Hi co-authors of draft-ietf-nvo3-overlay-problem-statement,

According to the suggestions made by some co-authors of 
draft-ietf-nvo3-overlay-problem-statement during the IETF85 meeting, I would 
like to suggest the following text about the path optimization issue to be 
added into the problem statement draft:

"VM migration across data centers usually require a LAN where the VM is located 
to be extended across these data centers. Since the LAN has been extended 
across multiple data center locations, the IP subnet associated with this LAN 
is also extended across these locations. As such, the traffic to/from the 
extended subnet (e.g., the traffic between cloud user sites and cloud data 
centers) would encounter suboptimal routing issues as described in the 
following sub-sections. Such suboptimal routing issue results in not only an 
unnecessary consumption of the bandwidth resources dedicated for data center 
interconnect, but also a possible decrease of cloud users' experiences due to 
increased path latency. Note that here the traffic to/from the extended subnet 
refers to L3VPN traffic between a remote L3VPN site (e.g., a cloud user site) 
and cloud data centers, rather than Internet traffic to/from cloud data 
centers. 

1.1. Suboptimal Routing for Incoming Traffic

Since the IP subnet has been extended across multiple data center locations, 
the location semantics of such subnet is lost. As a result, the incoming 
traffic towards a given server within the extended subnet may travel through 
suboptimal paths since the traffic is forwarded based on the corresponding 
subnet route. For instance, assume a server within a given subnet which has 
been extended across data center X and Y, is physically located at data center 
X, the incoming traffic towards that server may be forwarded to a default 
gateway router of that extended subnet located at data center Y before entering 
that subnet.

1.2. Suboptimal Routing for Outgoing Traffic

To avoid modifying a VM's default gateway address when migrating such VM across 
data centers, VRRP would usually be deployed between default gateway routers at 
different data center locations and therefore only one default gateway router 
would be elected as the VRRP Master for a given subnet that has been extended 
across these data center locations. As a result, assume one default gateway 
router at data center X is elected as the VRRP Master of a given subnet, the 
outgoing traffic originated from a given server of that subnet which is located 
at data center Y would have to be forwarded to that VRRP master at data center 
X through the extended LAN, which in turn forwards the traffic out of that 
extended subnet."

Best regards,
Xiaohu
_______________________________________________
nvo3 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nvo3

Reply via email to