[sldev] Optimization via _set_sbh_threshold

[Nicholaz Beresford]

Dirk,

in some of the threads about sbh I've read about occasional
drops in performance, and given the LBH is the more modern
version I'll definitely look into that direction.

I'm glad you caught this difference/detail, I guess I would
have missed that.


Nick


Dirk Moerenhout wrote:
> Nick,
> 
> indeed. Although I doubt they disabled SBH to promote LFH. I think
> they figured SBH would no longer play a role in todays workloads. I'm
> hence not so sure that SBH holds up when facing heavy load. SBH's
> locking code and ability to handle huge heaps (containing megabytes of
> small items) may be bad as it was written in a different age. They
> started working on LFH when 64MB was considered a lot of RAM. And as
> they turned off SBH they must've figured it would be no longer
> applicable for software with big memory footprints.
> 
> I'm for example not sure how SBH is doing alignment. For LFH it's
> clear that minimum size is 8 bytes and it's known to be ready for 64
> bits environments. I'm pretty sure it'll align well to avoid
> performance hits.
> 
> Dirk aka Blakar Ogre
> 
> On 6/13/07, Nicholaz Beresford <nicholaz at blueflash.cc> wrote:
>>
>> Hmm.... am I getting this right?
>>
>> LFH is an improvement over SBH but needs to be activated and
>> in order to promote it the turned off SBH by default?
>>
>> I'll test more in direction of activating LFH then ...
>>
>>
>> Nick
>>
>>
>>
>>
>>