[OmniOS-discuss] upgrade to 151014

[Bob Friesenhahn]

On Fri, 7 Aug 2015, Paul B. Henson wrote:
>
> Would enabling large blocks and bumping the record size to 1M improve
> the efficiency or performance of a filesystem with large files like
> this? Would there be any negative consequences for the small files it
> also contains? Based on the originating bug 5027 it would only really be

With the large blocks, it is necessary to wait for the whole large 
block to be read each time a fresh block is read.  This adds latency. 
The ARC would do caching in units of the large blocks.

The zfs read-ahead algorithm is based on reading zfs blocks and the 
large blocks will slow the acceleration rate (and tuning/resolution) 
of the read-ahead algorithm (but would have an intial head-start).

If the small files are smaller than 1MB or compression is 
enabled, then it seems like there should not be much impact for small 
files.

Be aware that copy-on-write is in units of zfs blocks and so 1MB leads 
to very large copy-on-write operations.  This could harm performance 
if files are updated in place.

The 1MB blocks are interesting to experiment with but I would not use 
them without observing positive impact in real usage.

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