Doable or a “Bridge Too Far”?

Berislav Todorovic, Juniper Networks

[email protected]
Agenda
• Intro – the IPv6 prelude and the MPLS fugue ...
• IPv6 and MPLS – current solutions for the IPv6 edge of an IPv4 core:
• 6PE, 6VPE, IPv6-over-TE tunnels ... or native IPv6
• MPLS over an IPv6-only core – MPLS transport challenges:
• LDPv6, RSVP-TEv6, BGP-LU(v6), SPRINGv6
• MPLS service challenges in an IPv6 core:
• L3 VPN
• L2 VPN – VPLS/VPWS/EVPN
• MVPN

Can we get rid of IPv4

actually ???
Prelude – IPv6 Takes up ...
• SPs introduce IPv6 to please the nerds because they have no other choice (out of “v4 stock”)
• RFC 6540 - IPv6 Support Required for All IP-Capable Nodes (2012!)
• ripe-554 - Requirements for IPv6 in ICT Equipment (2010-2012)
• IAB Statement on IPv6 (2016):
• We recommend that all (future) networking standards assume the use of IPv6,
and be written so they do not require IPv4.

Will the Internet world

(eventually?) default
to IPv6?
Fugue - MPLS is Still Out There ...
• MPLS - invented in the late 1990s
• Envisaged for an odd use case (ATM/IP interworking).
• But we use it for other reasons today:
• MPLS VPN
• Traffic Engineering
• Fast traffic restoration (FRR)
• Can’t we do it with: overlays, underlays ... ^.*lays$ ...
• MPLS doesn’t give up that easy, like its A** predecessor:
• Mobile operators just love it! 
• Reason – SDH-grade fast restoration, TE ...
Problem – it was developed assuming an IPv4 core
as its foundation!
Prelude Again –IPv4 Packet Forwarding
Although, IPv6 is the same thing ... Destination Intf
10.1.1.1 11
Destination Intf 11.2.2.2 21
14.3.3.3 1 14.3.3.3 31
11.2.2.2 2
10.1.1.1 3 11
IP 10.1.1.1 31
3
1
Destination Intf 21
10.1.1.1 1 2
11.2.2.2 2 IP 10.1.1.1
14.3.3.3 3

1 11.2.2.2
14.3.3.3 3
2
IP 10.1.1.1
Prelude Again –IPv6 Packet Forwarding
Destination Intf
Destination Intf 2001:a:b::/48 11
2a01:c::/32 2 2a01:c::/32 21
2b01:80:1::/56 1 2b01:80:1::/56 31
2001:a:b::/48 3 11
IP 2001:a:b:8::1 31
3
1
Destination Intf 21
2001:a:b::/48 1 2
2a01:c::/32 2 IP 2001:a:b:8::1
2b01:80:1::/56 3

1 11.2.2.2
14.3.3.3 3
2
IP 2001:a:b:8::1
Interlude – The Label ... RFC3031 just says “IP Packet”
Can be BOTH v4/v6 ...
Add a 4 byte tag to each IP packet ... ... or even an Ethernet frame (L2 packet)!

L2 Header IP Packet

Label (20 bits) CoS S TTL

L2 Header MPLS Header IP Packet

4 bytes
Fugue – MPLS Label Switching
Classical IP Forwarding (again IPv4 or IPv6 – doesn’t matter ... yet ...)
50 10.1.1.1
20
IFin LB-in IFout Destination
31 20 11 10.1.1.1
IFin LB-in IFout LB-out
1 50 3 20 11
31
3
1
Can this be IPv6 too? 21
Of course, but ... 2

1 11.2.2.2
14.3.3.3 3
50
2
10.1.1.1
Destination Intf Lb-Out
10.1.1.1 1 50
11.2.2.2 2 52
14.3.3.3 3 57
Label Stacking
Adding more labels to the packets ...

L2 Header
L2 Header IPPacket
Packet IPPacket
Packet

Label (20 bits) CoS S TTL

Label (20 bits) CoS S TTL

L2 Header MPLS Header Packet

4 bytes

L2 Header
L2 Header MPLSL3Header
Ln L2 L1 Packet
Packet

4 bytes
4*n bytes
Label Stack
 Coda - MPLS L3 VPNs – Forwarding Plane
Using the label-stacking technique to create VPNs
11 10.1.1.1 51 11 10.1.1.1
VPN A VPN B
VPN A 51 12 10.1.1.1 Site2
Site 1 Site2
CE – A2
CE1 – A2
Service P P
Label CE – A1
PE 2 CE – B2
VPN B PE 1
Site 1 Transport VPN A
Service Label Site 3
Label P PE 3

CE – B1
12 10.1.1.1 CE – A3
MPLS L2 VPNs – Forwarding Plane
Pushing Ethernet frames from A to B (“pseudowires”)
 Problem – Label Signaling (Control Plane)!
Unsolicited Downstream Label Distribution Label signaling
uses IPv4 !!!
50 10.1.1.1
20
IFin LB-in IFout Destination
31 20 11 10.1.1.1
IFin LB-in IFout LB-out
1 50 3 20 11
31
3
1
21
2
IFin Dest IFout LB-out
3 10.1.1.1 1 50
2 11.2.2.2 2 52
1 11.2.2.2
14.3.3.3 3
50
2

Destination Intf
10.1.1.1 1
11.2.2.2 2
14.3.3.3 3
MPLS VPN Site Signalling (Control Plane)
Exchanging IP network information among VPN sites

? RR
CE
CE
192.0.3.0
/24 A
PE1 PE2 A 192.0.4.0
/22

IP/MPLS
Backbone
MP-BGP
(over IPv4)
CE
100.64.32.0/
24 B PE3 PE4 CE
100.71.0.0
B /23

?
MPLS Control Plane – Signaling ...
• MPLS Transport Signaling – determines transport label values:
• LDP
• RSVP-TE
• BGP-LU
• SPRING (Segment Routing)
• Service Signaling – determines service label values and routing:
• Depends on service
• MPLS L3 VPNs – use MP-BGP
• MPLS L2 VPNs – can use either MP-BGP or LDP
• 6PE / 6VPE (IPv6-over-MPLS) – uses MP-BGP

Problem – most of those protocols assume an IPv4 core!

Even IPv6 services running over MPLS ... WHY?
IPv6 in MPLS Networks – Solutions So Far
• MPLS was created based assuming an underlying IPv4 infrastructure!
• Remember, this was in 1998: <sarcasm>
• No IPv4 address space issues How could we ever live
• No widespread broadband Internet connectivity. without those ... ???
• No smartphones, tables, phablets ... .*blets$ </sarcasm>
• No Instagram, Facebook, Linkedin, Snapchat ...

• Still, MPLS operators pushed to provide IPv6 connectivity since Day 1.

• Standard industry solutions based on an IPv6-agnostic core.
“Do not touch the (IPv4) core, run IPv6 on the edge only!”
• Options:
• Native IPv6 in the core (non-label switched)
• 6PE (RFC4798)
• 6VPE (RFC4659)
6PE (IPv6 Provider Edge) Architecture
(RFC4798)
IPv6-agnostic area

Forwarding plane
MPLS
IPv6 IPv6
IPv4

IPv6
customer MPLS IPv6
core customer

IPv6 Edge (6PE)

Control plane

IGPv4 (OSPF, IS-IS etc.)

EBGP EBGP
between betwen
IPv6 LDP (or RSVP) - between IPv4 endpoints IPv6
endpoints endpoints

MP-iBGP - between IPv4 endpoints

But what if we pull the IPv4 plug ???

IPv4
Why Pull the IPv4 Plug?
• Running an IPv6-only core – why not?
• IPv4 can move to the edges (4PE instead of 6PE).
• Trials done in the past – e.g. Peter Lothberg’s TeraStream:
• https://ripe67.ripe.net/presentations/131-ripe2-2.pdf
• Operationally simple:
• Maintain only one set of ACLs.
• Manage it only via IPv6
• Avoid overlaps on management IP addresses – e.g. for zillions CPEs ...
• Nice thought ... but nightmare for MPLS!
Test Topology
n = circuit number r = router number
IPv6 P2P: 2001:8:8:hex(n)::/64 IPv6 P2P: 2001:8:0::hex(r)/128

1 4 6 r 10 12
cpe1 pre1 n cr1 pre3 cpe3
3 8 18

5 8 9
rr1 rr2

2 7 17 20
6 10 11

4 9 19
cpe2 pre2 cr2 pre4 cpe4
2 5 7 11 13

with permission of the author Krzysztof Szarkowicz, this example was taken from the book (with slight modifications):
A. Sánchez-Monge and K. Szarkowicz, MPLS in the SDN era, 1st ed. O'Reilly Media Inc, 2015.
 Traceroute – IPv6 in Global Routing Table
beri@cpe2> show route table inet6.0 terse
A V Destination P Prf Metric 1 Metric 2 Next hop AS path
* ? 2001:8::/32 B 170 100 65000 I
>2001:8:8:2::2
* ? 2001:222::1/128 D 0 >lo0.0
* ? 2001:444::/32 B 170 100 65000 65444 I
>2001:8:8:2::2

beri@cpe2> traceroute 2001:444::1 source 2001:222::1

traceroute6 to 2001:444::1 (2001:444::1) from 2001:222::1, 64 hops max, 12 byte packets
1 pre2 (2001:8:8:2::2) 141.331 ms 99.055 ms 129.311 ms
2 pre1 (2001:8:8:5::1) 89.134 ms 81.426 ms 56.412 ms
MPLS Label=300160 CoS=0 TTL=1 S=0
3 cr1 (2001:8:8:6::2) 198.055 ms 172.842 ms 184.104 ms
MPLS Label=300448 CoS=0 TTL=1 S=0
4 pre3 (2001:8:8:10::1) 6.077 ms 7.645 ms 5.673 ms
5 2001:444::1 (2001:444::1) 15.286 ms 8.099 ms 10.956 ms
 Traceroute – IPv6 in a VPN
beri@cpe1> show route table inet6.0 terse
A V Destination P Prf Metric 1 Metric 2 Next hop AS path
* ? 2001:db9::1/128 D 0 >lo0.0
* ? 2001:dba::/32 B 170 100 65000 65002 I
>2001:8:f:1::

beri@cpe1> traceroute 2001:dba::1 source 2001:db9::1

traceroute6 to 2001:dba::1 (2001:dba::1) from 2001:db9::1, 64 hops max, 12 byte packets
1 2001:8:f:1:: (2001:8:f:1::) 84.542 ms 87.228 ms 70.557 ms
2 pre1 (2001:8:8:5::1) 83.897 ms 57.609 ms 150.652 ms
MPLS Label=300160 CoS=0 TTL=1 S=0
MPLS Label=16 CoS=0 TTL=1 S=1
3 cr1 (2001:8:8:6::2) 105.846 ms 110.714 ms 107.287 ms
MPLS Label=300448 CoS=0 TTL=1 S=0
MPLS Label=16 CoS=0 TTL=2 S=1
4 2001:8:f:2:: (2001:8:f:2::) 190.653 ms 190.920 ms 67.796 ms
5 2001:dba::1 (2001:dba::1) 35.943 ms 20.132 ms 11.525 ms
However ...
• Many things still rely on IPv4
• Transport signaling (LDPv6) is resolved, however ...
• Service signaling is not:
• None of the vendors support MP-BGP over IPv6 endpoints
• Even for VPNv6 (IPv6 VPN) address family BGP requires IPv4 endpoints!
• L2VPN – both LDP- and BGP-signaled L2VPNs can run over IPv6 infra
• However, most vendors mostly implemented the IPv4-only version.
• MVPN using P2MP LSPs – will require updates for IPv6 ...
• Implementation or standards issue?
• Well ...
RFC7439

[RFC7439] George, W., Ed., and C. Pignataro, Ed.,

"Gap Analysis for Operating IPv6-Only MPLS Networks",
RFC 7439, DOI 10.17487/RFC7439, January 2015,
<http://www.rfc-editor.org/info/rfc7439>.
 +---------+---------------------------------------+-----------------+
| Item | Gap | Addressed in |
RFC7439 +---------+---------------------------------------+-----------------+
| LDP | LSP mapping, LDP identifiers, LDP | [LDP-IPv6] |
Gap Summary | S.3.2.1 | discovery, LDP session establishment, |
| | next-hop address, and LDP TTL |
|
|
| | security | |
+---------+---------------------------------------+-----------------+
| mLDP | Inherits gaps from LDP, RFC 6512 | Inherits |
| S.3.2.2 | [RFC6512] | [LDP-IPv6], |
| | | additional |
| | | fixes TBD |
+---------+---------------------------------------+-----------------+
| GMPLS | RFC 6370 [RFC6370] Node ID derivation | TBD |
| S.3.2.6 | | |
+---------+---------------------------------------+-----------------+
| L2VPN | RFC 6074 [RFC6074] discovery, | TBD |
| S.3.3.1 | signaling | |
+---------+---------------------------------------+-----------------+
| L3VPN | RFC 4659 [RFC4659] does not define a | TBD |
| S.3.3.2 | method for 4PE/4VPE | |
+---------+---------------------------------------+-----------------+
| OAM | RFC 4379 [RFC4379] No IPv6 multipath | [IPv6-RAO] |
| S.3.4 | support, no IPv6 RAO, possible | |
| | dropped messages in IP version | |
| | mismatch | |
+---------+---------------------------------------+-----------------+
| MIB | RFC 3811 [RFC3811] no IPv6 textual | [MPLS-TC] |
| Modules | convention | |
| S.3.5 | | |
+---------+---------------------------------------+-----------------+
Conclusion
• If you run MPLS, you’ll still need IPv4 ...
• Standards will fill the gaps to run MPLS in IPv6-only networks.
• Implementations will certainly follow, but not that quickly.
• Will newer technologies (e.g. SPRINGv6 w/ overlay VPNs) phase out
MPLS?
• Possibly ... but not that fast either ...