Lustre discuss - May 2006 - mds storage capacity compared to ost storage capacity

On Mon, 2005-12-19 at 14:03 -0500, Chris Bachmann wrote:
> I''m getting ready to start building my storage cluster and I am
> beginning to assemble the hardware that I''ll need for this. The
one
> question I have is how big does the mds have to be compared to the
> osts? I don''t mind building a beast of a machine to handle it, but
I
> do need to know the minimum space per TB of ost space. Is there a
> simple formula that I can use?
The space needed on the MDS is related to how many inodes you need.  You
should plan for the number of files and directories, not the size of the
OST.  Lustre creates large inodes on the MDS filesystem, so it can
normally store all the metadata information right in the inode.  The
controlling factor will be how many inodes you have on your MDS.

I created an MDS on a 1.6TB storage device like so:

<mdsdev>
[...]
 <mkfsoptions>
  -T news
 </mkfsoptions>
[...]
</mdsdev>

... meaning the filesystem is created using mke2fs -T news.  This yields
a lot of inodes:

mds-1:~# dumpe2fs -h /dev/mds1-vg/mds1-lv | grep ^Inode
dumpe2fs 1.37 (21-Mar-2005)
Inode count:              390627328
Inodes per group:         32768
Inode blocks per group:   2048
Inode size:               256

I have almost 400 million inodes on the MDS.

I hope that helps.

-jwb

Jeffrey W. Baker wrote:
> On Mon, 2005-12-19 at 14:03 -0500, Chris Bachmann wrote:
> 
>>I''m getting ready to start building my storage cluster and I am
>>beginning to assemble the hardware that I''ll need for this. The
one
>>question I have is how big does the mds have to be compared to the
>>osts? I don''t mind building a beast of a machine to handle it,
but I
>>do need to know the minimum space per TB of ost space. Is there a
>>simple formula that I can use?
> 
> 
> The space needed on the MDS is related to how many inodes you need.  You
> should plan for the number of files and directories, not the size of the
> OST.  Lustre creates large inodes on the MDS filesystem, so it can
> normally store all the metadata information right in the inode.  The
> controlling factor will be how many inodes you have on your MDS.
For a sizing equation, we reccomend 4k on the MDS for each file on the 
filesystem. As Jeffery points out, number of inodes is also important.
cliffw

> 
> I created an MDS on a 1.6TB storage device like so:
> 
> <mdsdev>
> [...]
>  <mkfsoptions>
>   -T news
>  </mkfsoptions>
> [...]
> </mdsdev>
> 
> ... meaning the filesystem is created using mke2fs -T news.  This yields
> a lot of inodes:
> 
> mds-1:~# dumpe2fs -h /dev/mds1-vg/mds1-lv | grep ^Inode
> dumpe2fs 1.37 (21-Mar-2005)
> Inode count:              390627328
> Inodes per group:         32768
> Inode blocks per group:   2048
> Inode size:               256
> 
> I have almost 400 million inodes on the MDS.
> 
> I hope that helps.
> 
> -jwb
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss@lists.clusterfs.com
> https://lists.clusterfs.com/mailman/listinfo/lustre-discuss

I''m getting ready to start building my storage cluster and I am
beginning to
assemble the hardware that I''ll need for this. The one question I have
is
how big does the mds have to be compared to the osts? I don''t mind
building
a beast of a machine to handle it, but I do need to know the minimum space
per TB of ost space. Is there a simple formula that I can use?
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20051219/18d8a210/attachment.html