There’s also the ‘schedule-able’
command ‘wafl reallocate’ that does the same thing.
It’s very useful, though there is
some debate on exactly what happens when snapshots are taken into
consideration. Generally speaking, if there is adequate free space, and the
data in question is deemed ‘fragmented’ by the scanner, then it
will move it.
Not that it much matters, but adding disks
to an existing RG vs adding a new RG is about the same: when data is written
within an RG, it is striped across the disks in such a fashion that each disk
gets x number of contiguous blocks, then it writes to the next disk (64 or 128
depending on the version of ONTAP and disk type). Parity is calculated per
stripe. When writing across multiple RGs, WAFL decides how much data per RG to
write (tetris, but not the game) and breaks up the writes per RG in this
manner. I believe this is in 64MB chunks.
Either way, the data is written to the
location where the best example of contiguous free space exists.
Glenn
From:
owner-toasters@mathworks.com [mailto:owner-toasters@mathworks.com] On Behalf Of Glenn Dekhayser
Sent: Monday, January 22, 2007
3:22 PM
To: toasters@mathworks.com
Subject: RE: Aggregate expansion
All;
It was my understanding that you can
alleviate the pain caused by an uneven expansion of disks as mentioned below by
performing a wafl scan reallocate, anyone have info to contrary?
Glenn
From:
owner-toasters@mathworks.com [mailto:owner-toasters@mathworks.com] On Behalf Of Fox, Adam
Sent: Monday, January 22, 2007
2:30 PM
To: Suresh Rajagopalan;
toasters@mathworks.com
Subject: RE: Aggregate expansion
That depends greatly on the values of N
and n, as well as the RAID group size of the aggregate.
Done according to best practices, there
will be almost no difference. Done poorly, it can make all the
difference in the world.
My personal view (I'm in no position to
speak officially for NetApp) on this to add disks to an aggregate in one of 2
multiples:
1) a whole RG at a time
2) half a RG at a time.
This typically allows for a sufficient
number of free disks such that you should not expect any noticable
performance difference. I realize
that not all sites can implement this, but let's look at a worst practice:
N = n+1
Fill up the aggregate, then add 1
disk. Ouch! This hurts! So you've seen what I consider to be
the best
case for adding space, and you've seen the
worst case. So how close you are to these extremes should
give you can idea of what to expect.
I know this isn't a simple answer, but I
believe it to be accruate.
--
Adam Fox
adamfox@netapp.com
From: Suresh
Rajagopalan [mailto:SRajagopalan@williamoneil.com]
Sent: Monday, January 22, 2007
12:46 PM
To: toasters@mathworks.com
Subject: Aggregate expansion
Is there any difference between creating an aggregate on a
certain number of disks (say n) , and then later expanding the aggregate
to N disks, as opposed to creating the initial aggregate on N disks?
Suresh