re: Application of L3VPN for DCI//re: Preliminary L3VPN/VPN4DC agenda@ IETF82

[Xuxiaohu <[e-mail]>]

Luyuan,

> -----邮件原件-----
> 发件人: Luyuan Fang (lufang) [mailto:[e-mail]]
> 发送时间: 2011年11月4日 13:21
> 收件人: Pedro Marques; Xuxiaohu
> 抄送: L3VPN WG mailing list
> 主题: RE: Application of L3VPN for DCI//re: Preliminary L3VPN/VPN4DC
> agenda@ IETF82
> 
> Xiaohu,
> 
> VM mobility is certainly in scope for VPN4DC, it would not be much use
> otherwise.
> There are large modern data centers running l3 only today, VM mobility
> goes without say, it works.

Oh, sorry, it's my misunderstanding:)

> Assume your draft is with l2vpn WG, as you called out "IP only L2VPN
> solution".

Virtual Subnet is an approach for subnet extension by the use of L3VPN or ARP proxy technologies. To some extension, subnet extension could be understood as an IP-only L2VPN service due to the following two reasons:1) from the perspective of CE hosts, subnet extension makes them act as if they are located within the same L2VPN instance:2) only IP traffic is allowed to be transported in Virtual Subnet.
 
Since it's an application of L3VPN protocol as a data center interconnection solution, I think it should be talked at L3VPN WG. 

Best regards,
Xiaohu 

> Thanks,
> Luyuan
> 
> > -----Original Message-----
> > From: [e-mail] [mailto:[e-mail]] On Behalf
> > Of Pedro Marques
> > Sent: Friday, November 04, 2011 12:14 AM
> > To: Xuxiaohu
> > Cc: L3VPN WG mailing list
> > Subject: Re: Application of L3VPN for DCI//re: Preliminary
> L3VPN/VPN4DC
> > agenda@ IETF82
> >
> > Xiaohu,
> >
> > On Thu, Nov 3, 2011 at 8:54 PM, Xuxiaohu <[e-mail]> wrote:
> > >> There is a second reason why the data-center network topology is
> > >> virtualized which is the need to support VM moves. This implies
> that
> > >> the IP addresses used for communication between applications may be
> > >> anywhere across the data-center. Using a virtual topology is an
> > >> attractive way to solve this problem.
> > >
> > > I guess VM mobility is not the object for VPN4DC. For VM mobility
> > across data centers, subnet extension at least is required so as to
> > allow the IP addresses used for communication between applications to
> > be anywhere across the data-centers.
> >
> > There is no reason to think in terms of subnets. Given that there is
> > no locality the concept of subnet no longer makes any sense. Thus one
> > has a collection of host routes spread randomly across a large area.
> >
> > It is trivial to make all VM to VM communication routed. If you look
> > at the details in the l3vpn-end-system proposal, packets are routed
> > from the VM to the host-os and by the host-os which encapsulates the
> > packets. The latter has a collection of host routes. Thinking in terms
> > of subnets is not helpful, the concept is just not applicable.
> >
> > Using ARP, proxy or otherwise, is simply self inflicted pain and
> > suffering. There is no reason to care about the mac address of the
> > remote system.
> >
> > regards,
> >   Pedro.