[rds-devel] The meaning of MR invalidation

[Or Gerlitz]

Olaf Kirch wrote:
> On Wednesday 13 February 2008 15:48, Or Gerlitz wrote:

>> correct?

> Roughly. We don't allocate all FMRs right away, and the occasions
> when we trigger an unmap depends more on the amount of memory pinned
> than the actual number of FMRs.

OK, I think that with these assumption at hand, the design I suggested 
is close enough to what you would be doing so lets review it (next email...)

> Actually, it's also possible to grab a FMR off the dirty list.

This I am not sure to understand, as you said that basically there's a 
requirement to use fmr exactly ones before it gets unmapped, no?

Or