[Lustre-discuss] adding ost(s)

[Andreas Dilger]

On Jan 17, 2008  11:22 -0500, Aaron S. Knister wrote:
> I have a 33TB lustre filesystem that's 84% used. Today I plan to add an
> additional 16TB. When the 16TB is brought  online how will lustre handle
> it in terms of space allocation? Will it temporarily stop adding new files
> to the existing osts until the newer ones are relatively the same % full?

It won't stop using the new OSTs entirely, but they will be used only
in proportion to how much space is available on those OSTs compared to the
new OSTs.  On my home system, for example, there are different OST sizes:

	$ lfs df
	UUID                 1K-blocks      Used Available  Use% Mounted on
	mds-myth-0_UUID        9174328    675720   8498608    7% /myth[MDT:0]
	ost-myth-0_UUID      292223856 249675772  42548084   85% /myth[OST:0]
	ost-myth-1_UUID       94442984  92942296   1500688   98% /myth[OST:1]
	ost-myth-2_UUID      487388376 273782068 213606308   56% /myth[OST:2]
	ost-myth-3_UUID      487865304 273839020 214026284   56% /myth[OST:3]

	filesystem summary:  1361920520 890239156 471681364   65% /myth

Note ost-myth-1 is much smaller (it shares the same physical disk as
mds-myth-0, as did ost-myth-0 before it was migrated to a larger separate
disk).  The ost-myth-[23] were added more recently.

The available space on ost-myth-1 hovers around the same amount of free
space as new files are added and deleted, but the majority of new files
are created on ost-myth-2 and ost-myth-3.  Of the 230 files created in
the last week (2 objects each) the object distribution is:

ost-myth-0:  41
ost-myth-1:   2
ost-myth-2: 206
ost-myth-3: 211

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.