Red Hat

Hi,

I'm new here and have a quick question about Linux Raid and LVM.

I have a HP server with 2 physically separate raid arrays both contain 8 spindles which are configured in hardware as raid 5. When I installed Cent OS 5.7 they were seen, as you would expect, as two separate physical volumes by the OS.

Now CentOS I believe used LVM to pool both the volumes together into a logical volume group and within that group there are 2 logical volumes. I think LVM is not doing any kind of performance enhancement and is just joining the 2 physical volumes together, so writing data will fill the first physical volume and then once full will start using the other.

So my questions are:

Am I likely to see any speed benefits using software raid 0 so it uses 16 spindles simultaneously?

Will this configuration co-exist with LVM or will I need to destroy the LVM and switch over to software raid and redo partitions and mount points?

Cheers,
David.

using raid 0 will provide no fault tolerance !!

To change raid level you have to destroy the existing raid if it is hardware raid with hp array controller !!

If you create a raid 0 with all the disks how will OS read the raid partition a single partition or 16 different partitions ?

Sorry I should clarify, I plan to keep the hardware raid 5 but I wanted to know if it was feasible to put software raid 0 on top so both raid arrays are used simultaneously and protected by the hardware raid 5. The server in question is a file server so the CPU sits idle most of the time but just getting the best config for performance without sacrificing data integrity.

In a word, no. Software RAID doesn't combine two devices into one double-speed device. Though if you have two programs reading simultaneously, one may one device, and the other may get the other device, spreading out the load.

If you are having performance issues, perhaps RAID 5 is a bad choice. Other options might be to check the controller settings to see if you are doing write-through instead of write-back (using the controller's cache or bypassing it).

Another thing to consider is the chunk/stripe size in RAID 5. Depending on workload, it could be an issue. Did you align the file systems so that you are not on a boundary?

Also, even though you have dual RAID 5 arrays, you can and will lose data in a Software RAID 0 system during a failure if you lose 2 drives in an array or part of it becomes inaccessible, as there are parts of the same blocks of data on different parts of the software array. You will have used your disks in the least efficient manner possible.

I suggest you consider RAID 10 for your system if you plan on doing thing this way. Alternately, you may want to check/change your I/O elevator (scheduler), and determine what your most common workload for the disks will be, and then create a setup that will work best in that scenario.

Thanks for the replies guys. I was always planning to keep the hardware raid 5 as the last layer of protection so I always have that ability to loose a drive from each of the arrays and they 'should' still function until I can swap in a new drive.

I'm in the lucky position that I have a spare server that I can test all this out with so I may just bite the bullet and try the software raid 0 on top to see if I can get all those spindells flocking in the same direction.

Backup strategy at the moment is just rsync to a cifs share which seems to work quite well.

I repeat, software RAID 0 on Linux does not do this. It won't combine two devices into one double-speed device. It just gives you redundancy.