zfs discuss - Oct 2008 - Shutting down / exporting zpool without flushing slog devices

Hey folks,

This might be a daft idea, but is there any way to shut down solaris / zfs
without flushing the slog device?

The reason I ask is that we''re planning to use mirrored nvram slogs,
and in the long term hope to use a pair of 80GB ioDrives.  I''d like to
have a large amount of that reserved for write cache (potentially 20-30GB), to
facilitate rapid suspend to disk of multiple virtual machines.

I suspect there will be a few problems with this configuration though.  My first
concern is does ZFS have any limits on the maximum size of slog device it will
work with?  I know there is work being done to tweak the ZIL behaviour.  Does it
take the size of the slog device into account?

And secondly, I''m concerned about how the system will cope with a power
failure with such a large slog.  What I really want to avoid is ZFS flushing
large amounts of data to disk before it can shut down when there is really no
need for it to do so.

Are there any other issues I need to think about?

thanks,

Ross
--
This message posted from opensolaris.org

Ross wrote:
> Hey folks,
>
> This might be a daft idea, but is there any way to shut down solaris / zfs
without flushing the slog device?
>   
panic, sudden power loss, sledgehammer to the motherboard :-)
> The reason I ask is that we''re planning to use mirrored nvram
slogs, and in the long term hope to use a pair of 80GB ioDrives.  I''d
like to have a large amount of that reserved for write cache (potentially
20-30GB), to facilitate rapid suspend to disk of multiple virtual machines.
>
> I suspect there will be a few problems with this configuration though.  My
first concern is does ZFS have any limits on the maximum size of slog device it
will work with?  I know there is work being done to tweak the ZIL behaviour. 
Does it take the size of the slog device into account?
>   
ZFS doesn''t limit the size of the slog device.  If you overrun the
device, then the ZIL will revert to using the pool.  But remember
that the ZIL is a write-only log, except in the case of unexpected
reboot (see above). The actual space used will depend on the amount
of sync writes accumulated before the data is committed in the pool.
As a SWAG, plan for 10 seconds of sync write data -- the default
txg commit interval is 5 seconds, today.  I expect the high water mark
to be on the order of megabytes rather than gigabytes for most folks.
> And secondly, I''m concerned about how the system will cope with a
power failure with such a large slog.  What I really want to avoid is ZFS
flushing large amounts of data to disk before it can shut down when there is
really no need for it to do so.
>   
In general, you should expect no more than 5 seconds of data to
be flushed to the pool.
 -- richard
> Are there any other issues I need to think about?
>
> thanks,
>
> Ross
> --
> This message posted from opensolaris.org
> _______________________________________________
> zfs-discuss mailing list
> zfs-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
>