Lustre discuss - Jun 2006 - Two Driver on On e OST?

On Jun 02, 2006  10:31 +0800, pandy.song wrote:
> I can add one device to "ost1-test" via following command:
> lmc -m config.xml --add ost --node node_1 --lov lov-test --ost ost1-test
--fstype ext3 --dev /dev/hda
> 
> but how can i add another "/dev/hdb" to "ost1-test" ?
You can only have a single block device per OST, though this block device
may be a virtual device via DM, LVM/MD, hardware RAID, etc.  An OST
corresponds to exactly one ext3 filesystem.

It is possible to have multiple OSTs per OSS (node), so you can do:

lmc -m config.xml --add ost --node node_1 --lov lov-test --ost ost1-test
--fstype ext3 --dev /dev/hda
lmc -m config.xml --add ost --node node_1 --lov lov-test --ost ost2-test
--fstype ext3 --dev /dev/hdb

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

Thank Andreas Dilger for your detailed reply.

So I still try to attach two or more harddisks  in one 
node to one ost. How can I get correct config.xml via LMC tool?

I can add one device to "ost1-test" via following command:
lmc -m config.xml --add ost --node node_1 --lov lov-test --ost ost1-test
--fstype ext3 --dev /dev/hda

but how can i add another "/dev/hdb" to "ost1-test" ?


-----????-----
???:"Andreas Dilger" <adilger@clusterfs.com>
????:2006-06-02 01:12:32
???:"David Vasil" <dmvasil@ornl.gov>
??:lustre-discuss@clusterfs.com
??:Re: [Lustre-discuss] Two Driver on One OST?



On Jun 01, 2006  08:25 -0400, David Vasil wrote:
&gt; pandy.song wrote:
&gt; &gt;     One of my storage node has two disk driver, Must I
&gt; &gt; create two ost for two driver using LMC tool ? OR how
&gt; &gt; can I attach two harddisks fro one ost?
&gt; 
&gt; You could use LVM to create a single OST out of the two disks.  Is
there
&gt; a particular reason you dont want to create two separate OSTs?  While
&gt; LVM does have the ability to stripe across multiple disks using the -i
&gt; and -I flag to lvcreate, one would expect you to get better performance
&gt; by letting Lustre do the striping.

This is probably true regarding the performance, though we haven''t
tested
in this area specifically.  As for the underlying storage, Lustre can
use any block device that is available, and LVM is a good choice.  It is
also desirable to have some form of redundancy at the storage layer like
RAID1 or RAID5 unless the filesystem is only used for temporary files and
the loss of files is not serious.  There is currently a noticable performance
impact of using the kernel software RAID5 currently that we are working to fix.

One drawback of having many small OSTs vs. fewer larger ones is that the
free space is more fragmented.  If you run mostly with smaller files this
is fine, they will be distributed around the OSTs, but if you have very
large files then you will prematurely run out of space.

For all but the smallest systems we recommend OSTs of at least 500GB,
but preferrably close to 2TB in size.  With modern disks this may end
up being a single 500GB or 750GB disk, which is also OK, assuming they
have RAID of some form.

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

_______________________________________________
Lustre-discuss mailing list
Lustre-discuss@clusterfs.com
https://mail.clusterfs.com/mailman/listinfo/lustre-discuss

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20060601/f1c751ba/attachment-0001.html