SR-TE Path Midpoint RestorationHuawei TechnologiesHuawei Bld., No.156 Beiqing Rd.Beijing100095Chinahuzhibo@huawei.comFutureweiBoston, MAUSAHuaimo.chen@futurewei.comHuawei TechnologiesHuawei Bld., No.156 Beiqing Rd.Beijing100095Chinayaojunda@huawei.comJuniper Networks1194 N. Mathilda Ave.Sunnyvale, CA94089USAcbowers@juniper.netChina Telecom109, West Zhongshan Road, Tianhe DistrictGuangzhou510000Chinazhuyq8@chinatelecom.cnChina Mobile510000Chinaliuyisong@chinamobile.comSegment Routing Traffic Engineering (SR-TE) supports explicit
paths using segment lists containing adjacency-SIDs,
node-SIDs and binding-SIDs.
The current SR FRR such as TI-LFA provides fast re-route protection
for the failure of a node along a SR-TE path by the direct
neighbor or say point of local repair (PLR) to the failure.
However, once the IGP converges,
the SR FRR is no longer sufficient to forward traffic of the path
around the failure, since the non-neighbors
of the failure will no longer have a route to the failed node.
This document describes a mechanism for the restoration of the
routes to the failure of a SR-MPLS TE path after the IGP converges.
It provides the restoration of the routes to an adjacency segment,
a node segment and a binding segment of the path.
With the restoration of the routes to the failure,
the traffic is continuously sent to the neighbor of the
failure after the IGP converges.
The neighbor as a PLR fast re-routes the traffic around the
failure.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
when, and only when, they appear in all capitals, as shown here.Segment Routing Traffic Engineering (SR-TE) is a technology that implements
traffic engineering using a segment list. SR-TE supports the creation
of explicit paths using adjacency-SIDs, node-SIDs, anycast-SIDs, and binding-SIDs.
A node-SID in the segment list defining an SR-TE path indicates a loose
hop that the SR-TE path should pass through.
When the node fails, the network
may no longer be able to properly forward traffic on that SR-TE path. describes
an SR FRR mechanism that provides fast re-route protection for
the failure of a node on a SR-TE path by the direct
neighbor or say point of local repair (PLR) to the failure.
However, once the IGP converges,
the SR FRR is no longer sufficient
to forward traffic of the path around the failure, since the
non-neighbors of the failure will no longer have a route to the failed
node and drop the traffic.To solve this problem,
proposes that a hold timer should be configured on every router in a network.
After the IGP converges on the event of a node failure,
if the node-SID of the failed node becomes unreachable,
the forwarding changes should not be communicated to the forwarding planes
on all configured routers
(including PLRs for the failed node) until the hold timer expires.
This solution may not work for some cases such as
some of nodes in the network not supporting this solution.
This document describes a proxy forwarding mechanism
for the restoration of the
routes to the failure of a SR-MPLS TE path after the IGP converges.
It provides the restoration of the routes to an adjacency segment,
a node segment and a binding segment on a failed node along
the path.
With the restoration of the routes to the failure,
the traffic for the SR-MPLS TE path is continuously sent to
the neighbor of the failure after the IGP converges.
The neighbor as a PLR fast re-routes the traffic around the
failure.Segment Routing.Point of Local Repair.Link State Protocol Data Unit (PDU) in IS-IS.Link State Advertisement in OSPF.Link State, which is LSP or LSA.In the proxy forwarding mechanism, each neighbor of a possible
failed node advertises its SR proxy forwarding capability in
its network domain when it has the capability. This capability
indicates that the neighbor (the Proxy Forwarder) will forward
traffic on behalf of the failed node.
A router receiving the SR Proxy Forwarding capability from the
neighbors of a failed node will send traffic using the node-SID
of the failed node to the nearest Proxy Forwarder
after the IGP converges on the event of the failure.Once the affected traffic reaches a Proxy Forwarder, it
sends the traffic on the post-failure shortest path to the node
immediately following the failed node in the segment list.For a binding segment of a possible failed node, the node
advertises the information about the binding segment,
including the binding SID and the list of SIDs associated with
the binding SID, to its direct neighbors only.
Note that the information is not advertised in the network domain.After the node fails and the IGP converges on the failure,
the traffic with the binding SID of the failed node
will reach its neighbor having SR Proxy Forwarding capability.
Once receiving the traffic, the neighbor swaps the binding SID
with the list of SIDs associated with the binding SID and
sends the traffic along the post-failure shortest path to the
first node in the segment list.
This section describes the semantic of
protocol extensions/re-use for advertising
the SR proxy forwarding capability of a node in a network
domain and the information about each binding segment
(including its binding SID and the list of SIDs associated)
of a node to its direct neighbors.When a node P is able to do SR proxy forwarding
for its neighboring nodes for protecting
the failures of these
nodes, P advertises its SR proxy forwarding capability
for these nodes.
The mirror SID
for a node N (Neighbor of P) advertised by P
indicates the capability of P for N.
For a node X in the network, it learns the prefix/node SID of
node N, which is originated and advertised
by node N.
It creates a proxy prefix/node SID of node N for node P if node P is
capable of doing SR proxy forwarding for node N.
The proxy prefix/node SID of node N for node P is
a copy of the prefix/node SID
of node N originated by node N, but stored under
(or say, associated with) node P.
The route to the proxy prefix/node SID is through
proxy forwarding capable nodes.In normal operations, node X prefers to use the prefix/node SID of
node N. When node N fails, node X prefers to use the proxy prefix/node SID
of node N. Thus node X will forward the traffic targeting to the prefix/node
SID of node N
to node P when node N fails, and node P will do a SR proxy forwarding
for node N and forward the traffic towards its final destination without
going through node N.
Note that the behaviors of normal IP forwarding and routing convergences
in a network are not changed at all by the SR proxy forwarding.
For example, the next hop used by BGP is an IP address (or prefix).
The IGP and BGP converge in normal ways for changes in the network.
The packet with its IP destination to this next hop is forwarded
according to the IP forwarding table (FIB) derived from IGP
and BGP routes.Alternatively, P advertises its capability in its LS.
For OSPF, P advertises its information opaque LSA
with one bit (called PF bit) set to one indicating
that P has the capability for all its neighbors.
For IS-IS, P advertises its LSP with PF bit.
If node P can not do a SR proxy forwarding for all its neighboring nodes,
but for some of them, then
it advertises the node SID of each of the nodes as a proxy node SID,
indicating that it is able to do proxy forwarding for the node SID.For a binding segment (or binding for short) on a node A,
which consists of a binding SID and a list of segments,
node A advertises an LS containing
the binding (i.e., the binding SID and the list of the segments)
in a binding segment TLV.
The LS is advertised only to each of the node A's neighboring nodes.
For OSPFv2, the LS is a opaque LSA of LS type 9
(i.e., a link local scope LSA).
For IS-IS, the TLV is advertised in Circuit Scoped
Link State PDUs (CS-LSP) .
Alternatively, when a protocol (such as
PCE or BGP running on a controller) supports sending a binding
on a node A to A, this protocol may be extended to send the binding
with node A to A's neighbors if the controller knows the neighbors and
there are protocol (PCE or BGP) sessions between the controller and
the neighbors.
This section illustrates the proxy forwarding for
a binding SID through an example.
The proxy forwarding for a node SID and an adjacency SID
can refer to or
Appendix.
is an example network topology used to illustrate
the proxy forwarding mechanism for a binding SID.
Each node N has SRGB = [N000-N999]. RT1 is an ingress node of
SR domain.
RT3 is a failure node.
RT2 is a Point of Local Repair (PLR) node, i.e., a proxy forwarding node.
Label Stack 1 uses a node-SID and a binding SID.
The Binding-SID with label=100 at RT3 represents the ECMP-aware path
RT3->RT4->RT5.
So Label Stack 1, which consists of the node-SID for RT3 following by
Binding-SID=100, represents the ECMP-aware path
RT1->RT3->RT4->RT5.
RT2:10012| |Stack 3| |Stack 2| |Stack 1|
+-----------------+ +--------------+ +-------+ +-------+ +-------+
| RT2:[2000-2999] | |RT2->RT3:20023| | 10012 | | 1003 | | 1003 |
+-----------------+ +--------------+ +-------+ +-------+ +-------+
| RT3:[3000-3999] | |RT3->RT6:30036| | 20023 | | 3004 | | 100 |
+-----------------+ +--------------+ +-------+ +-------+ +-------+
| RT4:[4000=4999] | |RT3->RT7:30037| | 30034 | | 4005 | 100 is
+-----------------+ +--------------+ +-------+ +-------+ binding SID
| RT5:[5000-5999] | |RT3->RT4:30034| | 40045 | to
+-----------------+ +--------------+ +-------+ {30034,40045}
| RT6:[6000-6999] | |RT7->RT4:70074|
+-----------------+ +--------------+
| RT7:[7000-7999] | |RT4->RT5:40045|
+-----------------+ +--------------+]]>If the Point of Local Repair (PLR), for example, RT2,
has the capability to do SR proxy
forwarding for its neighboring nodes such as RT3,
it must advertise this capability.
When RT3 fails, RT2 needs to maintain
its SR proxy forwarding capability
for a period of time. When the proxy forwarding
table corresponding to the fault node is deleted,
the capability is withdrawn. The nodes in the network (for
example, RT1) learn the prefix/node SID advertised by RT3 and
the proxy forwarding capability for RT3
advertised by RT2. When RT3 is normal, the nodes prefer
prefix/node SID. When the RT3 fails, the proxy prefix/node SIDs
of RT3 for RT2 is preferred.For binding-SID 100,
which is associated with segment list {30034, 40045},
RT3 advertises the binding (i.e., 100 bond to {30034, 40045})
to its neighbors RT2, RT4 and RT7.
RT2 as PLR uses the binding to build an entry
for proxy forwarding for binding-SID 100
in its Proxy Forwarding Table for RT3.
The entry is used when RT3 fails.A SR proxy node P needs to build an independent
proxy forwarding table for each neighbor N.
The proxy forwarding table for node N
contains the following information:1: Node N's SRGB range and the difference between the SRGB
start value of node P and that of node N;2: Every adjacency-SID of N and Node-SID of the node pointed to by
node N's adjacency-SID.3: Every binding-SID of N and the label stack associated with
the binding-SID.Node P (PLR) uses a proxy forwarding table based on the next
segment to find a node N as a backup forwarding entry to the
adjacency-SID
and Node-SID of node N. When node N fails, the proxy forwarding table
needs to be maintained for a period of time, which is recommended for
30 minutes.Node RT3 in
is node N, and node RT2
is node P (PLR). RT2 builds the proxy forwarding table for
RT3. RT2 calculates the proxy forwarding table for RT3,
as shown in .
This Section shows through example
how a proxy node uses the SR proxy
forwarding mechanism to forward traffic to the destination node
when a node fails and the
next segment of label stack is a binding-SID.As shown in ,
Label Stack 1 {1003, 100} represents
SR-TE loose path RT1->RT3->RT4->RT5,
where 100 is a Binding-SID, which represents
segment list {30034, 40045}.When the node RT3 fails, the proxy forwarding SID implied or advertised by the
RT2 is preferred to forward the traffic of the RT1 to the PLR node RT2.
Node RT2 acts as a PLR node and uses Binding-SID to query the proxy
forwarding table locally built for RT3. The path returned is the
label forwarding path to RT3's next hop node (RT4), which bypasses
RT3. The specific steps are as follows:a. RT1 swaps label 1003 to out-label 2003 to RT3.b. RT2 receives the label forwarding packet whose top label of
label stack is 2003, and searches for the local Routing Table, the
behavior found is to lookup Proxy Forwarding table due to RT3
failure.c. RT2 uses Binding-SID:100 (label 2003 has pop) as the in-label to
lookup the Next Label record of the Proxy Forwarding Table, the
behavior found is to swap to Segment list {30034, 40045}.d. RT2 swaps Binding-SID:100 to Segment list {30034, 40045}, and
uses the 30034 to lookup the Next Label record of the Proxy
Forwarding table again. The behavior found is to forward the packet to RT4.e. RT2 queries the Routing Table to RT4, using primary or backup path
to RT4. The next hop is RT7.f. RT2 forwards packets
to RT7. RT7 queries the local routing table to forward the packet to RT4.The extensions to OSPF and IS-IS described in this
document result in two types of behaviors in data plane
when a node in a network fails.
One is that for a node, which is a upstream (except for
the direct upstream) node of the failed node along a SR-TE
path, it continues to send the traffic to the failed node
along the SR-TE path for an extended period of time.
The other is that for a node, which is the direct upstream
node of the failed node, it fast re-routes the traffic
around the failed node to the direct downstream node
of the failed node along the SR-TE path.
These behaviors are internal to a network and should not
cause extra security issues.The authors would like to thank Peter Psenak,
Acee Lindem, Les Ginsberg, Bruno Decraene and Jeff Tantsura
for their comments to this work.This Section shows through example
how a proxy node forward traffic to the destination node
when a node fails and the
next segment of label stack is an adjacency-SID or node-SID.As shown in ,
Label Stack 3 {10012, 20023, 30034, 40045}
uses only adjacency-SIDs and represents
the SR-TE strict explicit path
RT1->RT2->RT3->RT4->RT5.
When RT3 fails, node RT2 acts as a PLR, and uses next adjacency-SID (30034)
of the label stack to lookup the proxy forwarding table built by RT2
locally for RT3.
The path returned
is the label forwarding path to RT3's next hop node RT4, which
bypasses RT3. The specific steps are as follows:a. RT1 pops top adjacency-SID 10012, and forwards the packet to RT2;b. RT2 uses the label 20023 to identify the next hop node RT3, which
has failed. RT2 pops label 20023 and
queries the Proxy Forwarding Table corresponding to RT3 with
label 30034. The query result is 2004. RT2 uses 2004 as the incoming
label to query the label forwarding table. The next hop is RT7,
and the incoming label is changed to 7004.c. So the packet leaves RT2 out the interface to RT7 with label
stack {7004, 40045}. RT7 forwards it to RT4, where the original path
is rejoined.d. RT2 forwards packets
to RT7. RT7 queries the local routing table to forward the packet to RT4.As shown in ,
Label Stack 2 {1003, 3004, 4005}
uses only node-SIDs and represents the ECMP-aware path
RT1->RT3->RT4->RT5, where
1003 is the node SID of RT3.When the node RT3 fails, the proxy forwarding TLV advertised by the
RT2 is preferred to direct the traffic of the RT1 to the PLR node RT2.
Node RT2 acts as a PLR node and queries the proxy forwarding table
locally built for RT3. The path returned is the label forwarding
path to RT3's next hop node RT4, which bypasses RT3. The specific
steps are as follows:a. RT1 swaps label 1003 to out-label 2003 to RT3.b. RT2 receives the label forwarding packet whose top label of
label stack is 2003, and searches for the local Routing Table, the
behavior found is to lookup Proxy Forwarding table due to RT3
failure, RT2 pops label 2003.c. RT2 uses 3004 as the in-label to lookup Proxy Forwarding table,
The value of Map Label calculated based on SRGBDiffValue is 2004.
and the query result is forwarding the packet to RT4.d. Then RT2 queries the Routing Table to RT4, using the primary or backup
path to RT4. The next hop is RT7.e. RT2 forwards the packet
to RT7. RT7 queries the local routing table to forward the packet to RT4.f. After RT1 convergences, node SID 1003 is preferred
to the proxy SID implied/advertised by RT2.