Re: [virtio-dev] [PATCH] Add virtio parameter server device specification

[Hao Chen <chenhaosjtuacm@google.com>]

> Thanks for explaining. Would publishing a k-v store protocol spec for
> use with vsock?

Does "standard protocol over vsock" exist in virtio-spec? We can say
"we use this protocol on vsock for this device, and it works very
well", but I guess it is still different than writing it somewhere in
the spec.

On Mon, May 24, 2021 at 7:38 AM Stefan Hajnoczi <stefanha@redhat.com> wrote:
>
> On Mon, May 17, 2021 at 12:28:14PM -0700, Hao Chen wrote:
> > > Can you share a bit about the reason for defining a new VIRTIO device
> > > instead of running an existing key-value store over virtio-vsock or
> > > virtio-net?
> >
> > The reason for defining a virtio-parameter-server device is we are interacting
> > with some devices in a "key-value store" pattern. For example, we interact with
> > "vehicle device" in Android using this pattern.
> >
> > Ad-hoc protocol over vsock to a k-v store works, but we want to
> > standardize these
> > types of devices so that the VM doesn't need to know the
> > vendor-specific key-value
> > protocols for a specific device implementation.
>
> Thanks for explaining. Would publishing a k-v store protocol spec for
> use with vsock? That way it's a well-known protocol that anyone wishing
> to use k-v store functionality can use.
>
> One nice effect of defining a protocol over vsock instead of a new
> VIRTIO device is that implementation could be as simple as a
> Python/Go/Rust/etc library that's easy to integrate into applications. A
> VIRTIO device is more low-level and complex, making it more costly for
> others to implement.
>
> Stefan