dari milis meego-dev
dari Arjan van de Ven
On 5/11/2010 4:00, Ameya Palande wrote:
Hi,
I wanted to know why Btrfs is selected as default file system for MeeGo.
we made a positive choice for btrfs for a list of reasons
* It's the future of Linux filesystems. We had a case where the old guard
(ext3) is getting retired, and there are two new filesystmes on the table
(btrfs and ext4). We felt that if we picked ext4, we'd have all the pain of
a new filesystem, and we'd then change again a year later to btrfs.
* BTRFS has a strong focus on data integrity, while many other filesystems
focus mostly on metadata integrity. For most cases this
extra focus is free, however for database applications there is a moderate
performance impact.
This includes things like duplicating key metadata structures on disk
twice, having data checksums, can mark files as critical and for
duplication (RAID1)... but basically the whole COW design (never overwrite
data) means that you never get garbage in files.
* BTRFS supports on-disk compression, giving both a smaller footprint
(factory preload time!) as well as higher throughput on really slow
storage.
* BTRFS has writable snapshots. This opens the door to features in MeeGo 1.1
like atomic package updates (already a Fedora 13 feature with
btrfs) but also the "Restore to factory defaults" becomes easy: just blow
away the snapshot and create a new one and the device is as new.
You can even use it to have "true multiple users" in the system, both users
have the whole device for themselves with a boot time switch
* Performance for small files. Small files are very common, and BTRFS stores
these very efficiently on disk. Whereas other filesystems
generally need upto 3 IOs (and thus 3 seeks if your storage rotates) to
access a file, BTRFS stores everything together and needs only 1
IO.
* Built in defragmentation - performance feature for things like boot time
* Storage pools: Add and remove storage on the fly. (Yes even for phones
this can matter... lets say you have a small flash chip you populate in the
factory; this feature allows you to then expand to a secondary storage
during first use as if it's one big filesystem... no more /opt issues etc)
* ...
well it's a long list of things that made us chose btrfs over its
competition.
Arjan van de Ven wrote:
On 5/12/2010 12:22, Greg KH wrote:
On Tue, May 11, 2010 at 01:09:16PM +0100, Neil McGovern wrote:
On Tue, May 11, 2010 at 02:00:46PM +0300, Ameya Palande wrote:
Btrfs is highly experimental, and THE DISK FORMAT IS NOT YET
FINALIZED. You should say N here unless you are interested in
testing Btrfs with non-critical data.
I would tend to agree, btrfs looks great, but it's just not ready
yet.
ready for what? Looks to be ready for MeeGo :)
yup it is...
blanket "it's not just ready" not based on actual arguments are a bit
useless to me.
we've been using BTRFS in all builds for a long time now, and frankly we
have had more
issues with the ext3 side of the world than with btrfs ;)
2010/6/2 冴羽獠 (Ryo Saeba) <[email protected]>
>
>
> anjrit, warning-nya kok serem gitu. jiper juga gue, apa mending ke xfs aja
> ya.
>
> $ sudo mkfs.btrfs /dev/md0
>
> WARNING! - Btrfs Btrfs v0.19 IS EXPERIMENTAL
> WARNING! - see http://btrfs.wiki.kernel.org before using
>
> fs created label (null) on /dev/md0
> nodesize 4096 leafsize 4096 sectorsize 4096 size 7.28TB
> Btrfs Btrfs v0.19
>
>
>
>
> On Wed, Jun 2, 2010 at 10:41 PM, Andika Triwidada <[email protected]>wrote:
>
>>
>>
>> share nfs ke gadjah :D
>>
>> 2010/6/2 冴羽獠 (Ryo Saeba) <[email protected]>
>>
>> jadi softwareraid + btrfs aja ya? setuju semua?
>>>
>>> Disk /dev/sdh: 2000.4 GB, 2000398934016 bytes
>>> 255 heads, 63 sectors/track, 243201 cylinders
>>> Units = cylinders of 16065 * 512 = 8225280 bytes
>>> Sector size (logical/physical): 512 bytes / 512 bytes
>>> I/O size (minimum/optimal): 512 bytes / 512 bytes
>>> Disk identifier: 0x00000000
>>>
>>> Disk /dev/sdh doesn't contain a valid partition table
>>>
>>> Disk /dev/sdi: 2000.4 GB, 2000398934016 bytes
>>> 255 heads, 63 sectors/track, 243201 cylinders
>>> Units = cylinders of 16065 * 512 = 8225280 bytes
>>> Sector size (logical/physical): 512 bytes / 512 bytes
>>> I/O size (minimum/optimal): 512 bytes / 512 bytes
>>> Disk identifier: 0x00000000
>>>
>>> Disk /dev/sdi doesn't contain a valid partition table
>>>
>>> Disk /dev/sdj: 2000.4 GB, 2000398934016 bytes
>>> 255 heads, 63 sectors/track, 243201 cylinders
>>> Units = cylinders of 16065 * 512 = 8225280 bytes
>>> Sector size (logical/physical): 512 bytes / 512 bytes
>>> I/O size (minimum/optimal): 512 bytes / 512 bytes
>>> Disk identifier: 0x00000000
>>>
>>> Disk /dev/sdj doesn't contain a valid partition table
>>>
>>> Disk /dev/sdk: 2000.4 GB, 2000398934016 bytes
>>> 255 heads, 63 sectors/track, 243201 cylinders
>>> Units = cylinders of 16065 * 512 = 8225280 bytes
>>> Sector size (logical/physical): 512 bytes / 512 bytes
>>> I/O size (minimum/optimal): 512 bytes / 512 bytes
>>> Disk identifier: 0x00000000
>>>
>>> Disk /dev/sdk doesn't contain a valid partition table
>>>
>>> Disk /dev/sdl: 2000.4 GB, 2000398934016 bytes
>>> 255 heads, 63 sectors/track, 243201 cylinders
>>> Units = cylinders of 16065 * 512 = 8225280 bytes
>>> Sector size (logical/physical): 512 bytes / 512 bytes
>>> I/O size (minimum/optimal): 512 bytes / 512 bytes
>>> Disk identifier: 0x00000000
>>>
>>> Disk /dev/sdl doesn't contain a valid partition table
>>>
>>>
>>> --
>>> things left unsaid, http://ryosaeba.wordpress.com
>>> maxgain scams, http://maxgain.wordpress.com
>>>
>>
>>
>>
>>
>
>
> --
> things left unsaid, http://ryosaeba.wordpress.com
> maxgain scams, http://maxgain.wordpress.com
>
>
