zfs discuss - Aug 2008 - ZFS configuration and performance questions.

Hello! I''m new to ZFS and have some configuration questions.

What''s the difference, performance wise, in below configurations?
* In the first configuration, can I loose 1 disk? And, are the disks striped to
gain performance, as they act as one vdev?

* In the second configuration, can I loose 2 disks (one per raidz group)? Since
dynamic striping is done across vdevs, will this configuration have better
performance then configuration 1?

[b]Configuration 1[/b]
 raidz
  - dev1
  - dev2
  - dev3
  - dev4
  - dev5
  - dev6
  - dev7
  - dev8
  - dev9
  - dev10

[b]Configuration 2[/b]
 raidz
  - dev1
  - dev2
  - dev3
  - dev4
  - dev5
 raidz
  - dev6
  - dev7
  - dev8
  - dev9
  - dev10
 
 
This message posted from opensolaris.org

On 10 August, 2008 - Martin Svensson sent me these 0,9K bytes:
> Hello! I''m new to ZFS and have some configuration questions.
> 
> What''s the difference, performance wise, in below configurations?
> * In the first configuration, can I loose 1 disk?
Yes.
> And, are the disks striped to gain performance, as they act as one
> vdev?
No.
> * In the second configuration, can I loose 2 disks (one per raidz
> group)?
Yes.
> Since dynamic striping is done across vdevs, will this
> configuration have better performance then configuration 1?
Yes.

You most often trade disk space for performance.. Best performance is
gained with striped mirrors, but then you "lose" half of your disks in
space..

/Tomas
-- 
Tomas ?gren, stric at acc.umu.se, http://www.acc.umu.se/~stric/
|- Student at Computing Science, University of Ume?
`- Sysadmin at {cs,acc}.umu.se

Hi,

I''d like to share our positive experience on a POC.

We created a few iSCSI shares. Mounted on a Windows box. Then took snap shot
of one of them. On the next step we converted the snap shot in to a clone
and then tried to mount to the same Windows server. 

We all thought it will not work as drive ID''s are same and we had never
tried to take a snap shot of iSCSI volumes. 

Surprise it worked without a problem. 

We were able to test the volume a couple of minutes. 

My question is , is what we have done ok in the long run ? Can we use it for
production ?

Best regards



Mertol Ozyoney 
Storage Practice - Sales Manager

Sun Microsystems, TR
Istanbul TR
Phone +902123352200
Mobile +905339310752
Fax +902123352222
Email mertol.ozyoney at sun.com

I read this (http://blogs.sun.com/roch/entry/when_to_and_not_to) blog regarding
when and when not to use raidz. There is an example of a plain striped
configuration and a mirror configuration. (See below)

M refers to a 2-way mirror and S to a simple dynamic stripe.

Config	    Blocks Available	Random FS Blocks /sec
------------     ----------------	---------
M  2 x (50)    5000 GB			20000	  
S  1 x (100)   10000 GB		20000

Granted, the simple striped configuration is fast, and of course with no
redundancy. But I don''t understand how a mirrored configuration can
perform as good when you sacrifice half of your disks for redundancy.
Doesn''t a mirror perform as one device? Can someone please clarify the
example from the above, I think I am missing something?
 
 
This message posted from opensolaris.org

Diskspace may be lost on redundacy, but there''s still two or more  
devices in the mirror. Read requests can be spread across these.

--
Via iPhone 3G

On 11-ao?t-08, at 11:07, Martin Svensson <martin.svensson at hagemeyerau.co 
m> wrote:
> I read this (http://blogs.sun.com/roch/entry/when_to_and_not_to)  
> blog regarding when and when not to use raidz. There is an example  
> of a plain striped configuration and a mirror configuration. (See  
> below)
>
> M refers to a 2-way mirror and S to a simple dynamic stripe.
>
> Config        Blocks Available    Random FS Blocks /sec
> ------------     ----------------    ---------
> M  2 x (50)    5000 GB            20000
> S  1 x (100)   10000 GB        20000
>
> Granted, the simple striped configuration is fast, and of course  
> with no redundancy. But I don''t understand how a mirrored  
> configuration can perform as good when you sacrifice half of your  
> disks for redundancy. Doesn''t a mirror perform as one device? Can
> someone please clarify the example from the above, I think I am  
> missing something?
>
>
> This message posted from opensolaris.org
> _______________________________________________
> zfs-discuss mailing list
> zfs-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/zfs-discuss

On 11 August, 2008 - Martin Svensson sent me these 0,9K bytes:
> I read this (http://blogs.sun.com/roch/entry/when_to_and_not_to) blog
regarding when and when not to use raidz. There is an example of a plain striped
configuration and a mirror configuration. (See below)
> 
> M refers to a 2-way mirror and S to a simple dynamic stripe.
> 
> Config	    Blocks Available	Random FS Blocks /sec
> ------------     ----------------	---------
> M  2 x (50)    5000 GB			20000	  
> S  1 x (100)   10000 GB		20000
> 
> Granted, the simple striped configuration is fast, and of course with
> no redundancy. But I don''t understand how a mirrored configuration
can
> perform as good when you sacrifice half of your disks for redundancy.
> Doesn''t a mirror perform as one device? Can someone please clarify
the
> example from the above, I think I am missing something?
This is when reading.. and since both disks contain the same data, you
can pick either of them.. For reading block a and b, you can read a from
disk 1 and b from disk 2 at the same time..

/Tomas
-- 
Tomas ?gren, stric at acc.umu.se, http://www.acc.umu.se/~stric/
|- Student at Computing Science, University of Ume?
`- Sysadmin at {cs,acc}.umu.se

On Mon, 11 Aug 2008, Martin Svensson wrote:
>
> Granted, the simple striped configuration is fast, and of course 
> with no redundancy. But I don''t understand how a mirrored 
> configuration can perform as good when you sacrifice half of your 
> disks for redundancy. Doesn''t a mirror perform as one device? Can 
> someone please clarify the example from the above, I think I am 
> missing something?
There are a few wonderful things about the simple zfs mirror vdev. 
One of those really wonderful things is that a read can be satisfied 
by either device in the mirror and zfs will nicely schedule sequential 
reads so that both devices are used.  This almost doubles sequential 
read performance.  Similarly, for random read I/O, the mirrors provide 
the same number of spindles as a non-redundant stripe would.  So you 
get more IOPS when reading.

Zfs is also quite effective when writing data to a mirror pair.  I 
don''t know what zfs''s internal rules are, but there does not
appear to
be much actual overhead from writing to a mirror.  Zfs can decouple 
writes to devices in a mirror pair so that one device is allowed to 
get ahead of the other since it knows how to reconcile the differences 
in case the system crashes or the device stops responding.  This is 
virtually impossible to do with hardware RAID-1 (other than some 
possible NVRAM caching) since the two disks are truely mirrors of each 
other.

The main cost of zfs mirrors is the raw disk space consumed.  If raw 
disk space is more important to you than performance, then consider 
using raidz2 since it makes maximal use of disk as well as offering 
more data security.  Raidz and raidz2 waste precious IOPS whereas 
mirrors waste less-precious raw disk space.

Bob
=====================================Bob Friesenhahn
bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/