On 05/16/2015 04:18 PM, Gordan Bobic wrote:
> Hi all,
>
> I have been mulling over an idea and wanted to run it past people here
> to see if there are any strong opinions either way. Speifically, where
> images are provided, I've been pondering making the based on zfs
> (zfs-fuse for now, at least until kernel mode zfs is more tested.
>
> Yes, I am catually proposing/advocating using zfs even for the rootfs.
> I have been using nothing but zfs for all my non-root block devices
> for years now, and have been using it for my rootfs-es wherever
> possible since late last year without any problems. And zfs is one of
> those things that once you try it you simply cannot imagine how you
> ever managed to make do without it.
>
> There is certainly a fair amount of work involved in producing images,
> and most importantly kernels and initramfs-es to go with those images,
> but I think the benefits are worth it (where the kernel available for
> the devices supports at least fuse or at least the kernel sources are
> available.
>
> The main drawback is performance. zfs-fuse is going to be slower than
> ext*, there's no two ways about it. This is certainly going to be
> noticeable for those running with SSDs. It may not be particularly
> noticeable for those running on spinning rust but it will use more
> CPU, and the zfs ARC cache will not be treated as releasable like
> normal page cache under memory pressure (so have swap if you this is a
> problem).
>
> Thoughts?
>

I also recently thought about other filesystems for the images, but my
options were xfs (to align with upstream) or one of the FS-es that
supposedly work better for SD-cards (yaffs? f2fs? ubifs?).
But, as the pi normally boots without an initrd (which I like) and
adjusting it to have an initrd seems non-trivial (according to google) I
decided to leave it at ext4.

Things I'm still considering are nodiratime and data=writeback for ext4.

Jacco
_______________________________________________
users mailing list
[email protected]
http://lists.redsleeve.org/mailman/listinfo/users

Reply via email to