[Lustre-discuss] OSSs, OSTs and RAID (once again)

Nikolay Kvetsinski nkvecinski at gmail.com
Mon Jun 17 06:57:13 PDT 2013

Hello guys, first of all sorry for asking about OSS and OST RAID
configurations  I say few posts on the subject , but they don`t quite
answer my questions.  So my pickle is : I have a couple of storage
enclosures (16 bay storage controller + 16 bay JBOD attached.). Each 16 bay
storage controller has dual raid controller, so in order to use the maximum
hardware performance and maximum storage capacity,  I think to create 4
Luns (8 disks each in RAID 5). The first 2 Luns will be assigned to
controller "A" and the second 2 Luns to controller "B", total 4 OSS per
OST. This way I will break the "Golden rule" about Lusre 1MB RPC. I`ve
created the RAID with 128 KB chunk size(stripe size in my controllers
"language"). So my mkfs.lustre options are :

-E stride=128, stripe_width=(128 * 7 = 896)

In the manual it says, that if one cannot fit evenly into 1 MB, it should
choose such stripe_width that is close, but not larger that 1 MB, as I am.
If I choose bigger stripe size for the RAID controller (256 KB) I`ll go way
beyond 1MB. So in your expert opinions will my configuration suffer a lot
from this, and are my calculations correct. I can use different scheme for
the luns for example RAID5(4+1), RAID6(4+2) but in both cases I loose disks
and storage capacity and number of OSS per OST goes from 4 to 6.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20130617/3f6a720d/attachment.htm>

More information about the lustre-discuss mailing list