Re: Current state of filesystem features / improvements

[haad <haaaad%gmail.com@localhost>]

Hi,

On Thu, Sep 16, 2010 at 9:33 AM, Niels Dettenbach <nd%syndicat.com@localhost> 
wrote:
> This seems to be a topic from time to time here.
>
> What are the current states of (local) filesystem features for improvements 
> (performance in usage and filesystem-checks).
>
> Which features could be understand "as stable" for typical production use 
> i.e. on internet-servers (webservers, DBs, mailservers, bind etc.)?
>
> Which (ffs) features could be recommended to use (i.e. "softdeps") for 
> productive environments (and especially within xen pv NetBSD guests) to 
> improve performance / faster fsck?
>
> As i used softdeps last time (2 years ago) i got some os crashes (fs errors) 
> within a month.

Yeah because of their instability, Andrew Doran has removed softdeps
from NetBSD source tree some time, ago and they are not going to be
included in 6.0 anymore. I suggest to use wapbl (-o log) option for me
it works without any problems.

For Xen pv disks I would suggest LVM backend, our NetBSD works exactly
same as linux one.

>
> Is there a current "roadmap" for further fs improvements (i.e. ffs 
> improvements or new filesystems)? How does ffs compares to ext4 which seems a 
> real step forward within the linux world - especially regarding performance.

There is a ZFS port ongoing but that can't be considered stable in any way.

> It seems resize_ffs is the only tool to extend ffs (shrinking seems not 
> possible today) - is it reliable enough for productive usage today?

Yes this is one of our weak points, which is much more visible now
when we have a functional LVM in our tree.

>
> Is the relative new software RAID under NetBSD reliable and performant?

AFAIK it's used in production in bunch of servers and works great.

-- 


Regards.

Adam