From: Tariq Toukan <ttoukan.li...@gmail.com> Date: Tue, 14 Feb 2017 16:56:49 +0200
> Internally, I already implemented "dynamic page-cache" and > "page-reuse" mechanisms in the driver, and together they totally > bridge the performance gap. I worry about a dynamically growing page cache inside of drivers because it is invisible to the rest of the kernel. It responds only to local needs. The price of the real page allocator comes partly because it can respond to global needs. If a driver consumes some unreasonable percentage of system memory, it is keeping that memory from being used from other parts of the system even if it would be better for networking to be slightly slower with less cache because that other thing that needs memory is more important. I think this is one of the primary reasons that the MM guys severely chastise us when we build special purpose local caches into networking facilities. And the more I think about it the more I think they are right. One path I see around all of this is full integration. Meaning that we can free pages into the page allocator which are still DMA mapped. And future allocations from that device are prioritized to take still DMA mapped objects. Yes, we still need to make the page allocator faster, but this kind of work helps everyone not just 100GB ethernet NICs.
