Re: [nvo3] Draft NVO3 WG Charter

> -----邮件原件-----
> 发件人: nvo3-bounces@ietf.org [mailto:nvo3-bounces@ietf.org] 代表 Robert
> Raszuk
> 发送时间: 2012年2月18日 0:50
> 收件人: John E Drake
> 抄送: narten@us.ibm.com; Ronald Bonica; david.black@emc.com;
> nvo3@ietf.org; Adrian Farrel; Nitin Bahadur
> 主题: Re: [nvo3] Draft NVO3 WG Charter
> 
> Hi John,
> 
> > [JD]  This is an assertion.  It is also the misses the fact that MPLS
> > is only required to mux/demux packets at the edges of the VPN
> > network.
> 
> Let me observe that at least in some data centers the "edge of VPN" is
> likely to be the hypervisor (with OVS) on the host machine. The reason
> being that VMs residing on such host may belong to different customers.
> 
> Could you describe how existing technologies would work in such case ?
> Are you proposing to turn each host to be a PE ? Or are you worried that
> host has not sufficient forwarding power to act as PE so one would need
> a dedicated hardware to be a PE ?

In addition to the concern about not sufficient forwarding power, host-based L2 overlay will more heavily depends on the core multicast trees to deliver uu/mc/bc traffic since the number of PE nodes will increase compared to the ToR-based L2 overly without doubt. However, keeping too many multicast groups within the core implies a serious scaling challenge on both the control plane and the data plane. Otherwise(if head-end replication is used), host-based L2 overlay will consume more bandwidth compared to ToR-based L2 overlay. Unless there is no mc and bc, and even no uu traffic within data centers...

Best regards,
Xiaohu

> Best regards,
> R.
> 
> _______________________________________________
> nvo3 mailing list
> nvo3@ietf.org
> https://www.ietf.org/mailman/listinfo/nvo3