Interconnecting Millions Of Endpoints With Segment Routing
draft-filsfils-spring-large-scale-interconnect-03
The information below is for an old version of the document.
draft-filsfils-spring-large-scale-interconnect-03
Network Working Group C. Filsfils, Ed.
Internet-Draft D. Cai, Ed.
Intended status: Informational S. Previdi
Expires: April 5, 2017 Cisco
W. Henderickx
Alcatel-Lucent
R. Shakir
BT
D. Cooper
F. Ferguson
Level3
S. Lin
Microsoft
T. LaBerge
Cisco
B. Decraene
Orange
L. Jalil
Verizon
J. Tantsura
Ericsson
October 4, 2016
Interconnecting Millions Of Endpoints With Segment Routing
draft-filsfils-spring-large-scale-interconnect-03
Abstract
This document describes an application of Segment Routing to scale
the network to support hundreds of thousands of network nodes, and
tens of millions of physical underlay endpoints. This use-case can be
applied to the interconnection of massive-scale DC's and/or large
aggregation networks. Forwarding tables of midpoint and leaf nodes
only require a few tens of thousands of entries.
Status of This Memo
Filsfils, et al. Expires April 5, 2017 [Page 1]
Internet-Draft Segment Routing October 2016
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on May 1, 2016.
Copyright Notice
Copyright (c) 2014 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3
2. Reference Design . . . . . . . . . . . . . . . . . . . . . . . 3
3. Control Plane . . . . . . . . . . . . . . . . . . . . . . . . . 5
4. Illustration of the scale . . . . . . . . . .. . . . . . . . . 5
5. Optional Designs . . . . . . . . . . . . . . . . . . . . . . . 6
6. Deployment Model . . . . . . . . . . . . . . . . . . . . . . . .7
7. Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . ..7
8. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 8
9. Manageability Considerations . . . . . . . . . . . . . . . . . 8
10. Security Considerations . . . . . . . . . . . . . . . . . . . 8
11. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 9
12. References . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . .9
Filsfils, et al. Expires April 5, 2017 [Page 2]
Internet-Draft Segment Routing October 2016
Filsfils, et al. Expires April 5, 2017 [Page 3]
Internet-Draft Segment Routing October 2016
1 Introduction
This document describes how SR can be used to interconnect millions
of endpoints.
1.1. Terminology
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL
NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL"
in this document are to be interpreted as described in RFC 2119
[RFC2119].
Term Definition
----------- ------------------------------------------------
Agg Aggregation
BGP Border Gateway Protocol
DC Data Center
DCI Data Center Interconnect
ECMP Equal Cost MultiPathing
FIB Forwarding Information Base
LDP Label Distribution Protocol
LFIB Label Forwarding Information Base
MPLS Multi-Protocol Label Switching
PCE Path Computation Element
PCEP Path Computation Element Protocol
PW Pseudowire
SR Segment Routing
TI-LFA Topology Independent - Loop Free Alternative
2 Reference Design
+-------+ +--------+ +--------+ +-------+ +-------+
A DCI1 Agg1 Agg3 DCI3 Z
| DC1 | | M1 | | C | | M2 | | DC2 |
| DCI2 Agg2 Agg4 DCI4 |
+-------+ +--------+ +--------+ +-------+ +-------+
For example, an operator could do the following:
-Independent ISIS-OSPF/SR instance in core (C)
-Independent ISIS-OSPF/SR instance in Metro1 (M1)
-Independent ISIS-OSPF/SR instance in Metro2 (M2)
-BGP/SR in DC1
-BGP/SR in DC2
-Agg routes are redistributed from C to M and from M to DC domains.
Nothing else is distributed
Filsfils, et al. Expires April 5, 2017 [Page 4]
Internet-Draft Segment Routing October 2016
-Same homogenous SRGB throughout the domains (e.g. 16000-23999)
-Allocate unique SRGB sub-ranges to each metro and core domains:
16000-16999 to the core, 17000-17999 to the metro1, 18000-18999 to
the metro2. Specifically, Agg3 is 16003 and the anycast SID for
(Agg3, Agg4) is 16006. DCI3 is 17003 and the anycast SID for (DCI3,
DCI4) is 17006
-Re-use the same SRGB sub-range for each DC: e.g. 20000-23999.
Specifically A and Z are both 20001.
3. Control-plane
It is out of the scope of this document to describe how the SRTE
Policies are computed and programmed at the source nodes.
This section provides a high-level description of an implemented
control-plane.
The service orchestration programs A with a PW to a remote next-hop Z
with a given SLA contract (low-latency path, be disjoint from a
specific core plane, be disjoint from a different PW service, etc.).
A automatically detects that it does not have reachability to Z. It
then automatically sends a PCEP request to an SR PCE for an SRTE
policy that provides reachability to Z with the requested SLA.
The SR PCE is made of two components. A multi-domain topology and a
compute block. The multi-domain topology is continuously refreshed
from BGP-LS feeds from each domain. The compute block implements TE
algorithms designed specifically for SR path expression. Upon
receiving the PCEP request, the SR PCE computes the solution (e.g.
{16003, 16005, 18001} and provides it to A.
The SR PCE logs the request as a stateful query and hence recomputes
another solution upon any multi-domain topology changes that
invalidates the previous solution.
A receives the PCEP reply with the solution. A installs the received
SRTE policy in the dataplane. A automatically steers the PW on that
SRTE policy.
4. Illustration of the scale
1 core domain and 100 leaf domains
Core domain has 200 core nodes. Assume two nodes per each leaf
domain, with specific node segment and anycast segments, it's 300
Filsfils, et al. Expires April 5, 2017 [Page 5]
Internet-Draft Segment Routing October 2016
prefix segments in total.
Assume a core node connects only one leaf domain.
Each leaf domain has 6,000 leaf node segments.
Each leaf-node has 500 endpoints attached, thus 500 adjacency
segments.
In total, it is 3M endpoints per leaf domain.
Network wide scale:
6,000x100=600,000 nodes
6,000x100x500=300M endpoints
Per-node segment scale:
Leaf node segment scale: 6,000 (leaf node segments) + 300 (core node
segments) + 500 (adj segments) = 6,800
Core node segment scale: 6,000 (leaf domain segments) + 300 (core
domain segments) = 6,300
In the above calculation, it didn't count the link adjacency
segments, which is local to the node. Typically it should be <100.
Note, depends on the leaf node FIB capability, we could split the
leaf domain into multiple smaller domains. For the above example, we
can split the leaf domain to 6 smaller leaf domains. So each leaf
node only need to learn 1000 (leaf node segments) + 300 (core node
segments) + 500 (adj segments)= 1,800 segments.
5 Optional Designs
5.1 SRGB size
In the simplified illustrations of this document, we picked a small
homogenous SRGB range of 16000-23999. In practice, a large-scale
design would use a bigger range such as 16000-80000, or even larger.
5.2 Redistribution of Agg routes
The operator might choose to not redistribute the Agg routes into the
Metro/DC domains. In that case, more segments are required to express
an inter-domain path.
For example, A would use an SRTE policy {DCI1, Agg1, Agg3, DCI3, Z}
to reach Z instead of {Agg3, DCI3, Z} in the reference design.
5.3 Sizing of the domains and number of Tiers
The operator is free to choose among a small number of larger leaf
domains, a large number of small leaf domains or a mix of small and
Filsfils, et al. Expires April 5, 2017 [Page 6]
Internet-Draft Segment Routing October 2016
large domains.
The operator is free to use a 2-tier design (Core/Metro) or a 3-tier
(Core/Metro/DC).
5.4 Local Segments to Hosts/Servers
Local segments can be programmed at any leaf node (e.g. Z) in order
to identify locally-attached hosts (or VM's). For example, if Z has
bound a local segment 40001 to a local host ZH1, then A uses the
following SRTE Policy to reach that host: {16006, 17006, 20001,
40001}. Such local segment could represent the NID (Network Interface
Device) device in the context of the SP access network, or VM in the
context of the DC network.
5.5 Compressed SRTE policies
We earlier saw that A could reach Z with a low-latency SLA contract
via the SRTE policy {16001, 16002, 16003, 17006, 20001}.
It is clear that the control-plane solution can install an SRTE
policy {16002, 16003, 17006} at Agg1, collect the Binding SID
allocated by Agg1 to that policy (e.g. 4001) and hence program A with
the compressed SRTE policy {16001, 4001, 20001}.
From A, 16001 leads to Agg1. Once at Agg1, 4001 leads to the DCI pair
(DCI3, DCI4) via a specific low-latency path {16002, 16003, 17006}.
Once at that DCI pair, 20001 leads to Z.
Binding SID's allocated to "intermediate" SRTE policies allow to
compress "end-to-end" SRTE policies.
{16001, 4001, 20001} expresses the same path as {16001, 16002, 16003,
17006, 20001} but with 2 less segments.
Binding SID's also provide for an inherent churn protection.
When the core topology changes, the control-plane can update the low-
latency SRTE policy from Agg1 to the DCI pair to DC2 without updating
the SRTE policy from A to Z.
6 Deployment Model
It is expected that this design be deployed as a green field but as
well in interworking (brown field) with seamless-mpls design (draft-
ietf-mpls-seamless-mpls).
7 Benefits
Filsfils, et al. Expires April 5, 2017 [Page 7]
Internet-Draft Segment Routing October 2016
7.1 Inter-domain interconnection of millions of endpoints
We have illustrated how millions of endpoints across different
domains can be interconnected.
7.2 Simplified operation
We have eliminated two protocols (LDP, RSVP-TE) and have not added
any. The design leverage the core IP protocols: ISIS, OSPF, BGP, PCEP
with straightforward SR extensions.
7.3 Inter-domain SLA
We leverage TILFA sub-50msec FRR upon Link/Node/SRLG failure.
We leverage the optional use of Anycast SID's for further
availability improvement.
We have shown how inter-domain SLA's can be delivered: e.g. latency
vs cost optimized path, disjointness from bacbone planes,
disjointness from other services, disjointness between primary and
backup paths
We note that the existing inter-domain solutions (Seamless MPLS) do
not provide any support for SLA contracts. They just provide a best-
effort reachability across domains.
7.4 Scale
We have eliminated two protocols and not added any. We have
eliminated midpoint states on a per-service basis.
7.5 ECMP
Each policy (intra or inter-domain, with or without TE) is expressed
as a list of segments. As each segment is optimized for ECMP,
therefore the entire policy is optimized for ECMP. The ECMP gain of
anycast prefix segment should also be considered (e.g. 16001 load-
shares across any gateway from L1 leaf domain to Core and 16002 load-
shares across any gateway from Core to L2 leaf domain.
8. IANA Considerations
None
9. Manageability Considerations
TBD
10. Security Considerations
Filsfils, et al. Expires April 5, 2017 [Page 8]
Internet-Draft Segment Routing October 2016
TBD
11. Acknowledgements
We would like to thank Giles Heron, Alexander Preusche and Steve
Braaten for their contribution to the content of this document.
12. References
12.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
12.2. Informative References
[draft-ietf-mpls-seamless-mpls] Leymann, et al., "Seamless MPLS
Architecture", draft-ietf-mpls-seamless-mpls-07, (work in
progress), July 2015
[draft-francois-spring-segment-routing-ti-lfa-01] Pierre Francois, et
al., "Topology Independent Fast Reroute using Segment
Routing", draft-francois-spring-segment-routing-ti-lfa-01,
(work in progress), April 2015
Authors' Addresses
Clarence Filsfils (editor)
Cisco Systems, Inc.
Brussels
BE
Email: cfilsfil@cisco.com
Dennis Cai (editor)
Cisco Systems, Inc.
170, West Tasman Drive
San Jose, CA 95134
US
Email: dcai@cisco.com
Stefano Previdi
Cisco Systems, Inc.
Via Del Serafico, 200
Rome 00142
Italy
Email: sprevidi@cisco.com
Filsfils, et al. Expires April 5, 2017 [Page 9]
Internet-Draft Segment Routing October 2016
Wim Henderickx
Alcatel-Lucent
Email: wim.henderickx@alcatel-lucent.com
Rob Shakir
BT
Email: rob.shakir@bt.com
Dave Cooper
Level 3
Email: Dave.Cooper@Level3.com
Francis Ferguson
Level 3
Email: Francis.Ferguson@level3.com
Tim LaBerge
Cisco
Email: tlaberge@cisco.com
Steven Lin
Microsoft
Email: slin@microsoft.com
Bruno Decraene
Orange
Email: bruno.decraene@orange.com
Luay Jalil
Verizon
400 International Pkwy
Richardson, TX 75081
Email: luay.jalil@verizon.com
Jeff Tantsura
Ericsson
jeff.tantsura@ericsson.com
Filsfils, et al. Expires April 5, 2017 [Page 10]