3 ms·
Since it's ESP-based, what are the improvements here over a conventional IPsec VPN? It would also help to know how different it is from OpenVPN, for example, a
by huhtenberg 8y ago
Since it's ESP-based, what are the improvements here over a conventional IPsec VPN?
It would also help to know how different it is from OpenVPN, for example, and other VPN options including WireGuard.
I mean it's nice to see another VPN, but it'd be useful to know right off the bat how it compares to existing options.
- djsumdog 8y agoI was curious about that too. I know there were some WireGuard/OpenVPN benchmarks a while back. It'd be interesting to see these technologies compared for speed.
- gonzo 8y agoThe published wireguard results are questionable, as the reported results 1011Mbps on a documented -Gbps NIC is impossible even before considering the protocol/framing overhead, which is within 2 bytes of IPSec ESP using AES-GCM-128 with the standard nonce. OpenVPN is slow.
- wmf 8y agoThis is polling-based, so it's probably faster. But if you really need the performance I would lean towards VPP since it seems to be more tested.
- gonzo 8y agoFrom the readme: “~2.5 Mpps (or ~5 Gbps of IMIX traffic) per core on a modern CPU” Note CPU unspecified in above. A Xeon-D 1541 (2.1GHz) will do 2.44 Mpps, (Simple IMIX) and a i7-6950X will do 3.27 Mpps running AES-GCM-128 and VPP. A kernel-based Linux implementation will do a bit under 500 Kpps running the same algorithm. The keying algorithm seems interesting, and is vaguely Wireguard-like (would be more if it used a public-private keypair).