4 ms·
I have been thinking about it for kernel-bypass CPU-bypass packet capture: DMA them directly to a big-ass hugepages-mapped ring buffer, where unprivileged user
by ncmncm 6y ago
I have been thinking about it for kernel-bypass CPU-bypass packet capture: DMA them directly to a big-ass hugepages-mapped ring buffer, where unprivileged user processes can then pick off interesting ones.
I use onload now, from an isolated core, but freeing up that core would be better.
- touisteur 6y agoMmmh. I've been looking at the 'affordable' big-memory (4-6TB) HPE confs for just this kind of 'snapshot to disk every bit from all 10GbE NICs for the previous 10 minutes' feature.