slowdown of zfs (tx->tx)

[Nicolas Rachinsky]

* Steven Hartland <killing at multiplay.co.uk> [2013-01-14 00:05 -0000]:
> 
> ----- Original Message ----- From: "Nicolas Rachinsky"
> <fbsd-mas-0 at ml.turing-complete.org>
> To: "Steven Hartland" <killing at multiplay.co.uk>
> Cc: "freebsd-fs" <freebsd-fs at freebsd.org>
> Sent: Sunday, January 13, 2013 11:52 PM
> Subject: Re: slowdown of zfs (tx->tx)
> 
> 
> >* Steven Hartland <killing at multiplay.co.uk> [2013-01-11 13:58 -0000]:
> >>TBH looks like your just saturating your disks with the number of IOP's
> >>your doing.
> >
> >But now a backup takes forever (16hours and more) that took less than
> >30 minutes two weeks ago.
> >
> >I duplicated the complete setup to another (slower) server. There the
> >backups are slower than they were on this machine, but they are much
> >faster than on this machine.
              ^they are       ^now

> 
> Its not something silly like you have 4k disks which aren't 4k aligned
> is it?

No, these are 512 bytes/sector disks. And it started to become so slow
without any change (hardware or software). 

> IIRC you where using rsync, is there a reason why you aren't using
> zfs send - recv?

Most of the machines whose backups are put to this machine run with
linux and don't have zfs.

Nicolas
-- 
http://www.rachinsky.de/nicolas