3 ms·
Network Stack Specialization for Performance [pdf]
- pixl97 12y agoThis concerns me... >Finally, we evaluated whether Sandstorm handles high packet loss correctly. With 80 simultaneous clients and 1% packet loss, as expected, throughput plummets. FreeBSD achieves approximately 640Mb/s and Sandstorm roughly 25% less. This is not fundamental, but due to FreeBSD’s more fine-grained retransmit timer and its use of NewReno congestion control rather than Reno, which could also be implemented in Sandstorm. Which tells me an attacker can disable your infrastructure more easily with the high performance custom stack by sending specialized malformed data. If you're even thinking about putting a custom stack on the net, its worst possible performance should be the same as low performance stack you're replacing it with. An attacker always targets the weakest point.
- kev009 12y agoSorry but that's a pretty pedantic takeaway. This is high quality academic research at the forefront of CS/Systems Research, not a fancy bubbly website mandating you stop using nginx right now and switch to sandstorm. A "real" system would conceivably converge to be as good as kernel stack, and there are already projects like https://github.com/pkelsey/libuinet https://github.com/pkelsey/libuinet and https://github.com/rumpkernel/netmap-rumptcpip https://github.com/rumpkernel/netmap-rumptcpip where you can share kernel code.