2 ms·
The page size can't be more than half the L2 cache (else writeback to DRAM would be triggered, dramatically lowering performance). That means you have to use
by londons_explore 5y ago
The page size can't be more than half the L2 cache (else writeback to DRAM would be triggered, dramatically lowering performance). That means you have to use a relatively small page size, and lock contention in the kernel for those pages then limits throughput if you use many cores, even if the data is never read by pv.
See the comments in the code for info...