Messages in this thread | | | Date | Wed, 29 May 2013 11:06:49 +0530 | From | "Srivatsa S. Bhat" <> | Subject | Re: [RFC PATCH v2 00/15][Sorted-buddy] mm: Memory Power Management |
| |
On 05/29/2013 01:38 AM, Phillip Susi wrote: > > On 4/19/2013 3:12 AM, Srivatsa S. Bhat wrote: >> But going further, as I had mentioned in my TODO list, we can be >> smarter than this while doing compaction to evacuate memory regions >> - we can choose to migrate only the active pages, and leave the >> inactive pages alone. Because, the goal is to actually consolidate >> the *references* and not necessarily the *allocations* themselves. > > That would help with keeping references compact to allow use of the > low power states, but it would also be nice to keep allocations > compact, and completely power off a bank of ram with no allocations. >
That is a very good point, thanks! But one of the differences we have to keep in mind is that powering off a bank requires intervention from the OS (ie., OS should initiate the power-off, because we lose the contents on power-off) whereas going to lower power states can be mostly done automatically by the hardware (because it is content-preserving).
But powering-off unused banks of RAM (using techniques such as PASR - Partial Array Self Refresh) can give us more power-savings than just entering lower power states. So yes, keeping allocations consolidated has that additional advantage. And the sorted-buddy design of the page allocator helps us achieve that.
Thanks a lot for your inputs, Phillip!
Regards, Srivatsa S. Bhat
| |