[infinispan-dev] Lock amortization preliminary performance numbers

[Vladimir Blagojevic]

Hi Bryan,

See how it works with thread pool especially if you use one of the lock free structures to create a pool. My bet is that a ConcurrentLinkedQueue would be a good candidate. I'll commit BufferedConcurrentHashMap probably in a week or so. Ideally we can then compare BufferedConcurrentHashMap with thread pool approach.  

Regards,
Vladimir


On 2010-01-27, at 7:00 PM, Bryan Thompson wrote:

>  
> So, overall, this approach is working out very nicely for us in combination with WeakReferences.  One drawback of the approach is that hard references remain in the thread-local arrays until they are batched through the lock.  Depending on how threads are managed, it is possible for hard references to remain in the thread local queues long past their last touch.  For this reason we do not use this technique when the cache contains expensive resources where we need a better guarantee of timely discard when they resource is no longer in use by the  application.  It might be possible to address this concern with a protocol to attach and detach the thread-local buffers from threads on their way into / out of a thread pool.
>  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/infinispan-dev/attachments/20100128/547eb4a7/attachment-0002.html