g1: remembered set scan costs very high w/ LRU cache type of behavior

[Jon Masamitsu]

Peter Schuller wrote On 02/27/10 16:17,:

>>I'll see about building from source to maks sure.
>>    
>>
>
>I see the same behavior (RS scan times begin growing at ~ 1 gb heap)
>with a JDK7 built from sources cloned from hg just now (an hour or so
>ago).
>
>  
>
Thanks for doing the experiment.  We'll have to take a closer
look.