Okay, it’s faster. What’s your security audit look like? How many lines of code do you have?
How big is your attack surface area? How many large-scale attacks has your code survived?
Wireguard has done so well for so long, in part because it has kept a minimal attack surface area. The code has been audited multiple times. The code is battle-hardened. And fast.
Before I seriously consider anything else as a replacement, I’m going to need to see all of your numbers in this area, as compared to Wireguard.
This isn't the first VPN I've implemented; I gave OpenVPN the multi-peer DCO code that's in ovpn-dco-win and I did work on accelerating pfsense's in-kernel crypto.
WireGuard is not just one implementation; every one has their own version. And it has seen breakages, and it can suffer from resource haemorrhaging. The reason I raise WG up is because people conflate its performance because of its protocol, but don't understand that at the data-path, it's just AEAD with the cipher that's chosen and how that AEAD data plane is implemented.
The tunnel security here is simple (in terms of what it uses):
- it depends on OpenSSL (or AWS-LC) for handshaking TLS and key discovery, which is what openvpn would do too.
- the AEAD data channel is the same in terms of known usage of AES-256-GCM, ensuring that keys are flipped before the threshold of the counter (proto/crypt).
The security elsewhere will be how people deploy it and what trust environment they run it on, ie cert management and key management.
The weakness of stock openvpn is that it allows all connections into the client, that's why I added client-side ACLs to restrict what comes inbound. And also allow explicit setting to block IPv6 immediately (for those who don't want leakage).
Like any new software, people should only test it in an isolated environment first and see what it does. It's new code, so the numbers you are asking for in the prefix is not possible to provide.
More than 10x faster than stock openvpn, and 2-3x faster than DCO.
Faster than wireguard, both Go and kernel versions.
Lower CPU usage.
No BPF/io_uring.
Adds TLS encapsulation (for stealth), WebSocket for proxy, client-side ACLs (openvpn defaults to allow-any), client-side forwarding, pinned-keys, PQ key exchange.
Codebase can be built for Linux, macOS and Windows. GUI uses SDL3. Daemon uses between 5-30MB of memory depending on throughput level.
iperf TCP single stream over the tunnel type, in Gbps. Linux VMs in ESXi with ConnectX-5 SR-IOV pairs. On Linux with direct containers, it's even faster (more than 14Gbps).
Okay, it’s faster. What’s your security audit look like? How many lines of code do you have?
How big is your attack surface area? How many large-scale attacks has your code survived?
Wireguard has done so well for so long, in part because it has kept a minimal attack surface area. The code has been audited multiple times. The code is battle-hardened. And fast.
Before I seriously consider anything else as a replacement, I’m going to need to see all of your numbers in this area, as compared to Wireguard.
The code is there for review.
This isn't the first VPN I've implemented; I gave OpenVPN the multi-peer DCO code that's in ovpn-dco-win and I did work on accelerating pfsense's in-kernel crypto.
WireGuard is not just one implementation; every one has their own version. And it has seen breakages, and it can suffer from resource haemorrhaging. The reason I raise WG up is because people conflate its performance because of its protocol, but don't understand that at the data-path, it's just AEAD with the cipher that's chosen and how that AEAD data plane is implemented.
The tunnel security here is simple (in terms of what it uses): - it depends on OpenSSL (or AWS-LC) for handshaking TLS and key discovery, which is what openvpn would do too. - the AEAD data channel is the same in terms of known usage of AES-256-GCM, ensuring that keys are flipped before the threshold of the counter (proto/crypt).
The security elsewhere will be how people deploy it and what trust environment they run it on, ie cert management and key management.
The weakness of stock openvpn is that it allows all connections into the client, that's why I added client-side ACLs to restrict what comes inbound. And also allow explicit setting to block IPv6 immediately (for those who don't want leakage).
Like any new software, people should only test it in an isolated environment first and see what it does. It's new code, so the numbers you are asking for in the prefix is not possible to provide.
More than 10x faster than stock openvpn, and 2-3x faster than DCO. Faster than wireguard, both Go and kernel versions. Lower CPU usage.
No BPF/io_uring. Adds TLS encapsulation (for stealth), WebSocket for proxy, client-side ACLs (openvpn defaults to allow-any), client-side forwarding, pinned-keys, PQ key exchange.
Codebase can be built for Linux, macOS and Windows. GUI uses SDL3. Daemon uses between 5-30MB of memory depending on throughput level.
iperf TCP single stream over the tunnel type, in Gbps. Linux VMs in ESXi with ConnectX-5 SR-IOV pairs. On Linux with direct containers, it's even faster (more than 14Gbps).
On an A55 ARM NanoPi R5C, wireguard gets around 550Mbps, this can do >700Mbps.