<div dir="ltr">On 1 September 2014 09:10, loy wolfe <span dir="ltr"><<a href="mailto:loywolfe@gmail.com" target="_blank">loywolfe@gmail.com</a>></span> wrote:<br><div class="gmail_extra"><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
<div dir="ltr">If the neutron side MD is just for snabbswitch, then I thinks there is no change to be merged into the tree. Maybe we can learn from sriov nic, although backend is vendor specific, but the MD is generic, can support snabb, dpdkovs, ans other userspace vswitch, etc.</div>
</blockquote><div><br></div><div>That is an interesting idea.</div><div><br></div><div>The Snabb mech driver simply asks Neutron/Nova to bind the port with VIF_VHOSTUSER. If this is the requirement for other drivers too then it would seem that we have good potential for sharing code. Perhaps we could rename mech_snabb to mech_vhostuser, like we have already renamed the VIF code.</div>
<div><br></div><div>Going forward we would like the Snabb driver to become more mainstream in the way it manages its agent on the compute host. Currently we use a simple "brute force" approach [1] that is intended to protect us from synchronisation bugs (race conditions) and be compatible with multiple versions of OpenStack (e.g. the one we deploy with + the one we upstream towards). If we did not have to support a production version based on a stable OpenStack release then we might have been less conservative here.</div>
<div><br></div><div>Cheers,</div><div>-Luke</div><div><br></div><div>[1] Snabb NFV architecture: <a href="https://github.com/SnabbCo/snabbswitch/wiki/Snabb-NFV-Architecture">https://github.com/SnabbCo/snabbswitch/wiki/Snabb-NFV-Architecture</a></div>
<div><br></div><div><br></div></div></div></div>