[Lustre-discuss] Small scale OST configuration

Chris Jewell chris.jewell at warwick.ac.uk
Mon Sep 5 05:33:13 PDT 2011


Hi all,

I'd like to first introduce myself as a Lustre newbie who is interested in implementing Lustre on a small cluster to improve on NFSv3 limitations.  Let me describe my setup:

I have 11 execution nodes (104 cores total) which are each connected via 1GE to a fileserver (using 4x1GE aggregated link).  The fileserver currenty has a 4-disk 1.5TB RAID10 array as its main storage, plus a 2-disk 2TB RAID1 array, and a 4-disk 4TB RAID6 array.  Whilst the RAID10 array must stay intact (due to system OS, shared application partitions etc etc), I'm looking to possibly reconfigure the 6 (2TB) disks into arrays suitable for serving 3 Lustre filesystems (/home, /storage, /scratch).  This would be with a view to having the MGS and 3 MDT partitions on the main storage, with OST partitions spread across the 6 disks.  

The question is, what is most efficient?  Given that my RAID controller is hardware-based (Dell PERC6), would it be better to split my 6 disks into multiple OSTs, or keep them as one big lump?  I've not seen a discussion of this in the Lustre doc, so I'd like to gather a few pointers/opinions, if I may?

Thanks,

Chris


--
Dr Chris Jewell
Department of Statistics
University of Warwick
Coventry
CV4 7AL
UK
Tel: +44 (0)24 7615 0778








More information about the lustre-discuss mailing list