Comment by Veserv
2 months ago
Thank you for benchmarking.
2.5x slower than what they were replacing. Demanding evidence for claims strikes again.
2 months ago
Thank you for benchmarking.
2.5x slower than what they were replacing. Demanding evidence for claims strikes again.
Careful that VSOCK is an abstraction for disparate VM-to-host transports. GP only measured on KVM so if you run e.g. ESXi you’ll need to do your own benchmarks.
vsock isn't a replacement for TCP, because you can't assume that IP exists or is routable / not firewalled between the guest and the host.
Having said that, yes it also really ought to be faster. It's a decent, modern protocol so there's no particular reason for it, so with a bit of tuning somewhere it should be possible.
Couldn’t you just use a broadcast address and get the same result ?
VMs might not have a network connection at all, or (in a more normal secure configuration) have all their network traffic trunked onto a VLAN that avoids touching the host. Vsock is designed so it can only be used for traffic between the hypervisor/host and guests (or between guests on the same host). It's more akin to virtio or hypercalls than a traditional network.