[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

[linux-lvm] starting LVs on RAID stripe boundaries



Hi Linux-LVM,

I'm searching for some definitive information about how to make sure that logical
volumes start exactly at the beginning of a RAID stripe.

My hardware is an EMC Clariion CX3-40F array with a RHEL5 host.
The default stripe element size is 64 KB (128 blocks) on the CX3.

Both EMC and Red Hat recommend using fdisk or parted  to align the first partition
to the start of a stripe boundary (128th block, 128x512=64KB) :
Red Hat : http://kbase.redhat.com/faq/docs/DOC-2893
EMC : "EMC CLARiiON Best Practices", PDF available on emc.com

After doing that. the first partition is now aligned, next I need to make sure my logical
volumes are similarly aligned. However, after reading the LVM Howto on tldp.org,
the LVM wiki at redhat.com, and Alasdair Kergon's LinuxTag presentation at :
http://people.redhat.com/agk/talks/LVM2-LinuxTag2006/
I found no information on how to do that. 

I'll now use some reasonable (?) assumptions :
* the physical extent size is a multiple of 64K (vgcreate's default is 4MB) ;
* if physical extents are aligned, logical volumes are themselves aligned
  (lvcreate -l, --extents LogicalExtentsNumber) ;
* since PV, VG and LV infos are stored in PV metadata, the only possible PE
   misalignment culprit is PV metadata ;
* PV metadata size is a multiple of 64KB 
 ( http://orezpraw.com/orezpraw.php/blogosphere/your-filesystem-starts-where )

Since 64KB is my stripe size, and metadata size is a multiple of 64KB,
pvcreate should do the RightThing(tm) with default options, i.e. start the
first PE at a multiple of the stripe size.

In other words, aligning the partition the PV uses to the stripe boundary and
using a PE size multiple of the stripe size insure that LVs get themselves created
on stripe boundaries.

Comments, please ?

Thanks,

Fran├žois Cami


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]