Internet2 IPv6 Connector Changes - Moving Commercial IPv6 to CPS
Caren Litvanyi
lead network engineer
peering team
Internet2 NOC
Joint Techs Meeting
January 2008
Honolulu, HawaiÕi

"What is CPS?"
What is CPS?
How is Internet2Õs IPv6 peering done today?
Why are we proposing changing this?
What are the implications for Connectors?
What about R&E IPv6 peering?
What happens if my organization decides to not participate?
How will moving commercial IPv6 to CPS be implemented?
What is still being worked out?
How can I comment or participate?

CPS Background
CPS is Internet2Õs ÒCommercial Peering ServiceÓ.
ÒThrough CP Service members can leverage their existing Internet2 Network investments to help serve their commercial Internet needs, thereby saving money on commodity Internet charges.Ó
CPS is included in the base connection fee, so it is available for Internet2 Network connectors at no additional cost.
http://www.internet2.edu/network/cp/

CPS Background
Implemented as an ÒoverlayÓ on the existing Internet2 Layer3 Network.
Uses MPLS Layer3 VPN, VRF on same T640 routers.
Currently 4 commercial public peering sites:
PAIX New York 10GE
Equinix Chicago 10GE
PAIX Palo Alto 10GE
SIX (Seattle) 1GE
Also PNI (private peerings).

CPS Background
One of the additional goals of the CPS project was to improve commercial IPv6 connectivity for our R&E Members.
This was hoped to have the additional benefit of promoting the adoption and ÒmainstreamÓ implementation of IPv6.
The original intention was to connect commercial IPv6 peers with the traditional R&E network instance (with which all Connectors already peer).

IPv6 on the Internet2 Network
Internet2 has the allocation 2001:468::/32, and has delegated about 40 /40s to Connectors.
Allocations are documented at:  http://ipv6.internet2.edu/Abilene_Allocations.shtml
The R&E backbone has supported native IPv6 for years, and maintains native IPv6 BGP peerings with many Connectors, government-sponsored research organizations, international R&Es, ISPs, etc.
The IPv6 working group and IPv6 how-to workshops have been sponsored for years as well.

Commodity IPv6 on the Internet2 Network Today
Internet2 purchases a modest amount of IPv6 transit from Global Crossing.
Peers with commercial IPv6-enabled ISPs at PAIX and the NGIX.  These are routed out of Los Angeles and Salt Lake City respectively.
MAX also kindly provides additional commodity IPv6 transit to Internet2 on the east coast.
Internationals pass us some commercial prefixes.
All of the above are routed in inet6.0, the default R&E routing instance.

Why do we want to change this?
CPS gives us a great opportunity to have IPv6 connectivity with more commercial ISPs in more locations!
Exchange points switches generally carry IPv4 and IPv6 in the same VLAN
In Layer3-VPN implementations, an interface (or VLAN subinterface) must be defined to exist in one, and only one, VRF.  That means R&E or CPS, not both.
to separate the two, putting IPv4 in cps.inet.0 and IPv6 in inet6.0 would have taken two connections to each exchange point.
This would be expensive, in port fees and cross-connects.
Other options like leaking routes between instances seemed kludgy and messy to implement.

The Recommendation sent to NTAC
The best thing to do seemed to be to carry the IPv6 prefixes from CPS commercial peers in the CPS VRF - cps.inet6.0.
The CPS VRF would then carry all commercial unicast traffic, IPv4 and IPv6.
New IPv6 peers at the public exchange points.
Move the NGIX and PAIX connections, and transit.
This seems ÒcleanerÓ philosophically too - aligning commercial IPv4 and IPv6 routing.  Easy to explain.
Email sent to NTAC last autumn for comment.

The Implications for Connectors
If Connectors want to benefit from the enhanced commercial IPv6 connectivity provided by Internet2, they will need to establish a second IPv6 connection to Internet2, which connects them to the CPS VRF.
For a connector already participating in CPS, this can be done on that same VLAN, or we can do a second VLAN between the Connector and the Internet2 CPS instance.

The Implications for Connectors
Addressing.
If you have a delegation from Internet2Õs IPv6 block, it particularly makes sense to participate in the CPS IPv6.
In other words, this brings up the usual multi-homing and provider-independent addressing dilemma
We encourage Connectors to take this opportunity to obtain PI address space from ARIN as appropriate, whether participating in IPv6 via CPS or not.
Internet2 is happy to be your primary upstream for IPv6, R&E and commercial, so it is not necessary to get PI space right away if you currently use a delegation from Internet2.

What about R&E IPv6?
This will continue to be well-supported, monitored, and instrumented, just as it is today.
The master instance (inet6.0 in Juniper terms) will continue to carry IPv6 routes for R&E.
Specifically, all Connector-to-Connector traffic will continue to be in the master instance.
Peerings with internationals like GEANT will continue to be in the master instance, as well as ESnet, DREN, etc.
Granted, today, this is still slightly messy, as not all international R&Es have IPv6 policy aligned with their IPv4 policy.  You may get some commercial routes this way.

ButÉ do I have to participate?
No, you do not have to participate at all.  You may continue with just your IPv6 peering with the Internet2 R&E instance, but you will not get commercial prefixes this way in the future.
Note, you may establish an IPv6-ONLY connection to CPS - you do not have to be an IPv4 CPS participant!
If you use Internet2 delegated address space and choose not to participate, you will likely soon have ONLY R&E IPv6 connectivity.

The Implementation
Enable IPv6 in the CPS VRF -- done.
Get Connectors set up next.
This means you will not see much until most Connectors have the new IPv6 peering up with CPS.
This is so we can advertise our /32 and not lose anyone who wants to continue to have commercial IPv6 connectivity via Internet2.
Move current transit and commercial IPv6 peers.
Bring up new IPv6 peers at the public exchanges.

The Implementation Details
Internet2 staff and GlobalNOC staff will contact Connectors about their IPv6 connectivity, open a ticket.
Point-to-point addressing if delegated from Internet2 will look similar:
If you have  2001:468:ff:144::2  today
Your additional address will be:  2001:468:ffff:144::2
We generally plan to reuse ConnectorsÕ same access-lists, MD5, etc., but this can be as you wish.
You decide where you want the routes on your end - some may want them in their own Òcommercial VRFÓ, others, dump them into what is today their R&E IPv6.

What is still being worked out
Feedback from NTAC indicated a desire to maintain universal access to IPv6 performance measurement servers.
The addressing issues may bite us in the end, so we may need some additional changes there.
We do not yet have a specific timetable to implement this migration.

Comments?
Talk to Steve Wallace of Internet2 -- ssw@internet2.edu
Talk with NTAC members (Paul Schopis, chair).
Technical/Implementation details - ask the NOC  noc@net.internet2.edu, or me specifically -- litvanyi@grnoc.iu.edu
Find us here at JT!
Thank you!