DKNOG12 Rev1b PDF

SRv6
Record-Speed Standardization and Deployment
Pablo Camarillo
[email protected]
DKNOG12 - Copenhagen
Thank you
• Lead operators
• EcoSystem Partners
• Academic Partners
• IETF Partners
• Cisco SR Team
© 2021 Cisco and/or its affiliates. All rights reserved. Cisco Public.
segment-routing.net
• SRv6 uSID: training
• SRv6 Stateless Slicing: demo
• SRv6 Ultra-Scale SR Policy: 26 uSID push at linerate: demo
• BGP PIC Edge with SRv6 Summarization: ISIS UPA: demo
• Path Tracing: demo
Deployment Status
Record-Speed Deployment
• 3 years of commercial deployment (2019-2022)
• ~100M SRv6 subscribers
• ~100 deployments, with ~14 public reports
• Across markets (Web, SP, Enterprise) and geographies (Asia, EU, US)
SoftBank Rakuten Indosat MTN Uganda Noia
Iliad Alibaba China Telecom China Bank Cernet2
Free Bell Canada China Unicom Line

NEW
Alibaba - Full Stack SRv6 Deployment
• “Predictable network services to every single application, based on the full
stack SRv6 innovations across endpoints(container,host,P4 gateway), network
devices and controller/network service center” YuanChao Su, Alibaba
• SID  Service
• Massive IPv6 space (HyperScale)
• Seamless Deployment
− Many use-cases do not require SRH
Co-Development
✓ SRv6 Service Anchors on C8000
✓ 2. The uSID POC in C8000
Commitment to SR Lead-Operators
Standard-Based Technology
Vendor eco-system
Open-Source eco-system
Mature Standardization Much faster standardization than usual
Sign of the SRv6 Industry Endorsement

• Proposed Standard
− RFC 8402 SR Architecture
− RFC 8754 SRv6 DataPlane
− RFC 8986 SRv6 Network Programming
− RFC xxxx SRv6 ISIS Extension
− RFC xxxx SRv6 BGP Extension
− RFC xxxx SRv6 OAM
− RFC xxxx SR Policy
Rich Eco-System
• 25 HW implementations
− Cisco Systems, Nokia, Arrcus, Kaloom, NoviFlow, Huawei, ZTE, Juniper
− Broadcom, Barefoot, Intel SmartNIC, Marvell, Mellanox,
− Spirent, Ixia
− Multiple Interop Reports
• 11 open-source platforms/ Applications

− Linux, FD.io VPP, P4, iptables, nftables, snort, ExaBGP, GoBGP, GoBMP, Contiv-VPP,
SONiC, SAI, SERA
Many Successful Interops
• 2020/04: EANTC: SRv6 interop between Cisco, Huawei, Juniper, Arrcus, Ixia (link)
− Classic IPv6 nodes as SRv6 transit nodes
− SRv6-L3VPN for IPv4 and IPv6 services
− SRv6 TI-LFA FRR link protection with SRH insert
− SRv6 EVPN for E-Line and EVPN L3VPN services
− SRv6 TE SR Policy
• 2021/02: NetOne Systems (link)

− Cisco XR, Cisco NX, Juniper
• 2021/09: EANTC: SRv6 interop between Cisco, Huawei, Juniper, Nokia, Spirent (link)
− SRv6-Based Global IPv4 and IPv6 services
− SRv6-L3VPN for IPv4 and IPv6 services
− SRv6 TI-LFA FRR local SRLG protection with SRH insert
− SRv6 EVPN for E-Line and EVPN L3VPN services
− IGP Flex-Algo using TWAMP-measured link delays
Stateless & Scalable Network Partitioning
Partition 1
R1
Partition 2
• Stateless: DA, PARTITION-ID and DSCP are independent fields in the packet header
• Scalable: Routing, Partitioning & QoS are orthogonal spaces
• Seamless Deployment: PARTITION-ID is seamlessly hashed by legacy devices
• IETF Terminology: Network Resource Partitioning (draft-filsfils-spring-srv6-stateless-slice-id)
Demo Link
1. Select outgoing interface from
IPv6 Destination Address
Partition 1
R1
Partition 2
Demo Link
1. Select outgoing interface from 2. Select partition from
IPv6 Destination Address PARTITION-ID in IPv6 Flow label
Partition
Partition 1 1
Partition 1
R1
Partition 2
Demo Link
1. Select outgoing interface from 2. Select partition from 3. Select QoS queue
IPv6 Destination Address PARTITION-ID in IPv6 Flow label from IPv6 DSCP
Partition
Partition 1 1
Partition 1
Queue 2
R1
Partition 2
Demo Link
SRv6 uSIDs
SRv6 uSID Terminology
• Industry:
− SRv6 Micro Segment
− SRv6 uSID
− Briefly: uSID
• IETF: NEXT-C-SID
− Briefly: Next
− IETF document: draft-ietf-spring-srv6-srh-compression-01
− Training: link
Rakuten – SRv6 uSID in Deployment
• 5G End-to-End Network Slicing based on SRv6 uSID Flex-Algo
− SRv6 uSIDs are allocated from the ULA address range
− SRv6 uSID ISIS Flex-Algo: Low-Cost vs Low-Delay
− SRv6 uSID BGP services
− SRv6 uSID TILFA
− Cisco NCS5500 and NCS-540 series
• Innovation in partnership with Cisco

− BGP PIC Core and Edge with SRv6 Summarization: ISIS UPA (demo)
− SR BW counters for deterministic and scalable capacity planning and BW guarantee
Bell Canada - SRv6 uSID Deployment
• Bell promptly switched from SR-MPLS to SRv6 uSID
• Continued Simplification (remove MPLS dataplane)
• Better Routing Scale: Summarization
• Better HW Scale: linerate 26 uSID push for end-to-end SR Policy
• Seamless Deployment (6 uSID’s in DA without SRH)

• End-to-End IP Unified Dataplane from socket to Internet Peering
− SRv6-TE Policy: topological and service uSID’s
• Service Programming
• Reduce network service costs by up to 90%
footprint by 75%
power consumption by as much as 66%
SRv6 uSID - Feature Parity with SR-MPLS
• TILFA & uLoop Avoidance
• Flex-Algo Low-Cost/Low-Delay with VPN Automated Steering
− Performance Monitoring: Link Latency (for Low-Delay slice)
• SRv6-TE Policy: topological and service uSID’s
• L3VPN (IPv4 and IPv6), IPv4 Internet, IPv6 Internet, PW, BD
• VPN GW to interconnect with legacy VPN
• Seamless Inter-Domain with Summarization
− No need for complex BGP3107 mechanisms
SRv6 uSID - Rich Eco-System
• Cisco, NoviFlow, Arrcus, Nokia, Ciena
• Merchant: Silicon One, Broadcom, Marvell, Barefoot
• Open Source: Linux, FD.io, P4, eBPF, Cillium, SAI
SRv6 uSID – Pure IP Routing
• Classless Routing (CIDR, RFC7608)
“CIDR rules, even within an SR domain. For that reason, the fact that
the bottom 64 bits in the "address" look funny or change is simply
irrelevant. They are invisible to routing (which is done based on the
prefix).…”
Brian Carpenter, former IETF and IAB chair
https://mailarchive.ietf.org/arch/msg/ipv6/37bH3Ag7jaNgKwnZY_mKIxWRwdM/
IP and uSID – Pure IP Forwarding
• Longest-Prefix-Match (LPM)
• Benefits
− Straightforward Implementation
− Support for different uSID sizes in the same uSID list
− Capability to process multiple uSIDs with one single LPM
• Other technologies cannot do this
− An MPLS lookup processes one single label at a time
− All MPLS labels have a fixed 20-bit size
− GSID (CSID-REPLACE) has the same limitations as MPLS
Illustration – LPM flexibility
• 1 single LPM to resolve across different scales of services at PE2
CE Tenant YELLOW
CE Tenant GREEN
2 CE D1
CE D2 Tenant PURPLE
CE D3
FIB entries@2: /48: A few unicast/anycast personalities @ PE2
fc00:0:0200::/48 uNode(2) /64: A few thousands of VRF’s @ PE2

fc00:0:0200:f100::/64 uDT4 (L3VPN-YELLOW)
fc00:0:0200:f200::/64 uDT4 (L3VPN-GREEN)
fc00:0:0200:fff0:D1::/80 uDX2 (L2VPN-PW-PURPLE-D1) /80: 100k of PW’s @ PE2
fc00:0:0200:fff0:D2::/80 uDX2 (L2VPN-PW-PURPLE-D2)
fc00:0:0200:fff0:D3::/80 uDX2 (L2VPN-PW-PURPLE-D3)
Perfect SRv6 Integration
• uSID reuses SRH (RFC8754) without any change
• uSID strictly applies the SRv6 Network Programming (RFC8986)
• uSID can be bound to any Network Programming Instruction
SRv6 uSID offers the best SRv6 Compression
SRv6: 208 bytes
SRv6 uSID: 64 bytes
uSID (NEXT):
average compression 62%
GSID (REPLACE):
average compression 39%
uSID is 59% more efficient than GSID (CSID-Replace)

62/39 = 1.59
Better HW Scale
SRv6 uSID MPLS
Linerate steering into SR Policy of N SID’s (NCS-5700, J2) N=26 N=~12
Consumed counters associated to a remote ISIS node 1 4
Consumed dataplane entries associated to remote ISIS node 1 4
Demo Link
Better Routing Scale
SRv6 uSID MPLS

Unique Nodes in the SR domain 15M-240M 0.8M
Unique Services per node 512k 200k
ISIS Summarization Yes No
BGP3107 complexity tax to scale ISIS Host Routes No Yes
Seamless Deployment
• Many use-cases leverage classic IPinIP
• The outer DA already holds 6 uSIDs !! D. Cai – Alibaba podcast - SRv6 uSID
• SDN coherence
− The SDN controller packs the uSID policy in the outer DA
− The host/access node only sees an opaque DA
• VPN GW to interconnect with legacy VPN
Path Tracing
The exact path from A to M is not known
B F J
A C G M
D H
• 7 possible “valid” ECMP path
− ABFM, ABGM, ACFM, ACGM, ACHM, ADGM, ADHM … E
B F J
A C G M
D H
• 7 possible “valid” ECMP path
− ABFM, ABGM, ACFM, ACGM, ACHM, ADGM, ADHM … E
• The path may be invalid B F J

− Routing or FIB corruption @ B
A C G M
• Timestamp at each hop
• Interface Load at each hop D H
Stamping Trajectory in PT Header
IPv6 header: A to M
PT header: AB,
Payload
A C G M
D H
IPv6 header: A to M
PT header: (AB, Time, Load) BG,

Payload
A C G M
D H
IPv6 header: A to M
PT header: (AB, Time, Load) (BG, Time, Load) GM,
Payload
A C G M
D H
The PT idea
• Stamping in the Packet Header
• Implemented in the most basic HW pipeline
− Linerate for any packet
• Ultra-MTU-efficient: only 3 bytes per hop!
− 12-bit Interface, 8-bit Timestamp, 4-bit Load
• For IPv6, with or without SRH
− MPLS solution also designed
• Interwork with legacy node
Dataplane Encapsulation
• Minimize NPU parsing
• Minimize # of Read/Write
• Minimize depth of Read/Write
• Maximize Read/Write at fixed positions
• Avoid Header Insert/Resize
• Minimize MTU
Minimize HW complexity by leveraging SDN analytics
• Analytics
− translates the list of collected IDs into a path
− deduces the timing and load history at each hop
− Highlights hotspots
SDN App
• Consistency check: FIB (PT) vs RIB (PCE)
• Feedback loop to applications
− Trigger a change of path (SR, MTCP)
− Trigger a change of rate
B
A C G M
D H
Product, Deployment & EcoSystem
• Cisco Shipping in CY22
− PT Demo and Training (link)
• Strong Operator Interest
• Rich Eco-System
• Rich Open-Source
• At IETF: draft-filsfils-spring-path-tracing
Conclusion
Simplicity Always Prevails
43
Simplicity Always Prevails LDP
RSVP-TE
BGP 3107
MPLS
UDP/VxLAN
NSH
44
RSVP-TE
BGP 3107
MPLS
UDP/VxLAN
NSH
Furthermore with more scale
45
RSVP-TE
BGP 3107
MPLS
UDP/VxLAN
NSH
Furthermore with more scale and functionality
46
Stay up-to-date
SRv6 Part III

Coming by
Sept 2022
amzn.com/B01I58LSUO amazon.com/dp/B07N13RDM9
twitter.com/SegmentRouting segment-routing.net
facebook.com/SegmentRouting/ linkedin.com/groups/8266623
[email protected]

DKNOG12 Rev1b PDF

Uploaded by

Copyright:

Available Formats

DKNOG12 Rev1b PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DKNOG12 Rev1b PDF

Uploaded by

Copyright:

Available Formats

SRv6

Record-Speed Standardization and Deployment

SoftBank Rakuten Indosat MTN Uganda Noia

Iliad Alibaba China Telecom China Bank Cernet2

Free Bell Canada China Unicom Line

Sign of the SRv6 Industry Endorsement

• 11 open-source platforms/ Applications

• 2021/02: NetOne Systems (link)

• Innovation in partnership with Cisco

• Seamless Deployment (6 uSID’s in DA without SRH)

Brian Carpenter, former IETF and IAB chair

FIB entries@2: /48: A few unicast/anycast personalities @ PE2

fc00:0:0200::/48 uNode(2) /64: A few thousands of VRF’s @ PE2

SRv6: 208 bytes

SRv6 uSID: 64 bytes

uSID is 59% more efficient than GSID (CSID-Replace)

SRv6 uSID MPLS

Linerate steering into SR Policy of N SID’s (NCS-5700, J2) N=26 N=~12

Consumed counters associated to a remote ISIS node 1 4

Consumed dataplane entries associated to remote ISIS node 1 4

SRv6 uSID MPLS

• The path may be invalid B F J

PT header: (AB, Time, Load) BG,

Furthermore with more scale

Furthermore with more scale and functionality

SRv6 Part III

You might also like