Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 6 Dec 2016 22:26:21 +0200
From:      Alexander Motin <mav@FreeBSD.org>
To:        freebsd-fs@FreeBSD.org
Cc:        Alex Tutubalin <lexa@lexa.ru>, Andriy Gapon <avg@FreeBSD.org>
Subject:   Re: 11-STABLE vs 11.0-RELENG test
Message-ID:  <8b4ba98d-03d3-f671-33b2-ed12d3b4fb7c@FreeBSD.org>

next in thread | raw e-mail | index | archive | help
On 06/12/2016 18:17, Alex Tutubalin wrote:
> Followup, same box as in 1st message, but with different HBA and 5 HDD:
>
> Summary:
>  1) read speed depends on 'on-disk' data, not by code used while read.
>  2) FreeBSD 11.0-releng (svn up today) creates 'fast' big files (420+
Mb/s read
> speed)
>  3)  FreeBSD 11-STABLE creates 'slow' big files. (~200  Mb/s)
>  4) zfs send slow-dataset | zfs recv   (under 11.0-releng) creates
> 'intermediate' files (320Mb/s)
>  5) file copy slow-file ... (under 11.0-releng) creates fast copy
(400+Mb/s)
>
> So, ZFS write code in 11-STABLE looks broken.

I've reproduced this issue with quick test on my lab system configured
with 12-disk RAIDZ2 pool.  I've measured write and read back (with and
without prefetch) speeds for pool recreated on different FreeBSD head
revisions:
		r309625	r305456	r305330	r305322
write		702	701	1115	1120
read w/ pref	232	228	518	512
read w/o pref	128	126	242	240

I suspect we could obtain the problem here:

r305331 | mav | 2016-09-03 13:04:37 +0300 (сб, 03 сент. 2016) | 45 lines

MFV r304155: 7090 zfs should improve allocation order and throttle
allocations

I'll take closer look on that change tomorrow.

-- 
Alexander Motin



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?8b4ba98d-03d3-f671-33b2-ed12d3b4fb7c>