Lustre discuss - Jul 2006 - Understanding Lustre Terminology

Hi,
I need some explanation of the Lustre terminology and its capacity in
terms of storage.
 
1) In a Lustre File System can there be only one OSS which has Multiple
OSTs?
 
2) If I configure one OST on one physical machine with an underlying
storage would that be considered as one OSS? or If there are 3 OSTs on 3
different Physical systems would each of the system be considered as an
OSS or would I just consider there to be one OSS with 3 OSTs?
 
3) Again I might have an underlying storage with 2TB of data split
accross multiple disks in a raid pattern. Now is an OST for this machine
aware of this RAID pattern and during the Lustre file system
configuration can I specify any OSD management. Does an OST see the
entire underlying storage as one aggregate storage? As in does an OST
manage just one OSD or multiple OSTs?
 
Thanks
Priya
 
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20060717/3f3b029c/attachment.html

On Jul 17, 2006  14:57 -0700, Priya Muralidharan wrote:
> 1) In a Lustre File System can there be only one OSS which has Multiple
> OSTs?
OSS = node which is exporting Object Storage Target(s) to network
OST = Object Storage Target, equivalent to a single ldiskfs (ext3)
      filesystem on a single Linux block device
> 2) If I configure one OST on one physical machine with an underlying
> storage would that be considered as one OSS?
Yes, this would be 1 OST on 1 OSS.
> If there are 3 OSTs on 3 different Physical systems would each of the
> system be considered as an OSS
Yes, this would 3 OSTs on 3 OSSes.
> 3) Again I might have an underlying storage with 2TB of data split
> accross multiple disks in a raid pattern. Now is an OST for this machine
> aware of this RAID pattern and during the Lustre file system
> configuration can I specify any OSD management.
If there is a single (possibly virtual) block device with a single
ldiskfs (ext3) filesystem this is a single OST.  The block device can
be any kind of Linux block device, e.g. device mapper logical volume,
MD RAID0, 1, 5 device (though I would STRONGLY recommend AGAINST a
RAID0 device), a single SCSI/SATA/FC disk, external hardware RAID
device, etc.
> Does an OST see the entire underlying storage as one aggregate storage?
> As in does an OST manage just one OSD or multiple OSTs?
One OST is mapped to a single OSD == a single block device.


On a related note - it is possible that the ldiskfs filesystem can
be made a little bit aware of the underlying RAID configuration.
The mke2fs command can use the "-R stride=NNNN" parameter to better
align the filesystem metadata with the RAID chunk size so that the
metadata is spread evenly over the disks in the RAID set.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.