Re: Hammer Benchmark Fun

["Justin C. Sherrill" <justin@xxxxxxxxxxxxxxxxxx>]

On Sat, January 8, 2011 2:58 am, Alex Hornung wrote:
> On 08/01/11 06:12, Siju George wrote:
>> Hi,
>>
>> Some one send this to me
>>
>> http://www.phoronix.com/scan.php?page=article&item=dragonfly_hammer&num=1
>> [...]
>>
>> http://en.wikipedia.org/wiki/Phoronix_Test_Suite#Controversies
>
> While I don't want to get into how useful or useless their tests are, it
> would have been nice to see those same tests with a different scheduler,
> to see if it made any impact. Especially in the blogbench case it made a
> huge difference during my initial testing.
>
> Overall I think our I/O performance has improved a lot since previous
> releases and we are at a comparable level to, at least, other BSDs. All
> of this is anecdotal evidence and gut feeling of course :)

Looking at those tests, there was a huge jump in relative performance for
Hammer once they did multi-threaded tests - and this was on uniprocessor
systems!

Of course, comparing a filesystem with infinite snapshots and history
retention to one that doesn't do it at all is never going to work out the
same for speed.  It's like comparing elephants and chimpanzees in a weight
contest.