume (disk or partition) will define the size of the array, so any extra space on larger volumes in the array is not used.

You should not see any issues with drives of different speeds other than the fact the the slower drive(s) will reduce average performance.

If one of the drives is much much faster than the others then RAID4 is better then RAID5 as it keeps the parity on one one disk. If you select the faster drive for the parity area you essentially have RAID0 on the other two+ drives for read access, and parity on the faster drive (for every write, the parity drive much be written to too, as with the distributed parity blocks of RAID5). But I doubt that the difference between 15K and 10K drives will make much difference to overall performance either way here. RAID4 offers the same protection as RAID5 (it can survive one drive failing), though is not always offered as it is very rarely desirable in practise.

share|improve this answer

The problem is that it has a damaged disc on RAID 5 and the solution that I seek is to build another SCSI disk to avoid having to invest in a new complet SAS system (which I recommend) is much more reliable. Thank you very much for your answers, try to get the system change. Sorry for my english :).

share|improve this answer
    
Then you'd better choose a disk as close as possible as the original. Usually RAID vendors can sell you (very expensive) obsolete drives : for instance I have 73GB 10K RPM drives in my client spare stock :) –  wazoox Jul 1 '09 at 14:30

At best any striped RAID configuration will support reading one stripe per physical disk per revolution. There is no inherent reason why it won't work but anything reading in a round robin will be held back by the slowest disk.

Asynchronous or random I/O will be mostly a function of the total IOPS available from the physical array but any applications with a dependency on previous reads (i.e. non-async blocking I/O) may experience bottlenecks from the slower disks. Parity writes waiting on the slower disks will be held up until all I/O finishes.

Disk capacity will limit the volume size to the maximum that can be supported by the smallest disk. Depending on the capabilities of the controller you may be able to use the remaining space on the other disks for a different array. Some controllers will support a M:M relationship between disks and arrays and others won't.

At a guess I would expect it to work with some but not all I/O being held up by the 10k disks. Some RAID configurations require synchronised disk spindles but RAID-5 does not.

share|improve this answer

It's generally strongly advised to use disks of the same model and the same firmware revision in a RAID array, to minimize risk. It's technically possible to mix about any disks together, however your array will be slowed down by the slowest disk; the slowing down can be even worse than expected, because data striping is done the same way on all disks; if one is slower, the others may miss access and make an additional rotation, adding even more latency, giving you abysmal performance. This will add stress the the disks too, greatly adding to the risk of failure.

So to be short : Don't do it. Unless it's to test, or learn about RAID, but never ever put such a setup in production.

share|improve this answer
3  
If using the same drive model, I generally recommend getting the drives from a mix of suppliers. You don't want all the drives in an array having come from the same batch and that batch to be one that had QA problems! I do not have a lot of experience with hardware RAID, but with software RAID under Linux I've had no problems with mixing drive models and even using drives from different manufacturers (though performance has never particularly been my reason for using RAID). –  David Spillett Jul 1 '09 at 16:35
    
I also mixed SCSI and P-ATA harddisks, when a SCSI disk in a Raid1 died. No problems occured for a long time. I love softwareraid 0 –  ThorstenS Jul 7 '09 at 3:03

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.