[OpenIndiana-discuss] What happens when a ZIL drive dies?

[Bob Friesenhahn]

On Thu, 7 Jun 2012, Jesus Cea wrote:
>>
>> The implementation is obviously very much simplified by using
>> discrete contiguous ZILs.  Otherwise there is a lot more complexity
>> and activity to track which I/Os have completed.  With the current
>> approach, only one write is necessary to record that a batch of
>> writes have been committed to the pool.
>
> Uhm.... and how is it done when you don't have a LOG device?. That is,
> when the ZILs are in the harddisk?. There moving the disk arm should
> be something to avoid/reduce.

As discussed at length previously on the zfs-discuss list, rotational 
latency is the real performance killer with ZIL on a hard disk. Even 
if the head does not move at all, it is necessary to wait for the hard 
disk to spin back around before the next write can be committed.

This list is really not the best place for zfs discussions.  There are 
already two existing lists for such discussions.

Bob
-- 
Bob Friesenhahn
bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/