Restrictions for lock coarsening?

[Tom Rodriguez]

It was definitely done as a result of benchmarking and there were  
definitely regressions.  I don't have any information about which  
benchmarks showed problems though.  I think it was done early enough  
that other changes and improvements to biased locking might have  
helped.  Maybe that's just wishful thinking on my part.

tom

On Jan 8, 2009, at 2:41 AM, Christian Thalinger wrote:

> On Tue, 2009-01-06 at 15:27 -0800, Tom Rodriguez wrote:
>> It's been a while for me too so my answers aren't the final word  
>> but I
>> don't think there's any implementation reason why a previously
>> unbiasable object couldn't be converted back to a biasable one.  I
>> think it's mainly a matter of policy and performance.  The better
>> solution would be to remove the startup delay all together but I  
>> don't
>> think anyone has a good idea one what it would take to do that.  The
>> original decision was motivated by not allowing any regression, not
>> matter how small, in the startup benchmarks.  It's possible it's  
>> not a
>> big deal now but I don't think anyone has done the investigation to
>> know one way or another.
>
> Have there been any regressions back then?  Or was the delay added
> conservatively?
>
> - Christian
>