zfs discuss - May 2006 - Sequentiality & direct access to a file

How does (or does) ZFS maintain sequentiality of the blocks of a file.  
If I mkfile on a clean UFS, I likely will get contiguous blocks for my 
file, right?  A customer I talked to recently has a desire to access 
large volumes of sequential data.  They were concerned about maintaining 
this file as a string of sequential blocks on disk (for performance 
reasons) as they update it.

This brought up the second part of the question - how does ZFS deal with 
things that look and feel like directio?  When processes update blocks 
"in place", like a database might do, are new blocks allocated, or the
existing blocks left alone?  If they are left in place, doesn''t that 
screw around with the transactional nature of ZFS?

Is this question at all clear?

--SCott

Hello Scott,

Wednesday, May 24, 2006, 9:42:06 PM, you wrote:

SD> How does (or does) ZFS maintain sequentiality of the blocks of a file.
SD> If I mkfile on a clean UFS, I likely will get contiguous blocks for my
SD> file, right?  A customer I talked to recently has a desire to access 
SD> large volumes of sequential data.  They were concerned about maintaining
SD> this file as a string of sequential blocks on disk (for performance 
SD> reasons) as they update it.

SD> This brought up the second part of the question - how does ZFS deal with
SD> things that look and feel like directio?  When processes update blocks
SD> "in place", like a database might do, are new blocks allocated,
or the
SD> existing blocks left alone?  If they are left in place, doesn''t
that
SD> screw around with the transactional nature of ZFS?

SD> Is this question at all clear?

"updated" blocks will be actually written to new place - so if you
update random blocks in a file then reading this file sequentially
will not be that sequential.


-- 
Best regards,
 Robert                            mailto:rmilkowski at task.gda.pl
                                       http://milek.blogspot.com

On Thu, May 25, 2006 at 12:01:18AM +0200, Robert Milkowski
wrote:
> Hello Scott,
> 
> Wednesday, May 24, 2006, 9:42:06 PM, you wrote:
> 
> SD> How does (or does) ZFS maintain sequentiality of the blocks of a
file.
> SD> If I mkfile on a clean UFS, I likely will get contiguous blocks for
my
> SD> file, right?  A customer I talked to recently has a desire to access
> SD> large volumes of sequential data.  They were concerned about
maintaining
> SD> this file as a string of sequential blocks on disk (for performance 
> SD> reasons) as they update it.
> 
> SD> This brought up the second part of the question - how does ZFS deal
with
> SD> things that look and feel like directio?  When processes update
blocks
> SD> "in place", like a database might do, are new blocks
allocated, or the
> SD> existing blocks left alone?  If they are left in place,
doesn''t that
> SD> screw around with the transactional nature of ZFS?
> 
> SD> Is this question at all clear?
> 
> "updated" blocks will be actually written to new place - so if
you
> update random blocks in a file then reading this file sequentially
> will not be that sequential.
On the other hand, if you are reading the file sequentially, ZFS has
very good read-ahead algorithms.

Cheers,
- jonathan

-- 
Jonathan Adams, Solaris Kernel Development

Scott Dickson writes:
 > How does (or does) ZFS maintain sequentiality of the blocks of a file.  
 > If I mkfile on a clean UFS, I likely will get contiguous blocks for my 
 > file, right?  A customer I talked to recently has a desire to access 

you would get up to maxcontig worth of sequential block.
But only inasmuch as there is a single thread in the system
doing the allocating writes.  When more threads compete for
blocks we end up with less contiguity.

With ZFS, i believe that as part of a transaction group (a big sync every 5 
second) dirtied blocks of a file are reordered by
offsets then sent to disk in that order.
It would be interesting to see if it truly works that way
but basically ZFS may reduce the randomness as part of the sync.


-r
 > 
 > --SCott
 > _______________________________________________
 > zfs-discuss mailing list
 > zfs-discuss at opensolaris.org
 > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss