Rob Bradford

@sdake The performance you get with vhost-user-net is highly dependent on how you configure that backend you talk to with it. DPDK is the most common one and has a...

Okay, let's wait until the kernel of the CI is running a version that integrates this feature.

@jlevon I guess without ioeventfd/ioregionfd the BARs for the NVMe device cannot be mapped into the guest address space as there would be no way to know if the NVMe...

@cloud-hypervisor/cloud-hypervisor-maintainers We've been offered assistance from an Intel team to help improve our user documentation. There is still ongoing discussion around technology (e.g. the possible use of sphinx to collate...

Before proceeding I want consensus from the @cloud-hypervisor/cloud-hypervisor-maintainers team

@liuw @michael2012z Okay, thanks for your feedback. I'll ask the team to work out away to use tooling so that the docs can be kept in the source tree.

Definitely never a TCP stream. Doing it over an existing FD might be tricky for the receiving side.

I think you could prototype having the code pass the fds across for send/receive migration.

I wouldn't support implementing `TcpStream` handling in Cloud Hypervisor - the semantics of the way it works with sockets is quite different to the UNIX socket and the current design...

We intentionally don't want to implement TLS and authentication and TCP socket management in Cloud Hypervisor as that is huge complexity that is a burden for anyone that doesn't need...