Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 20 May 2011 16:09:34 +0300
From:      Andriy Gapon <avg@FreeBSD.org>
To:        lev@FreeBSD.org
Cc:        Kirk McKusick <mckusick@mckusick.com>, freebsd-fs@FreeBSD.org
Subject:   Re: Snapshots fail on large FFS2 volumes regulary -- how to backup /usr/home?!
Message-ID:  <4DD6680E.9040006@FreeBSD.org>
In-Reply-To: <795474996.20110520122933@serebryakov.spb.ru>
References:  <1606289061.20110519211755@serebryakov.spb.ru>	<201105200316.p4K3G6EU039569@chez.mckusick.com> <795474996.20110520122933@serebryakov.spb.ru>

next in thread | previous in thread | raw e-mail | index | archive | help
on 20/05/2011 11:29 Lev Serebryakov said the following:
> Hello, Kirk.
> You wrote 20 мая 2011 г., 7:16:06:
> 
>> Given the size of your storage, you should consider using ZFS
>> which is better able to handle such large systems better.
>   Yes, I know, that everybody loves ZFS now, but it doesn't have two
>  characteristics which is important for my installation:
> 
>   (1) nodump flag or any other way to mark directories and files as
>   not-importand for backup. "zfs send" is all-or-nothing solution, and
>   now my users use "nodump" to reduce backup sizes greatly.

Two options:
a) you don't have to zfs send all filesystems, just the ones that you really need;
and you can easily create many filesystems with ZFS; you can tag filesystems that
you do not want to backup with user properties.

b) you can use something else for backups

Besides, zfs send / receive best works for replicating data.  Storing results of
zfs send for later restoration is not a good idea, IMO.

>   (2) Incremental backups with a little of local information (zfs send
>   can send difference between snapshots, but system needs to store old
>   snapshot for this).
>   Second one is not so important yet, because there is a lot of free space,
>   but "zfs send" could not do anything with (1) :(
> 
>     All other backups solutions doesn't store full FS information, as
>   works on file level, not FS one :(

This sounds more like a theoretical than practical objection.
If you don't lose any information that you actually need, then a solution works.
Take a look at e.g. archivers/star.

>> My second suggestion is that you try building UFS2 with 32K
>> blocks and 4K fragments. That will reduce the number of resources
>> needed to take the snapshot.
>   I'll try this. But I remember, that some time ago (about 7.1-STABLE)
>  there was deadlock in kernel memory allocator when different UFSes
>  on system uses different block sizes...
> 


-- 
Andriy Gapon



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4DD6680E.9040006>