Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 07 Dec 2010 13:31:27 +0200
From:      Alexander Motin <mav@FreeBSD.org>
To:        Pawel Jakub Dawidek <pjd@FreeBSD.org>
Cc:        svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org, Ivan Voras <ivoras@freebsd.org>, John Baldwin <jhb@freebsd.org>
Subject:   Re: svn commit: r216230 - head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs
Message-ID:  <4CFE1B0F.90908@FreeBSD.org>
In-Reply-To: <20101207110410.GE1700@garage.freebsd.pl>
References:  <201012061218.oB6CI3oW032770@svn.freebsd.org> <AANLkTine9rGq_cM4ruFXYq=-F7cMXcQAr-zKHuWoQs2z@mail.gmail.com> <20101206195327.GD1936@garage.freebsd.pl> <201012061518.49835.jhb@freebsd.org> <4CFD514E.8010103@FreeBSD.org> <20101207095137.GC1700@garage.freebsd.pl> <4CFE0B9E.3060808@FreeBSD.org> <20101207110410.GE1700@garage.freebsd.pl>

next in thread | previous in thread | raw e-mail | index | archive | help
On 07.12.2010 13:04, Pawel Jakub Dawidek wrote:
> On Tue, Dec 07, 2010 at 12:25:34PM +0200, Alexander Motin wrote:
>> It is really nice that we support bigger sector sizes. But unluckily we
>> are not the only OS in universe. Disks with data may move between
>> systems, partition could be shared, etc. We must keep compatibility --
>> period. Can you predict what happen if we try to use some FAT partition
>> created by Windows (using 512bytes sectors) after we set disk sector
>> size to 4K? I have feeling that we won't even read partition table
>> properly, not speaking about FAT. Even GEOM classes supporting big
>> sector sizes depend on that size to be constant -- otherwise they will
>> just be unable to locate their own metadata in last sector.
>
> First valid argument, thank you:)
>
> BTW. What Ivan did changes ashift for existing ZFS pools as well, so it
> breaks them too.

I can't say anything about it. Ivan told me that it's not. You may 
discuss it between yourselves and I'll listen. :)

> If we decide to align other things to stripesize we can still break
> compatibility with other operating systems.

Not necessary. Some places indeed may have some legacy requirements, for 
example, in theory MBR want partition to be aligned to "track boundary" 
(but I've seen many pre-formatted SD cards with MBR violating it to 
align partition to flash sector). Same time for BSD label I see no 
problem to align partitions any way we want. I also see no problems to 
make FAT cluster, UFS block/fragment, etc, to match some sizes.

> Also stripesize is really not good idea. For RAID5 it might be like
> 64kB or larger, which is definiately too large for ashift in ZFS or
> fragment size in UFS.

I agree that minimal I/O size of 64K or 128K may be too much. In this 
case I've proposed Ivan to limit maximum used stripesize with some lower 
value, possibly tunable. But he preferred to not introduce new constants.

I think we should not depend that stripesize should be small or big or 
power-of-2 or anything else. World is not uniform. Even for RAID it can 
theoretically vary from 1 sector to half of the disk. We have to 
reconsider it each time we are going to use it, taking to account local 
limitations and preferences.

-- 
Alexander Motin



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4CFE1B0F.90908>