r205231

[Alexander Motin]

On 18.09.2013 14:08, Andriy Gapon wrote:
> Guys,
>
> I would like to ask your opinion of r205231, specifically the part that splits
> the state lists.
> The change is quite large.  I can admit that I do not fully understand it.  And
> it introduces many differences to the upstream code which makes merges quite a
> bit harder.
>
> Unfortunately, Kip is not available for a discussion and I am not sure when he
> will be.
>
> Is there anybody who is in the know about this change?
> Has anyone besides Kip evaluated significance of its impact on performance?
> Should we perhaps consider reverting this change for the sake of staying closer
> to the upstream?  Or, conversely, propose a refinement of this change to the
> upstream?

Not explicitly about this case and almost flaming, but my internal 
(non-scientific) feeling is that every lock in ZFS in congested and 
affects performance. While optimizing block level I was able to reach 
almost a million IOPS, in no way I can push more then 100K random read 
disk IOPS through the ZFS. And unless my memory is leaking exactly 
ARC-related locks were standing on the way according to profiling. If 
you wish to play with it, we could manage you access to my 1M IOPS iron.

-- 
Alexander Motin