Storage Configuration using R710 and MD1200

Ken Nishimura nishimur at opus.scs.agilent.com
Mon May 3 16:56:22 CDT 2010


Trying to bring up a new DAS fileserver comprised of a R710 as host
with a PERC H800 connected to a MD1200 with about 12TB (raw) of drives.

We currently have a Nexsan ATABoy connected to a HP fileserver over SCSIU320
which we are replacing.

Currently, the ATABoy whose controller is on the disk array is configured to
look like one large LUN of 4.8TB -- internally it is RAID5 over 14 spindles.

We currently use that single LUN as a single PE on lvm, and use lvm to
create a bunch of logical volumes which are then used as needed.

Question:  What is recommended?  We want to have a bunch of filesystems of
approximately 500 GB in size.  I can see several scenarios:

1) Replicate what we have.  Configure the PERC to create one huge virtual
disk which is RAIDed on the PERC controller as a single physical drive
to lvm.  Use lvm to create logical volumes as needed

2) Use the PERC controller to make several large (>>500 GB) virtual drives
and then partition each of those to create the disk lumps we want.

3) Use the PERC controller to make a large number of ~500 GB virtual drives
and use them each individually.

What works best?  Any other ideas?

Some points:

Under (1), the single drive would show as /dev/sda, and we would use
lvm to make that some /dev/vg1.  The names of each of the volumes under
/dev/vg1 is fully under our control.  We could assign a sensible name to
each of the volumes (e.g. /dev/vg1/source_tree instead of /dev/sda?)

Under (2) or (3), the device name created by the PERC controller is
somewhat hit or miss -- you know what you get, but depending of how
things are reconfigured, could you not end up with a different device name?
That would be a pain.  I suppose one could use udev, but that seems like
another set of files to maintain.

Am I missing something?

Thanks,

Ken



More information about the Linux-PowerEdge mailing list