SRv6 Deployment
ConsiderationCAICTChinatianhui@caict.ac.cnCAICTChinazhaofeng@caict.ac.cnChina TelecomChinaxiechf.bri@chinatelecom.cnChina UnicomChinalitong@chinaunicom.cnChina UnicomChinamajc16@chinaunicom.cnHuawei TechnologiesChinapengshuping@huawei.comHuawei TechnologiesChinalizhenbin@huawei.comHuawei TechnologiesChinaxiaoyaqun@huawei.comSRv6 has significant advantages over SR-MPLS and has attracted more
and more attention and interest from network operators and verticals.
Smooth network migration towards SRv6 is a key focal point and this
document provides network design and migration guidance and recommendations on
solutions in various scenarios. Deployment cases with SRv6 are also
introduced.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.SRv6 is the instantiation of Segment Routing deployed on the IPv6
data plane. Therefore, in order to
support SRv6, the network must first be enabled for IPv6. Over the past
several years, IPv6 has been actively promoted all over the world, and
the deployments of IPv6 have been ever-increasing which provides the
basis for the deployments of SRv6.With IPv6 as its data plane, for network migration towards SRv6, both
software and hardware need to be upgraded. Compared with other new
protocols, only IGP and BGP need to be extended to support SRv6, which
significantly simplifies the software upgrade required. While the
hardware needs to support the new SRv6 header SRH, the design of
SRv6 assures compatibility with the existing IPv6 network as an SRv6 SID
is designed as a 128-bit IPv6 address and the encapsulation of an SRv6
packet is the same as an IPv6 packet. When only L3VPN over SRv6 BE
(Best-Effort) is deployed, there will be no SRH. Therefore, no
additional hardware capabilities are required but only software upgrade
for protocol extensions.As the number of services supported by SRv6 increase, e.g. SFC,
network slicing, iOAM etc., more SIDs in the SRH may impose new
requirements on the hardware. Besides upgrading the hardware, various
solutions have
already been proposed to relieve the imposed pressure on the hardware,
such as Binding SID (BSID) etc. to guarantee the compatibility with the
existing network. On the other hand SRv6 has many more advantages over
SR-MPLS for the network migration to support new services.This document summarizes the advantages of SRv6 and provides network
migration guidance and recommendations on solutions in various
scenarios.Compared with SR-MPLS, SRv6 has significant advantages especially in
large scale networking scenarios.The increasing complexity of service deployment is of concern for
network operators, especially in large-scale networking scenarios.
With solutions such as multi-segment PW and Option A , the number of service-touch points has increased,
and the services, with associated OAM features cannot be deployed
end-to-end.With Seamless MPLS or SR-MPLS, since the MPLS label itself does
not have reachability information, it must be attached to a
routable address. The 32-bit host route needs to leak across
domains. For an extreme case, as shown in Figure 1a, in a large
scale networking scenario, millions of host route LSPs might need
to be imported, which places big challenges on the capabilities of
the edge nodes.With SRv6, owing to its native IP feature of route aggregation
as shown in Figure 1b, the aggregated routes can be imported
across network domains. For large scale networking, only very few
aggregated routes are needed in order to start end-to-end
services, which also reduces the scalability requirements on the
edge nodes.In the SR cross-domain scenario, in order to set up end-to-end SR
tunnels, the SIDs in each domain need to be imported to other
domains.With SR-MPLS, SRGB and Node SID need overall network-wide
planning, and in the cross-domain scenario, it is difficult or
sometimes even impossible to perform as the node SIDs in different
domains may collide. BGP Prefix SID can be used for the
cross-domain SID import, but the network operator must be careful
when converting the SID to avoid SID collision. Moreover, the
pre-allocated SRGB within each domain needs to consider the total
number of devices in all other domains, which raises difficulties
for the network-wide planning.With SRv6, owing to its native IP feature of route
reachability, if the IPv6 address space is carefully planned, and
the aggregated routes are imported by using BGP4+ (BGP IPv6), the
services will auto-start in the cross-domain scenario.The MPLS label itself does not hold any reachability information,
so it must be attached to a routable address, which means that the
matching relationship between the label and FEC needs to be maintained
along the path.SR-MPLS uses the MPLS data plane. When the network migrates to
SR-MPLS, there are two ways, as shown in Figure 2:MPLS/SR-MPLS Dual stack: the entire network is upgraded first
and then deploy SR-MPLS.MPLS and SR-MPLS interworking: mapping servers are deployed at
some of the intermediate nodes and then removed once the entire
network is upgradedRegardless of which migration option is chosen, big changes
in a wide area is required at the initial stage therefore causing a
long time-to-market.In contrast, the network can be migrated to SRv6 on demand.
Wherever the services need to be turned on, only the relevant devices
need to be upgraded to enable SRv6, and all other devices only need to
support IPv6 forwarding and need not be aware of SRv6. When Traffic
Engineering (TE) services are needed, only the key nodes along the
path need to be upgraded to support SRv6.With SRv6, the service deployment can be significantly simplified in some scenarios. When the customer of the VPN service carrier (Provider Carrier) is itself a VPN service carrier (Customer Carrier), it becomes the scenario of Carrier's Carrier. For this scenario, with SRv6, the service deployment can be significantly simplified. To achieve better scalability, the CEs of the Provider Carrier (i.e. the PEs of the Customer Carriers) only distribute the internal network routes to the PEs of the Provider Carrier. The customers' routes of the Customer Carriers (i.e. from CE3 and CE4) will not be distributed into the network of the Provide Carrier. Therefore, LDP or Labeled BGP will be run between the CEs of the Provider Carrier (i.e. CE1 and CE2 in the Figure 3) and the PEs of the Provider Carrier (i.e. PE1 and PE2 in the Figure 3), and LDP will be run between the CEs of the Provider Carrier (i.e. the PEs of the Customer Carriers) and the PEs of the Customer Carrier (i.e. PE3 and PE4 in the Figure 3). MP-BGP will be run between the PEs of the Customer Carrier. The overall service deployment is very complex. If SRv6 is deployed by the Customer Carrier and the Provider Carrier, no LDP will be ever needed. The Locator routes and Loopback routes of the Customer Carriers can be distributed into the network of the Provider Carrier via BGP, and within each carrier's network only IGP is needed. The end-to-end VPN services can be provided just based on the IPv6 interconnections, and the customer carrier is just like a normal CE to the provider carrier, which significantly simplified the VPN service deployment. In a MPLS network, generally RSVP-TE is deployed in the P nodes of the network, and LDP is running between these P nodes and the PE nodes. Customers access to VPN services via the PE nodes. This scenario is called LDP over TE, which is a typical deployment for carriers who want to achieve the TE capability over MPLS network while keep scalability. However, such network configuration and service deployment are very complex. With SRv6 which can provide both TE capability and IP reachability, the service deployment can be significantly simplified. Only IGP and BGP are needed in the network to launch VPN services.
Address Planning is a very important factor for s successful network design, especially an IPv6 network, which will directly affect the design of routing, tunnel, and security. A good address plan can bring big benefit for service deployment and network operation.
If a network has already deployed IPv6 and set up IPv6 subnets, one of the subnets can be selected for the SRv6 Locator planning, and the existing IPv6 address plan will not be impacted.
If a network has not yet deployed IPv6 and there has not been an address plan, it needs to perform the IPv6 address planning first taking the following steps,
to decide the IPv6 address planning principlesto choose the IPv6 address assignment methodsto assign the IPv6 address in a hierarchical manner
For an SRv6 network, in the first step for IPv6 address planning, the following principles are suggested to follow, Unification: all the IPv6 addresses SHOULD be planned altogether, including service addresses for end users, platform addresses (for IPTV, DHCP servers), and network addresses for network devices interconnection.Uniqueness: every single address SHOULD be unique.Separation: service addresses and network addresses SHOULD be planned separately; the SRv6 Locator subnet, the Loopback interface addresses and the link addresses SHOULD be planned separately.Aggregatability: when being distributed across IGP/BGP domains, the addresses in the preassigned subnets (e.g. SRv6 Locator subnet, the Loopback interface subnet) SHOULD be aggregatable, which will make the routing easier.Security: fast tracability of the assigned addresses SHOULD be facilitated, which will make the traffic filtering easier. Evolvablity: enough address space SHOULD be reserved for each subset for future service development.Considering the above-mentioned IPv6 address planning principles, it has been adopted in some deployment cases to set Locator length 96bits, function length 20bits, and args 12bits.
When Locator is imported in ISIS, the system will automatically assign END SID with Flavors such as PSP (Penultimate Segment Pop) and distribute the Locator subnet route through ISIS.
The Flavor PSP, that is, SRH is popped at penultimate segment, provides the following benefits,Reduce the load of ultimate segment endpoint. Ultimate segment endpoint tends to have heavy load since it needs to handle the inner IP/IPv6/Ethernet payload and demultiplex the packet to the right overlay service.Support of incremental deployment on existing network where the ultimate segment endpoint is low-end device that is not fully capable of handling SRH.Incremental deployment is the key for a smooth network migration to
SRv6. In order to quickly launch SRv6 network services and enjoy the
benefits brought by SRv6, the recommended incremental SRv6 deployment
steps are given as follows. These are based on practical deployment
experience earned from the use cases described in .The referenced network topology is shown in Figure 5.Step1. All the network devices are upgraded to support IPv6.Step 2. According to service demands, only a set of selected PE
devices are upgraded to support SRv6 in order to immediately deploy SRv6
overlay VPN services. For instance, in Figure 3, PE1 and PE2 are
SRv6-enabled.Step 3. Besides the PE devices, some P devices are upgraded to
support SRv6 in order to deploy loose TE which enables network path
adjustment and optimization. SFC is also a possible service provided by
upgrading some of the network devices.Step 4. All the network devices are upgraded to support SRv6. In this
case, it is now possible to deploy strict TE, which enables the
deterministic networking and other strict security inspection.As the network migration to SRv6 is progressing, in most cases
SRv6-based services and SR-MPLS-based services will coexist.As shown in Figure 6, in the Non-Standalone (NSA) case specified by
3GPP Release 15, 5G networks will be supported by existing 4G
infrastructure. 4G eNB connects to CSG 2, 5G gNB connects to CSG 1, and
EPC connects to RSG 1.To support the 4G services, network services need to be provided
between CSG 2 and RSG 1 for interconnecting 4G eNB and EPC, while for
the 5G services, network services need to be deployed between CSG 1 and
RSG 1 for interconnecting 5G gNB and EPC. Meanwhile, to support X2
interface between the eNB and gNB, network services also need to be
deployed between the CSG 1 and CSG 2.As shown in Figure 6, in most of the current network
deployments, MPLS-based network services may have already existed
between CSG 2 and RSG 1 for interconnecting 4G eNB and EPC for 4G
services.When 5G services are to be supported, more stringent network services
are required, e.g. low latency and high bandwidth. SRv6-based network
services could be deployed between CSG 1 and RSG 1 for interconnecting
5G gNB and EPC.In order to perform smooth network migration, a dual-stack solution
can be adopted which deploys both SRv6 and MPLS stack in one node.With the dual-stack solution, only CSG 1 and RSG 1 need to be
upgraded with SRv6/MPLS dual stack. In this case, CSG 1 can immediately
start SRv6-based network services to RSG 1 for support of 5G services,
but continue to use MPLS-based services to CSG 2 for X2 interface
communications. The upgrade at CSG 1 will not affect the existing 4G
services supported by the MPLS-based network services between CSG 2 and
RSG 1. RSG1 can provide MPLS services to CSG2 for 4G services as well as
SRv6 services to CSG 1 for 5G services.With the current network, the launch of leased line service is slow,
the network operation and maintainence is complex, and the configuration
points are many. SRv6 can solve the issues above. There have already
been several successful SRv6 deployments following the incremental
deployment guidance shown in Section 3.China Telecom Si'chuan (Si'chuan Telecom) has enabled SRv6 at the
PE node of the Magic-Mirror DC in Mei'shan, Cheng'du, Pan'zhihua and
other cities. The SRv6 BE tunnel has been deployed through the 163
backbone network which has the IPv6 capability. It enables the fast
launch of the Magic-Mirror video service, the interconnection of the
DCs in various cities, and the isolation of video services. The
deployment case is shown in Figure 7.As shown in Figure 7, IGP (some cities such as Chengdu
deploy ISIS, while other cities such as Panzhihua deploy OSPF) and
IBGP are deployed between PE and CR, and EBGP is deployed between CRs
of cities in order to advertise the aggregation route. EBGP VPNv4
peers are set up between PEs in different cities to deliver VPN
private network routes.The packet enters the SRv6 BE tunnel from the egress PE of DC, and
the packet is forwarded according to the Native IP of the 163 backbone
network. When the packet reaches the peer PE, the SRH is decapsulated,
and then the IP packet is forwarded in the VRF according to the
service SID (for example, End.DT4).In order to further implement the path selection, ASBRs can be
upgraded to support SRv6. Different SRv6 policies are configured on
the DC egress PE so that different VPN traffic reaches the peer PE
through the 163 backbone network and the CN2 backbone network
respectively.China Unicom has deployed SRv6 L3VPN over 169 IPv6 backbone network
from Guangzhou to Beijing to provide inter-domain Cloud VPN service.
The deployment case is shown in Figure 8. In Guangzhou and Beijing metro networks, routers exchange
basic routing information using IGP(OSPF/ISIS). The prefixes of IPv6
loopback address and SRv6 locator of routers are different, and both
of them need to be imported into the IGP. The 169 backbone is a native
IPv6 network. Between metro and backbone, the border routers establish
EBGP peer with each other, e.g. CR1 with CR2, CR3 with CR4, to form
basic connectivity. All of these constitute the foundation of overlay
services, and have not been changed.PE1 and PE2 establish EBGP peer and advertise VPNv4 routes with
each other. If one site connects to two PEs, metro network will use
multi RD, community and local preference rules to choose one best
route and one backup.After basic routing among networks and VPN routes between the two
PEs are all ready, two PEs encapsulate and forward VPN traffic within
SRv6 tunnel. The tunnel is SRv6 best effort (BE) tunnel. It introduces
only outer IPv6 header but not SRH header into traffic packets. After
encapsulation, the packet is treated as common IPv6 packet and
forwarded to the egress PE, which performs decapsulation and forwards
the VPN traffic according to specific VRF.Guangdong Unicom has also lauched the SRv6 L3VPN among Guangzhou,
Shenzhen, and Dongguan, which has passed the interop test between
different vendors.With SRv6 enabled at the PE devices, the VPN service can be
launched very quickly without impact on the existing traffic. With
SRv6 TE further deployed, more benefits of using SRv6 can be
exploited. There are no IANA considerations in this document.TBD.
The section on the PSP use cases is inspired from the discussions over the mailing list. The authors would like to acknowledge the constructive discussions from Daniel Voyer, Jingrong Xie, etc.. Email:Email:Email: wenhuizhi@huawei.comEmail: huruizhao@huawei.comEmail: maojianwei@huawei.com