We have some swap related topics we'll like to discuss at mm summit. We'll like to get everyone's opinions to guide future swap related work that we have in mind. 1. DAX swap - For swap space on very fast solid state block devices, the swapped pages can be accessed directly using DAX mechanism without incurring the overhead to allocate a page in RAM and swap them in. The direct access swap space should speed things up in many cases. One remaining issue is if the pages are accessed frequently, we may still need to promote them back to RAM. We’ll like to discuss several possible approaches and their pros and cons to see what is the most viable: (i) Using performance monitoring unit to measure the access frequency (ii) Extend the LRU list to such DAX swap space (iii) A page scanning daemon 2. Improving Swap Read Ahead - The current swap read ahead is done in the same order that the pages are swapped out. However, the order of page access may have no relations with the order that the pages are accessed, especially for sequential memory access. We’ll like to discuss detection mechanism for sequential memory access and using a VMA based read ahead for such case. 3. Improving Swap out path - Optimization of the swap out paths by reducing the contentions on the locks on swap device, radix tree by introducing finer grained lock on cluster, splitting up the radix tree and getting swap pages and releasing swap pages in batches. We'll like to address any issues if our proposed patchset has not been merged by the time of mm summit. 4. Huge Page Swapping - Now, the transparent huge page (THP) will be split before swapping out and collapsed back to THP after swapping in. This will waste CPU cycles and reduce effectiveness (utilization) of THP. To resolve these issue, we propose to avoid splitting THP during swap out/in. At the same time, this give us the opportunity to further optimize the performance of THP swap out/in with large read/write size and reduced TLB flushing etc. to take advantage of the new high speed storage device. Thanks. Ying Huang & Tim Chen -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
