Re: ZFS: alignment/boundary for partition type freebsd-zfs

From: O. Hartmann <ohartmann_at_walstatt.org>
Date: Tue, 26 Dec 2017 21:24:12 +0100
Am Tue, 26 Dec 2017 09:31:53 -0800 (PST)
"Rodney W. Grimes" <freebsd-rwg_at_pdx.rh.CN85.dnsmgr.net> schrieb:

> > On Tue, Dec 26, 2017 at 10:04 AM, O. Hartmann <ohartmann_at_walstatt.org>
> > wrote:
> >   
> > > Am Tue, 26 Dec 2017 11:44:29 -0500
> > > Allan Jude <allanjude_at_freebsd.org> schrieb:
> > >  
> > > > On 2017-12-26 11:24, O. Hartmann wrote:  
> > > > > Running recent CURRENT on most of our lab's boxes, I was in need to  
> > > replace and  
> > > > > restore a ZFS RAIDZ pool. Doing so, I was in need to partition the  
> > > disks I was about  
> > > > > to replace. Well, the drives in question are 4k block size drives with  
> > > 512b emulation  
> > > > > - as most of them today. I've created the only and sole partiton on  
> > > each 4 TB drive  
> > > > > via the command sequence
> > > > >
> > > > > gpart create -s GPT adaX
> > > > > gpart add -t freebsd-zfs -a 4k -l nameXX adaX
> > > > >
> > > > > After doing this on all drives I was about to replace, something drove  
> > > me to check on  
> > > > > the net and I found a lot of websites giving "advices", how to prepare  
> > > large, modern  
> > > > > drives for ZFS. I think the GNOP trick is not necessary any more, but  
> > > many blogs  
> > > > > recommend to perform
> > > > >
> > > > > gpart add -t freebsd-zfs -b 1m -a 4k -l nameXX adaX
> > > > >
> > > > > to put the partition boundary at the 1 Megabytes boundary. I didn't do  
> > > that. My  
> > > > > partitions all start now at block 40.
> > > > >
> > > > > My question is: will this have severe performance consequences or is  
> > > that negligible?  
> > > > >
> > > > > Since most of those websites I found via "zfs freebsd alignement" are  
> > > from years ago,  
> > > > > I'm a bit confused now an consideration performing all this  
> > > days-taking resilvering  
> > > > > process let me loose some more hair as the usual "fallout" ...
> > > > >
> > > > > Thanks in advance,
> > > > >
> > > > > Oliver
> > > > >  
> > > >
> > > > The 1mb alignment is not required. It is just what I do to leave room
> > > > for the other partition types before the ZFS partition.
> > > >
> > > > However, the replacement for the GNOP hack, is separate. In addition to
> > > > aligning the partitions to 4k, you have to tell ZFS that the drive is 4k:
> > > >
> > > > sysctl vfs.zfs.min_auto_ashift=12
> > > >
> > > > (2^12 = 4096)
> > > >
> > > > Before you create the pool, or add additional vdevs.
> > > >  
> > >
> > > I didn't do the sysctl vfs.zfs.min_auto_ashift=12 :-(( when I created the
> > > vdev. What is
> > > the consequence for that for the pool? I lived under the impression that
> > > this is necessary
> > > for "native 4k" drives.
> > >
> > > How can I check what ashift is in effect for a specific vdev?
> > >  
> > 
> > It's only necessary if your drive stupidly fails to report its physical
> > sector size correctly, and no other FreeBSD developer has already written a
> > quirk for that drive.  Do "zdb -l /dev/adaXXXpY" for any one of the
> > partitions in the ZFS raid group in question.  It should print either
> > "ashift: 12" or "ashift: 9".  
> 
> And more than likely if you used the bsdinstall from one of
> the distributions to setup the system you created the ZFS
> pool from it has the sysctl in /boot/loader.conf as the
> default for all? recent?  bsdinstall's is that the 4k default
> is used and the sysctl gets written to /boot/loader.conf
> at install time so from then on all pools you create shall
> also be 4k.   You have to change a default during the
> system install to change this to 512.


I never used any installation scripts so far.

Before I replaced the pool's drives, I tried to search for informations on how-to. This
important tiny fact must have slipped through - or it is very bad documented. I didn't
find a hint in tuning(7), which is the man page I consulted first.

Luckily, as Allan Jude stated, the disk recognition was correct (I guess stripesize
instead of blocksize is taken?). 

>  
> > -aLAn  
> 
> > _______________________________________________
> > freebsd-current_at_freebsd.org mailing list
> > https://lists.freebsd.org/mailman/listinfo/freebsd-current
> > To unsubscribe, send any mail to "freebsd-current-unsubscribe_at_freebsd.org"
> >   
> 



-- 
O. Hartmann

Ich widerspreche der Nutzung oder Übermittlung meiner Daten für
Werbezwecke oder für die Markt- oder Meinungsforschung (§ 28 Abs. 4 BDSG).

Received on Tue Dec 26 2017 - 19:25:01 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:41:14 UTC