| RIFT Working Group | T. Przygienda, Ed. |
| Internet-Draft | Juniper Networks |
| Intended status: Standards Track | A. Sharma |
| Expires: September 2, 2018 | Comcast |
| A. Atlas | |
| J. Drake | |
| Juniper Networks | |
| Mar 01, 2018 |
RIFT: Routing in Fat Trees
draft-przygienda-rift-05
This document outlines a specialized, dynamic routing protocol for Clos and fat-tree network topologies. The protocol (1) deals with automatic construction of fat-tree topologies based on detection of links, (2) minimizes the amount of routing state held at each level, (3) automatically prunes the topology distribution exchanges to a sufficient subset of links, (4) supports automatic disaggregation of prefixes on link and node failures to prevent black-holing and suboptimal routing, (5) allows traffic steering and re-routing policies, (6) allows non-ECMP forwarding, (7) automatically re-balances traffic towards the spines based on bandwidth available and ultimately (8) provides mechanisms to synchronize a limited key-value data-store that can be used after protocol convergence to e.g. bootstrap higher levels of functionality on nodes.
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 https://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 September 2, 2018.
Copyright (c) 2018 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 (https://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.
Clos and Fat-Tree have gained prominence in today's networking, primarily as result of the paradigm shift towards a centralized data-center based architecture that is poised to deliver a majority of computation and storage services in the future. Today's routing protocols were geared towards a network with an irregular topology and low degree of connectivity originally but given they were the only available mechanisms, consequently several attempts to apply those to Clos have been made. Most successfully BGP [RFC7938] has been extended to this purpose, not as much due to its inherent suitability to solve the problem but rather because the perceived capability to modify it "quicker" and the immanent difficulties with link-state based protocols to perform in large scale densely meshed topologies.
In looking at the problem through the lens of its requirements an optimal approach does not seem however to be a simple modification of either a link-state (distributed computation) or distance-vector (diffused computation) approach but rather a mixture of both, colloquially best described as "link-state towards the spine" and "distance vector towards the leafs". In other words, "bottom" levels are flooding their link-state information in the "northern" direction while each switch generates under normal conditions a default route and floods it in the "southern" direction. Obviously, such aggregation can blackhole in cases of misconfiguration or failures and this has to be addressed somehow.
For the visually oriented reader, Figure 1 presents a first simplified view of the resulting information and routes on a RIFT fabric. The top of the fabric is holding in its link-state database the nodes below it and routes to them. In the second row of the database we indicate that a partial information of other nodes in the same level is available as well; the details of how this is achieved should be postponed for the moment. Whereas when we look at the "bottom" of the fabric we see that the topology of the leafs is basically empty and they only hold a load balanced default route to the next level.
The balance of this document details the resulting protocol and fills in the missing details.
. [A,B,C,D]
. [E]
. +-----+ +-----+
. | E | | F | A/32 @ A
. +-+-+-+ +-+-+-+ B/32 @ B
. | | | | C/32 @ C
. | | +-----+ | D/32 @ D
. | | | |
. | +------+ |
. | | | |
. [A,B] +-+---+ | | +---+-+ [A,B]
. [D] | C +--+ +-+ D | [C]
. +-+-+-+ +-+-+-+
. 0/0 @ [E,F] | | | | 0/0 @ [E,F]
. A/32 @ A | | +-----+ | A/32 @ A
. B/32 @ B | | | | B/32 @ B
. | +------+ |
. | | | |
. +-+---+ | | +---+-+
. | A +--+ +-+ B |
. 0/0 @ [C,D] +-----+ +-----+ 0/0 @ [C,D]
Figure 1: RIFT information distribution
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.
This section presents the terminology used in this document. It is assumed that the reader is thoroughly familiar with the terms and concepts used in OSPF and IS-IS, [ISO10589] as well as the according graph theoretical concepts of shortest path first (SPF) computation and directed acyclic graphs (DAG).
. +--------+ +--------+
. | | | | ^ N
. |Spine 21| |Spine 22| |
.Level 2 ++-+--+-++ ++-+--+-++ <-*-> E/W
. | | | | | | | | |
. P111/2| |P121 | | | | S v
. ^ ^ ^ ^ | | | |
. | | | | | | | |
. +--------------+ | +-----------+ | | | +---------------+
. | | | | | | | |
. South +-----------------------------+ | | ^
. | | | | | | | All TIEs
. 0/0 0/0 0/0 +-----------------------------+ |
. v v v | | | | |
. | | +-+ +<-0/0----------+ | |
. | | | | | | | |
.+-+----++ optional +-+----++ ++----+-+ ++-----++
.| | E/W link | | | | | |
.|Node111+----------+Node112| |Node121| |Node122|
.+-+---+-+ ++----+-+ +-+---+-+ ++---+--+
. | | | South | | | |
. | +---0/0--->-----+ 0/0 | +----------------+ |
. 0/0 | | | | | | |
. | +---<-0/0-----+ | v | +--------------+ | |
. v | | | | | | |
.+-+---+-+ +--+--+-+ +-+---+-+ +---+-+-+
.| | (L2L) | | | | Level 0 | |
.|Leaf111~~~~~~~~~~~~Leaf112| |Leaf121| |Leaf122|
.+-+-----+ +-+---+-+ +--+--+-+ +-+-----+
. + + \ / + +
. Prefix111 Prefix112 \ / Prefix121 Prefix122
. multi-homed
. Prefix
.+---------- Pod 1 ---------+ +---------- Pod 2 ---------+
Figure 2: A two level spine-and-leaf topology
We will use this topology (called commonly a fat tree/network in modern DC considerations [VAHDAT08] as homonym to the original definition of the term) in all further considerations. It depicts a generic "fat-tree" and the concepts explained in three levels here carry by induction for further levels and higher degrees of connectivity. However, this document will deal with designs that provide only sparser connectivity as well.
[RFC7938] gives the original set of requirements augmented here based upon recent experience in the operation of fat-tree networks.
Following list represents possible requirements and requirements under discussion:
Finally, following are the non-requirements:
Derived from the above requirements we present a detailed outline of a protocol optimized for Routing in Fat Trees (RIFT) that in most abstract terms has many properties of a modified link-state protocol [RFC2328][RFC1142] when "pointing north" and path-vector [RFC4271] protocol when "pointing south". Albeit an unusual combination, it does quite naturally exhibit the desirable properties we seek.
The singular property of RIFT is that it floods northbound "flat" link-state information so that each level understands the full topology of levels south of it. In contrast, in the southbound direction the protocol operates like a path vector protocol or rather a distance vector with implicit split horizon since the topology constraints make a diffused computation front propagating in all directions unnecessary.
To account for the "northern" and the "southern" information split the link state database is partitioned into "north representation" and "south representation" TIEs, whereas in simplest terms the N-TIEs contain a link state topology description of lower levels and and S-TIEs carry simply default routes. This oversimplified view will be refined gradually in following sections while introducing protocol procedures aimed to fulfill the described requirements.
All protocol elements are carried over UDP. Once QUIC [QUIC] achieves the desired stability in deployments it may prove a valuable candidate for TIE transport.
All packet formats are defined in Thrift models in Section 8.
Future versions may include a [PROTOBUF] schema.
LIE exchange happens over well-known administratively locally scoped IPv4 multicast address [RFC2365] or link-local multicast scope for IPv6 [RFC4291] and SHOULD be sent with a TTL of 1 to prevent RIFT information reaching beyond a single L3 next-hop in the topology. LIEs are exchanged over all links running RIFT.
Unless Section 4.2.9 is used, each node is provisioned with the level at which it is operating and its PoD (or otherwise a default level and "undefined" "PoD are assumed; meaning that leafs do not need to be configured at all). Nodes in the spine are configured with an "undefined" PoD. This information is propagated in the LIEs exchanged.
A node tries to form a three way adjacency if and only if (definitions of LEAF_ONLY are found in Section 4.2.9)
].
Rule in Paragraph 3 MAY be optionally disregarded by a node if PoD detection is undesirable or has to be disregarded.
A node configured with "undefined" PoD membership MUST, after building first northbound adjacency making it participant in a PoD, advertise that PoD as part of its LIEs.
LIEs arriving with a TTL larger than 1 MUST be ignored.
A node SHOULD NOT send out LIEs without defined level in the header but in certain scenarios it may be beneficial for trouble-shooting purposes.
LIE exchange uses three-way handshake mechanism [RFC5303]. Precise finite state machines will be provided in later versions of this specification. LIE packets contain nonces and may contain an SHA-1 [RFC6234] over nonces and some of the LIE data which prevents corruption and replay attacks. TIE flooding reuses those nonces to prevent mismatches and can use those for security purposes in case it is using QUIC [QUIC]. Section 7 will address the precise security mechanisms in the future.
Topology and reachability information in RIFT is conveyed by the means of TIEs which have good amount of commonalities with LSAs in OSPF.
TIE exchange mechanism uses port indicated by each node in the LIE exchange and the interface on which the adjacency has been formed as destination. It SHOULD use TTL of 1 as well.
TIEs contain sequence numbers, lifetimes and a type. Each type has a large identifying number space and information is spread across possibly many TIEs of a certain type by the means of a hash function that a node or deployment can individually determine. One extreme point of the design space is a prefix per TIE which leads to BGP-like behavior vs. dense packing into few TIEs leading to more traditional IGP trade-off with fewer TIEs. An implementation may even rehash at the cost of significant amount of re-advertisements of TIEs.
More information about the TIE structure can be found in the schema in Section 8.
As a central concept to RIFT, each node represents itself differently depending on the direction in which it is advertising information. More precisely, a spine node represents two different databases to its neighbors depending whether it advertises TIEs to the north or to the south/sideways. We call those differing TIE databases either south- or northbound (S-TIEs and N-TIEs) depending on the direction of distribution.
The N-TIEs hold all of the node's adjacencies, local prefixes and northbound policy-guided prefixes while the S-TIEs hold only all of the node's adjacencies and the default prefix with necessary disaggregated prefixes and southbound policy-guided prefixes. We will explain this in detail further in Section 4.2.8 and Section 4.2.4.
The TIE types are symmetric in both directions and Table 1 provides a quick reference to the different TIE types including direction and their function.
| TIE-Type | Content |
|---|---|
| node N-TIE | node properties, adjacencies and information helping in complex disaggregation scenarios |
| node S-TIE | same content as node N-TIE except the information to help disaggregation |
| Prefix N-TIE | contains nodes' directly reachable prefixes |
| Prefix S-TIE | contains originated defaults and de-aggregated prefixes |
| PGP N-TIE | contains nodes north PGPs |
| PGP S-TIE | contains nodes south PGPs |
| KV N-TIE | contains nodes northbound KVs |
| KV S-TIE | contains nodes southbound KVs |
As an example illustrating a databases holding both representations, consider the topology in Figure 2 with the optional link between node 111 and node 112 (so that the flooding on an east-west link can be shown). This example assumes unnumbered interfaces. First, here are the TIEs generated by some nodes. For simplicity, the key value elements and the PGP elements which may be included in their S-TIEs or N-TIEs are not shown.
Spine21 S-TIEs:
Node S-TIE:
NodeElement(layer=2, neighbors((Node111, layer 1, cost 1),
(Node112, layer 1, cost 1), (Node121, layer 1, cost 1),
(Node122, layer 1, cost 1)))
Prefix S-TIE:
SouthPrefixesElement(prefixes(0/0, cost 1), (::/0, cost 1))
Node111 S-TIEs:
Node S-TIE:
NodeElement(layer=1, neighbors((Spine21, layer 2, cost 1, links(...)),
(Spine22, layer 2, cost 1, links(...)),
(Node112, layer 1, cost 1, links(...)),
(Leaf111, layer 0, cost 1, links(...)),
(Leaf112, layer 0, cost 1, links(...))))
Prefix S-TIE:
SouthPrefixesElement(prefixes(0/0, cost 1), (::/0, cost 1))
Node111 N-TIEs:
Node N-TIE:
NodeElement(layer=1,
neighbors((Spine21, layer 2, cost 1, links(...)),
(Spine22, layer 2, cost 1, links(...)),
(Node112, layer 1, cost 1, links(...)),
(Leaf111, layer 0, cost 1, links(...)),
(Leaf112, layer 0, cost 1, links(...))))
Prefix N-TIE:
NorthPrefixesElement(prefixes(Node111.loopback)
Node121 S-TIEs:
Node S-TIE:
NodeElement(layer=1, neighbors((Spine21,layer 2,cost 1),
(Spine22, layer 2, cost 1), (Leaf121, layer 0, cost 1),
(Leaf122, layer 0, cost 1)))
Prefix S-TIE:
SouthPrefixesElement(prefixes(0/0, cost 1), (::/0, cost 1))
Node121 N-TIEs:
Node N-TIE:
NodeLinkElement(layer=1,
neighbors((Spine21, layer 2, cost 1, links(...)),
(Spine22, layer 2, cost 1, links(...)),
(Leaf121, layer 0, cost 1, links(...)),
(Leaf122, layer 0, cost 1, links(...))))
Prefix N-TIE:
NorthPrefixesElement(prefixes(Node121.loopback)
Leaf112 N-TIEs:
Node N-TIE:
NodeLinkElement(layer=0,
neighbors((Node111, layer 1, cost 1, links(...)),
(Node112, layer 1, cost 1, links(...))))
Prefix N-TIE:
NorthPrefixesElement(prefixes(Leaf112.loopback, Prefix112,
Prefix_MH))
Figure 3: example TIES generated in a 2 level spine-and-leaf topology
The mechanism used to distribute TIEs is the well-known (albeit modified in several respects to address fat tree requirements) flooding mechanism used by today's link-state protocols. Albeit initially more demanding to implement it avoids many problems with diffused computation update style used by path vector. As described before, TIEs themselves are transported over UDP with the ports indicates in the LIE exchanges and using the destination address (for unnumbered IPv4 interfaces same considerations apply as in equivalent OSPF case) on which the LIE adjacency has been formed.
On reception of a TIE with an undefined level value in the packet header the node SHOULD issue a warning and indiscriminately discard the packet.
Precise finite state machines and procedures will be provided in later versions of this specification.
In a somewhat analogous fashion to link-local, area and domain flooding scopes, RIFT defines several complex "flooding scopes" depending on the direction and type of TIE propagated.
Every N-TIE is flooded northbound, providing a node at a given level with the complete topology of the Clos or Fat Tree network underneath it, including all specific prefixes. This means that a packet received from a node at the same or lower level whose destination is covered by one of those specific prefixes may be routed directly towards the node advertising that prefix rather than sending the packet to a node at a higher level.
A node's node S-TIEs, consisting of all node's adjacencies and prefix S-TIEs with default IP prefix and disaggregated prefixes, are flooded southbound in order to allow the nodes one level down to see connectivity of the higher level as well as reachability to the rest of the fabric. In order to allow a E-W disconnected node in a given level to receive the S-TIEs of other nodes at its level, every *NODE* S-TIE is "reflected" northbound to level from which it was received. It should be noted that east-west links are included in South TIE flooding; those TIEs need to be flooded to satisfy algorithms in Section 4.2.5. In that way nodes at same level can learn about each other without a lower level, e.g. in case of leaf level. The precise flooding scopes are given in Table 2. Those rules govern as well what SHOULD be included in TIDEs towards neighbors. East-West flooding scopes are identical to South flooding scopes.
Node S-TIE "reflection" allows to support disaggregation on failures describes in Section 4.2.8 and flooding reduction in Section 4.2.3.8.
| Packet Type vs. Peer Direction | South | North |
|---|---|---|
| node S-TIE | flood self-originated only | flood if TIE originator's level is higher than own level |
| non-node S-TIE | flood self-originated only | flood only if TIE originator is equal peer |
| all N-TIEs | never flood | flood always |
| TIDE | include TIEs in flooding scope | include TIEs in flooding scope |
| TIRE | include all N-TIEs and all peer's self-originated TIEs and all node S-TIEs | include only if TIE originator is equal peer |
As an example to illustrate these rules, consider using the topology in Figure 2, with the optional link between node 111 and node 112, and the associated TIEs given in Figure 3. The flooding from particular nodes of the TIEs is given in Table 3.
| Router floods to | Neighbor | TIEs |
|---|---|---|
| Leaf111 | Node112 | Leaf111 N-TIEs, Node111 node S-TIE |
| Leaf111 | Node111 | Leaf111 N-TIEs, Node112 node S-TIE |
| Node111 | Leaf111 | Node111 S-TIEs |
| Node111 | Leaf112 | Node111 S-TIEs |
| Node111 | Node112 | Node111 S-TIEs |
| Node111 | Spine21 | Node111 N-TIEs, Leaf111 N-TIEs, Leaf112 N-TIEs, Spine22 node S-TIE |
| Node111 | Spine22 | Node111 N-TIEs, Leaf111 N-TIEs, Leaf112 N-TIEs, Spine21 node S-TIE |
| ... | ... | ... |
| Spine21 | Node111 | Spine21 S-TIEs |
| Spine21 | Node112 | Spine21 S-TIEs |
| Spine21 | Node121 | Spine21 S-TIEs |
| Spine21 | Node122 | Spine21 S-TIEs |
| ... | ... | ... |
The initial exchange of RIFT is modeled after ISIS with TIDE being equivalent to CSNP and TIRE playing the role of PSNP. The content of TIDEs and TIREs is governed by Table 2.
RIFT does not purge information that has been distributed by the protocol. Purging mechanisms in other routing protocols have proven to be complex and fragile over many years of experience. Abundant amounts of memory are available today even on low-end platforms. The information will age out and all computations will deliver correct results if a node leaves the network due to the new information distributed by its adjacent nodes.
Once a RIFT node issues a TIE with an ID, it MUST preserve the ID as long as feasible (also when the protocol restarts), even if the TIE looses all content. The re-advertisement of empty TIE fulfills the purpose of purging any information advertised in previous versions. The originator is free to not re-originate the according empty TIE again or originate an empty TIE with relatively short lifetime to prevent large number of long-lived empty stubs polluting the network. Each node will timeout and clean up the according empty TIEs independently.
Upon restart a node MUST, as any link-state implementation, be prepared to receive TIEs with its own system ID and supercede them with equivalent, newly generated, empty TIEs with a higher sequence number. As above, the lifetime can be relatively short since it only needs to exceed the necessary propagation and processing delay by all the nodes that are within the TIE's flooding scope.
Under certain conditions nodes issue a default route in their South Prefix TIEs with metrics as computed in Section 4.3.5.1.
A node X that
originates in its south prefix TIE such a default route IIF
The term "all other nodes at X's' level" describes obviously just the nodes at the same level in the POD with a viable lower layer (otherwise the node S-TIEs cannot be reflected and the nodes in e.g. POD 1 nad POD 2 are "invisible" to each other).
A node originating a southbound default route MUST install a default discard route if it did not compute a default route during N-SPF.
Several nodes can, but strictly only under conditions defined below, run a hashing function based on TIE originator value and partition flooding between them.
Steps for flooding reduction and partitioning:
and then
Additional rules for flooding reduction and partitioning:
The mechanism can be implemented optionally in each node. The capability is carried in the node S-TIE (and for symmetry purposes in node N-TIE as well but it serves no purpose there currently).
Obviously flooding reduction does NOT apply to self originated TIEs. Observe further that all policy-guided information consists of self-originated TIEs.
In a fat tree, it can be sometimes desirable to guide traffic to particular destinations or keep specific flows to certain paths. In RIFT, this is done by using policy-guided prefixes with their associated communities. Each community is an abstract value whose meaning is determined by configuration. It is assumed that the fabric is under a single administrative control so that the meaning and intent of the communities is understood by all the nodes in the fabric. Any node can originate a policy-guided prefix.
Since RIFT uses distance vector concepts in a southbound direction, it is straightforward to add a policy-guided prefix to an S-TIE. For easier troubleshooting, the approach taken in RIFT is that a node's southbound policy-guided prefixes are sent in its S-TIE and the receiver does inbound filtering based on the associated communities (an egress policy is imaginable but would lead to different S-TIEs per neighbor possibly which is not considered in RIFT protocol procedures). A southbound policy-guided prefix can only use links in the south direction. If an PGP S-TIE is received on an east-west or northbound link, it must be discarded by ingress filtering.
Conceptually, a southbound policy-guided prefix guides traffic from the leaves up to at most the north-most layer. It is also necessary to to have northbound policy-guided prefixes to guide traffic from the north-most layer down to the appropriate leaves. Therefore, RIFT includes northbound policy-guided prefixes in its N PGP-TIE and the receiver does inbound filtering based on the associated communities. A northbound policy-guided prefix can only use links in the northern direction. If an N PGP TIE is received on an east-west or southbound link, it must be discarded by ingress filtering.
By separating southbound and northbound policy-guided prefixes and requiring that the cost associated with a PGP is strictly monotonically increasing at each hop, the path cannot loop. Because the costs are strictly increasing, it is not possible to have a loop between a northbound PGP and a southbound PGP. If east-west links were to be allowed, then looping could occur and issues such as counting to infinity would become an issue to be solved. If complete generality of path - such as including east-west links and using both north and south links in arbitrary sequence - then a Path Vector protocol or a similar solution must be considered.
If a node has received the same prefix, after ingress filtering, as a PGP in an S-TIE and in an N-TIE, then the node determines which policy-guided prefix to use based upon the advertised cost.
A policy-guided prefix is always preferred to a regular prefix, even if the policy-guided prefix has a larger cost. Section 8 provides normative indication of prefix preferences.
The set of policy-guided prefixes received in a TIE is subject to ingress filtering and then re-originated to be sent out in the receiver's appropriate TIE. Both the ingress filtering and the re-origination use the communities associated with the policy-guided prefixes to determine the correct behavior. The cost on re-advertisement MUST increase in a strictly monotonic fashion.
When a node X receives a PGP S-TIE or a PGP N-TIE that is originated from a node Y which does not have an adjacency with X, all PGPs in such a TIE MUST be filtered. Similarly, if node Y is at the same layer as node X, then X MUST filter out PGPs in such S- and N-TIEs to prevent loops.
Next, policy can be applied to determine which policy-guided prefixes to accept. Since ingress filtering is chosen rather than egress filtering and per-neighbor PGPs, policy that applies to links is done at the receiver. Because the RIFT adjacency is between nodes and there may be parallel links between the two nodes, the policy-guided prefix is considered to start with the next-hop set that has all links to the originating node Y.
A policy-guided prefix has or is assigned the following attributes:
The specific action to apply based upon a community is deployment specific. Here are some examples of things that can be done with communities. The length of a community is a 64 bits number and it can be written as a single field M or as a multi-field (S = M[0-31], T = M[32-63]) in these examples. For simplicity, the policy-guided prefix is referred to as P, the processing node as X and the originator as Y.
Once a policy-guided prefix has completed ingress filtering and policy, it is almost ready to store and use. It is still necessary to adjust the cost of the prefix to account for the link from the computing node X to the originating neighbor node Y.
There are three different policies that can be used:
The default policy is Minimum Unequal-Cost but well-known communities can be defined to get the other behaviors.
Regardless of the policy used, a node MUST store a PGP cost that is at least 1 greater than the PGP cost received. This enforces the strictly monotonically increasing condition that avoids loops.
Two databases of PGPs - from N-TIEs and from S-TIEs are stored. When a PGP is inserted into the appropriate database, the usual tie-breaking on cost is performed. Observe that the node retains all PGP TIEs due to normal flooding behavior and hence loss of the best prefix will lead to re-evaluation of TIEs present and re-advertisement of a new best PGP.
A node must re-originate policy-guided prefixes and retransmit them. The node has its database of southbound policy-guided prefixes to send in its S-TIE and its database of northbound policy-guided prefixes to send in its N-TIE.
Of course, a leaf does not need to re-originate southbound policy-guided prefixes.
PGPs may overlap with prefixes introduced by automatic de-aggregation. The topic is under further discussion. The break in connectivity that leads to infeasibility of a PGP is mirrored in adjacency tear-down and according removal of such PGPs. Nevertheless, the underlying link-state flooding will be likely reacting significantly faster than a hop-by-hop redistribution and with that the preference for PGPs may cause intermittent black-holes.
A node has three sources of relevant information. A node knows the full topology south from the received N-TIEs. A node has the set of prefixes with associated distances and bandwidths from received S-TIEs. A node can also have a set of PGPs.
To compute reachability, a node runs conceptually a northbound and a southbound SPF. We call that N-SPF and S-SPF.
Since neither computation can "loop" (with due considerations given to PGPs), it is possible to compute non-equal-cost or even k-shortest paths and "saturate" the fabric to the extent desired.
N-SPF uses northbound and east-west adjacencies in North Node TIEs when progressing Dijkstra. Observe that this is really just a one hop variety since South Node TIEs are not re-flooded southbound beyond a single level (or east-west) and with that the computation cannot progress beyond adjacent nodes.
Default route found when crossing an E-W link is used IIF
This rule forms a "one-hop default route split-horizon" and prevents looping over default routes while allowing for "one-hop protection" of nodes that lost all northbound adjacencies.
Other south prefixes found when crossing E-W link MAY be used IIF
i.e. the E-W link can be used as the gateway of last resort for a specific prefix only. Using south prefixes across E-W link can be beneficial e.g. on automatic de-aggregation in pathological fabric partitioning scenarios.
A detailed example can be found in Section 5.4.
For N-SPF we are using the South Node TIEs to find according adjacencies to verify backlink connectivity. Just as in case of IS-IS or OSPF, two unidirectional links are associated together to confirm bidirectional connectivity.
S-SPF uses only the southbound adjacencies in the south node TIEs, i.e. progresses towards nodes at lower levels. Observe that E-W adjacencies are NEVER used in the computation. This enforces the requirement that a packet traversing in a southbound direction must never change its direction.
S-SPF uses northbound adjacencies in north node TIEs to verify backlink connectivity.
Ultimately, it should be observed that in presence of a "ring" of E-W links in a level neither SPF will provide a "ring protection" scheme since such a computation would have to deal necessarily with breaking of "loops" in generic Dijkstra sense; an application for which RIFT is not intended. It is outside the scope of this document how an underlay can be used to provide a full-mesh connectivity between nodes in the same layer that would allow for N-SPF to provide protection for a single node loosing all its northbound adjacencies (as long as any of the other nodes in the level are northbound connected).
Using south prefixes over horizontal links is optional and can protect against pathological fabric partitioning cases that leave only paths to destinations that would necessitate multiple changes of forwarding direction between north and south.
After the SPF is run, it is necessary to attach according prefixes. For S-SPF, prefixes from an N-TIE are attached to the originating node with that node's next-hop set and a distance equal to the prefix's cost plus the node's minimized path distance. The RIFT route database, a set of (prefix, type=spf, path_distance, next-hop set), accumulates these results. Obviously, the prefix retains its type which is used to tie-break between the same prefix advertised with different types.
In case of N-SPF prefixes from each S-TIE need to also be added to the RIFT route database. The N-SPF is really just a stub so the computing node needs simply to determine, for each prefix in an S-TIE that originated from adjacent node, what next-hops to use to reach that node. Since there may be parallel links, the next-hops to use can be a set; presence of the computing node in the associated Node S-TIE is sufficient to verify that at least one link has bidirectional connectivity. The set of minimum cost next-hops from the computing node X to the originating adjacent node is determined.
Each prefix has its cost adjusted before being added into the RIFT route database. The cost of the prefix is set to the cost received plus the cost of the minimum cost next-hop to that neighbor. Then each prefix can be added into the RIFT route database with the next_hop_set; ties are broken based upon type first and then distance. RIFT route preferences are normalized by the according thrift model type.
for each S-TIE
if S-TIE.layer > X.layer
next_hop_set = set of minimum cost links to the S-TIE.originator
next_hop_cost = minimum cost link to S-TIE.originator
end if
for each prefix P in the S-TIE
P.cost = P.cost + next_hop_cost
if P not in route_database:
add (P, type=DistVector, P.cost, next_hop_set) to route_database
end if
if (P in route_database) and
(route_database[P].type is not PolicyGuided):
if route_database[P].cost > P.cost):
update route_database[P] with (P, DistVector, P.cost, next_hop_set)
else if route_database[P].cost == P.cost
update route_database[P] with (P, DistVector, P.cost,
merge(next_hop_set, route_database[P].next_hop_set))
else
// Not preferred route so ignore
end if
end if
end for
end for
Figure 4: Adding Routes from S-TIE Prefixes
An exemplary implementation for node X follows:
Each policy-guided prefix P has its cost and next_hop_set already stored in the associated database, as specified in Section 4.2.4.3; the cost stored for the PGP is already updated to considering the cost of the link to the advertising neighbor. By definition, a policy-guided prefix is preferred to a regular prefix.
for each policy-guided prefix P:
if P not in route_database:
add (P, type=PolicyGuided, P.cost, next_hop_set)
end if
if P in route_database :
if (route_database[P].type is not PolicyGuided) or
(route_database[P].cost > P.cost):
update route_database[P] with (P, PolicyGuided, P.cost, next_hop_set)
else if route_database[P].cost == P.cost
update route_database[P] with (P, PolicyGuided, P.cost,
merge(next_hop_set, route_database[P].next_hop_set))
else
// Not preferred route so ignore
end if
end if
end for
Figure 5: Adding Routes from Policy-Guided Prefixes
Under normal circumstances, node's S-TIEs contain just the adjacencies, a default route and policy-guided prefixes. However, if a node detects that its default IP prefix covers one or more prefixes that are reachable through it but not through one or more other nodes at the same level, then it MUST explicitly advertise those prefixes in an S-TIE. Otherwise, some percentage of the northbound traffic for those prefixes would be sent to nodes without according reachability, causing it to be black-holed. Even when not black-holing, the resulting forwarding could 'backhaul' packets through the higher level spines, clearly an undesirable condition affecting the blocking probabilities of the fabric.
We refer to the process of advertising additional prefixes as 'de-aggregation' or 'dis-aggregation'.
A node determines the set of prefixes needing de-aggregation using the following steps:
To summarize the above in simplest terms: if a node detects that its default route encompasses prefixes for which one of the other nodes in its level has no possible next-hops in the level below, it has to disaggregate it to prevent black-holing or suboptimal routing. Hence a node X needs to determine if it can reach a different set of south neighbors than other nodes at the same level, which are connected to it via at least one common south or east-west neighbor. If it can, then prefix disaggregation may be required. If it can't, then no prefix disaggregation is needed. An example of disaggregation is provided in Section 5.3.
A possible algorithm is described last:
A node X computes its SPF based upon the received N-TIEs. This results in a set of routes, each categorized by (prefix, path_distance, next-hop-set). Alternately, for clarity in the following procedure, these can be organized by next-hop-set as ( (next-hops), {(prefix, path_distance)}). If partial_neighbors isn't empty, then the following procedure describes how to identify prefixes to disaggregate.
disaggregated_prefixes = {empty }
nodes_same_layer = { empty }
for each S-TIE
if (S-TIE.layer == X.layer and
X shares at least one S-neighbor with X)
add S-TIE.originator to nodes_same_layer
end if
end for
for each next-hop-set NHS
isolated_nodes = nodes_same_layer
for each NH in NHS
if NH in partial_neighbors
isolated_nodes = intersection(isolated_nodes,
partial_neighbors[NH].nodes)
end if
end for
if isolated_nodes is not empty
for each prefix using NHS
add (prefix, distance) to disaggregated_prefixes
end for
end if
end for
copy disaggregated_prefixes to X's S-TIE
if X's S-TIE is different
schedule S-TIE for flooding
end if
Figure 6: Computation to Disaggregate Prefixes
Each disaggregated prefix is sent with the accurate path_distance. This allows a node to send the same S-TIE to each south neighbor. The south neighbor which is connected to that prefix will thus have a shorter path.
Finally, to summarize the less obvious points partially omitted in the algorithms to keep them more tractable:
Ultimately, complex partitions of superspine on sparsely connected fabrics can lead to necessity of transitive disaggregation through multiple layers. The topic will be described and standardized in later versions of this document.
Each RIFT node can optionally operate in zero touch provisioning (ZTP) mode, i.e. it has no configuration (unless it is a superspine at the top of the topology or it MUST operate as leaf and/or support leaf-2-leaf procedures) and it will fully configure itself after being attached to the topology. Configured nodes and nodes operating in ZTP can be mixed and will form a valid topology if achievable. This section describes the necessary concepts and procedures.
RIFT identifies each node via a SystemID which is a 64 bits wide integer. It is relatively simple to derive a, for all practical purposes collision free, value for each node on startup. As simple examples either system MAC and two random bytes can be used or an IPv4/IPv6 router ID interface address recycled as System ID. The router MUST ensure that such identifier is not changing very frequently (at least not without sending all its TIEs with fairly short lifetimes) since otherwise the network may be left with large amounts of stale TIEs in other nodes (though this is not necessarily a serious problem if the procedures suggested in Section 7 are implemented).
ZTP forces us to think about miscabled or unusually cabled fabric and how such a topology can be forced into a "lattice" structure which a fabric represents (with further restrictions). Let us consider a necessary and sufficient physical cabling in Figure 7. We assume all nodes being in the same PoD.
. +---+
. | A | s = SUPERSPINE_FLAG
. | S | l = LEAF_ONLY
. ++-++ l2l = LEAF_2_LEAF
. | |
. +--+ +--+
. | |
. +--++ ++--+
. | E | | F |
. | +-+ | +-----------+
. ++--+ | ++-++ |
. | | | | |
. | +-------+ | |
. | | | | |
. | | +----+ | |
. | | | | |
. ++-++ ++-++ |
. | I +-----+ J | |
. | | | +-+ |
. ++-++ +--++ | |
. | | | | |
. +---------+ | +------+ |
. | | | | |
. +-----------------+ | |
. | | | | |
. ++-++ ++-++ |
. | X +-----+ Y +-+
. |l2l| | l |
. +---+ +---+
Figure 7: Generic ZTP Cabling Considerations
First, we need to anchor the "top" of the cabling and that's what the SUPERSPINE_FLAG at node A is for. Then things look smooth until we have to decide whether node Y is at the same level as I, J or at the same level as Y and consequently, X is south of it. This is unresolvable here until we "nail down the bottom" of the topology. To achieve that we use the the leaf flags. We will see further then whether Y chooses to form adjacencies to F or I, J successively.
A node starting up with UNDEFINED_VALUE (i.e. without a CONFIGURED_LEVEL or any leaf or superspine flag) MUST follow those additional procedures:
A node starting with LEVEL_VALUE being 0 (i.e. it assumes a leaf function or has a CONFIGURED_LEVEL of 0) MUST follow those additional procedures:
Precise finite state machines will be provided in later versions of this specification.
The procedures defined in Section 4.2.9.4 will lead to the RIFT topology and levels depicted in Figure 8.
. +---+
. | As|
. | 64|
. ++-++
. | |
. +--+ +--+
. | |
. +--++ ++--+
. | E | | F |
. | 63+-+ | 63+-----------+
. ++--+ | ++-++ |
. | | | | |
. | +-------+ | |
. | | | | |
. | | +----+ | |
. | | | | |
. ++-++ ++-++ |
. | I +-----+ J | |
. | 62| | 62| |
. ++--+ +--++ |
. | | |
. +---------+ | |
. | | |
. ++-++ +---+ |
. | X | | Y +-+
. | 0 | | 0 |
. +---+ +---+
Figure 8: Generic ZTP Topology Autoconfigured
In case we imagine the LEAF_ONLY restriction on Y is removed the outcome would be very different however and result in Figure 9. This demonstrates basically that auto configuration prevents miscabling detection and with that can lead to undesirable effects when leafs are not "nailed" and arbitrarily cabled.
. +---+
. | As|
. | 64|
. ++-++
. | |
. +--+ +--+
. | |
. +--++ ++--+
. | E | | F |
. | 63+-+ | 63+-------+
. ++--+ | ++-++ |
. | | | | |
. | +-------+ | |
. | | | | |
. | | +----+ | |
. | | | | |
. ++-++ ++-++ +-+-+
. | I +-----+ J +-----+ Y |
. | 62| | 62| | 62|
. ++-++ +--++ ++-++
. | | | | |
. | +-----------------+ |
. | | |
. +---------+ | |
. | | |
. ++-++ |
. | X +--------+
. | 0 |
. +---+
Figure 9: Generic ZTP Topology Autoconfigured
The autoconfiguration mechanism computes a global maximum of levels by diffusion. The achieved equilibrium can be disturbed massively by all nodes with highest level either leaving or entering the domain (with some finer distinctions not explained further). It is therefore recommended that each node is multi-homed towards nodes with respective HAL offerings. Fortuntately, this is the natural state of things for the topology variants considered in RIFT.
Overload Bit MUST be respected in all according reachability computations. A node with overload bit set SHOULD NOT advertise any reachability prefixes southbound except locally hosted ones.
The leaf node SHOULD set the 'overload' bit on its node TIEs, since if the spine nodes were to forward traffic not meant for the local node, the leaf node does not have the topology information to prevent a routing/forwarding loop.
Since the leafs do see only "one hop away" they do not need to run a full SPF but can simply gather prefix candidates from their neighbors and build the according routing table.
A leaf will have no N-TIEs except its own and optionally from its east-west neighbors. A leaf will have S-TIEs from its neighbors.
Instead of creating a network graph from its N-TIEs and neighbor's S-TIEs and then running an SPF, a leaf node can simply compute the minimum cost and next_hop_set to each leaf neighbor by examining its local interfaces, determining bi-directionality from the associated N-TIE, and specifying the neighbor's next_hop_set set and cost from the minimum cost local interfaces to that neighbor.
Then a leaf attaches prefixes as in Section 4.2.6 as well as the policy-guided prefixes as in Section 4.2.7.
The protocol supports a southbound distribution of key-value pairs that can be used to e.g. distribute configuration information during topology bring-up. The KV S-TIEs can arrive from multiple nodes and hence need tie-breaking per key. We use the following rules
Observe that if a node goes down, the node south of it looses adjacencies to it and with that the KVs will be disregarded and on tie-break changes new KV re-advertised to prevent stale information being used by nodes further south. KV information in southbound direction is not result of independent computation of every node but a diffused computation.
Certain use cases seem to necessitate distribution of essentialy KV information that is generated in the leafs in the northbound direction. Such information is flooded in KV N-TIEs. Since the originator of northbound KV is preserved during northbound flooding, overlapping keys could be used. However, to omit further protocol complexity, only the value of the key in TIE tie-broken in same fashion as southbound KV TIEs is used.
RIFT MAY incorporate BFD to react quickly to link failures. In such case following procedures are introduced:
A well understood problem in fabrics is that in case of link losses it would be ideal to rebalance how much traffic is offered to switches in the next layer based on the ingress and egress bandwidth they have. Current attempts rely mostly on specialized traffic engineering via controller or leafs being aware of complete topology with according cost and complexity.
RIFT presents a very light weight mechanism that can deal with the problem in an approximative way based on the fact that RIFT is loop-free.
In a first step, a node can compare the amount of northbound bandwith available to neighbors at the same level and modify metric on its advertised default route (or even other routes) to present a different distance leading to e.g. e.g. weighted ECMP forwarding on leafs. We call such a distance Bandwidth Adjusted Distance or BAD. This is best illustrated by a simple example.
. | x | |
. | x | |
. +-+---+-+ +-+---+-+
. | | | |
. |Node111| |Node112|
. +-+---+++ ++----+++
. |x || || ||
. || |+---------------+ ||
. || +---------------+| ||
. || || || ||
. || +------------+| || ||
. || |+------------+ || ||
. |x || || ||
. +-+---+++ +--++-+++
. | | | |
. |Leaf111| |Leaf112|
. +-------+ +-------+
Figure 10: Balancing Bandwidth
All links in Figure 10 are assumed to have the same bandwith for simplicity. Node 111 sees in the node S-TIE of 112 that Node 112 has twice the amount of bandwidth going northbound and therefore Node 111 will advertise its default route cost (BAD) as twice the default which without further failures would lead to Leaf 111 and Leaf 112 distributing 2/3 of the traffic to Node 111 and 1/3 to Node 112.
Further, in Figure 10 we assume that Leaf111 lost one of the parallel links to Node 111 and with that wants to push more traffic onto Node 112. This leads to local modification of the received BADs and each node can choose the ratio here independently based on understanding of e.g. traffic distribution between E-W and N-S or queue occupancy. If we assume that 50% of the leaf's traffic is for Leaf112 and 50% exits northbound we would modify the BADs accordingly to the bandwidth available towards each of them and end in Leaf 111 with a weight of 4 to Node 111 and weight of 1 to Node 112 which gives us roughly 4/5 of the traffic going to Node 112.
Future version of this document will provide the precise algorithm to compute BADs from all other nodes at the same level using the same algorithm as Section 4.2.3.8 while ignoring overloaded nodes.
Observe that since BAD is only computed for default routes any disaggregated prefixes or PGP are not affected.
Observe further that a change in available bandwidth will only affect one level down in the fabric, i.e. blast radius of bandwidth changes is contained.
Due to its loop free properties a node could take during S-SPF into account the available bandwidth on the nodes in lower layers and modify the amount of traffic offered to next level's "southbound" nodes based as what it sees is the total achievable maximum flow through those nodes. It is worth observing that such computations will work better if standardized but does not have to be necessarily. As long the packet keeps on heading south it will take one of the available paths and arrive at the intended destination.
Future versions of this document will fill in more details.
Recently, alternative architecture to reuse labels as segment identifiers [I-D.ietf-spring-segment-routing] has gained traction and may present use cases in DC fabric that would justify its deployment. Such use cases will either precondition an assignment of a label per node (or other entities where the mechanisms are equivalent) or a global assignment and a knowledge of topology everywhere to compute segment stacks of interest. We deal with the two issues separately.
Global segment identifiers are normally assumed to be provided by some kind of a centralized "controller" instance and distributed to other entities. This can be performed in RIFT by attaching a controller to the superspine nodes at the top of the fabric where the whole topology is always visible, assign such identifiers and then distribute those via the KV mechanism towards all nodes so they can perform things like probing the fabric for failures using a stack of segments.
Some segment routing use cases seem to precondition full knowledge of fabric topology in all nodes which can be performed albeit at the loss of one of highly desirable properties of RIFT, namely minimal blast radius. Basically, RIFT can function as a flat IGP by switching off its flooding scopes. All nodes will end up with full topology view and albeit the N-SPF and S-SPF are still performed based on RIFT rules, any computation with segment identifiers that needs full topology can use it.
Beside blast radius problem, excessive flooding may present significant load on implementations. RIFT can be extended beside the mechanism in Section 4.2.3.8 to provide an algorithm for globally optimized flooding minimalization should demand for such a use case solidify.
RIFT can optionally allow special leaf East-West adjacencies under additional set of rules. The leaf supporting those procedures MUST:
This will allow the E-W leaf nodes to exchange traffic strictly for the prefixes advertised in each other's north prefix TIEs (since the southbound computation will find the reverse direction in the other node's TIE and install its north prefixes).
Losing full, flat topology information at every node will have an impact on some of the end-to-end network services. This is the price paid for minimal disturbance in case of failures and reduced flooding and memory requirements on nodes lower south in the level hierarchy.
Multi-Topology (MT)[RFC5120] and Multi-Instance (MI)[RFC6822] is used today in link-state routing protocols to support several domains on the same physical topology. RIFT supports this capability by carrying transport ports in the LIE protocol exchanges. Multiplexing of LIEs can be achieved by either choosing varying multicast addresses or ports on the same address.
BFD interactions in Section 4.3.4 are implementation dependent when multiple RIFT instances run on the same link.
RIFT does not precondition that its nodes have reachable addresses albeit for operational purposes this is clearly desirable. Under normal operating conditions this can be easily achieved by e.g. injecting the node's loopback address into North Prefix TIEs.
Things get more interesting in case a node looses all its northbound adjacencies but is not at the top of the fabric. In such a case a node that detects that some other members at its level are advertising northbound adjacencies MAY inject its loopback address into southbound PGP TIE and become reachable "from the south" that way. Further, a solution may be implemented where based on e.g. a "well known" community such a southbound PGP is reflected at level 0 and advertised as northbound PGP again to allow for "reachability from the north" at the cost of additional flooding.
Based on the rules defined in Section 4.2.5, Section 4.2.3.7 and given presence of E-W links, RIFT can provide a one-hop protection of nodes that lost all their northbound links or in other complex link set failure scenarios. Section 5.4 explains the resulting behavior based on one such example.
This section describes RIFT deployment in the example topology without any node or link failures. We disregard flooding reduction for simplicity's sake.
As first step, the following bi-directional adjacencies will be created (and any other links that do not fulfill LIE rules in Section 4.2.2 disregarded):
Consequently, N-TIEs would be originated by Node 111 and Node 112 and each set would be sent to both Spine 21 and Spine 22. N-TIEs also would be originated by Leaf 111 (w/ Prefix 111) and Leaf 112 (w/ Prefix 112 and the multi-homed prefix) and each set would be sent to Node 111 and Node 112. Node 111 and Node 112 would then flood these N-TIEs to Spine 21 and Spine 22.
Similarly, N-TIEs would be originated by Node 121 and Node 122 and each set would be sent to both Spine 21 and Spine 22. N-TIEs also would be originated by Leaf 121 (w/ Prefix 121 and the multi-homed prefix) and Leaf 122 (w/ Prefix 122) and each set would be sent to Node 121 and Node 122. Node 121 and Node 122 would then flood these N-TIEs to Spine 21 and Spine 22.
At this point both Spine 21 and Spine 22, as well as any controller to which they are connected, would have the complete network topology. At the same time, Node 111/112/121/122 hold only the N-ties of level 0 of their respective PoD. Leafs hold only their own N-TIEs.
S-TIEs with adjacencies and a default IP prefix would then be originated by Spine 21 and Spine 22 and each would be flooded to Node 111, Node 112, Node 121, and Node 122. Node 111, Node 112, Node 121, and Node 122 would each send the S-TIE from Spine 21 to Spine 22 and the S-TIE from Spine 22 to Spine 21. (S-TIEs are reflected up to level from which they are received but they are NOT propagated southbound.)
An S Tie with a default IP prefix would be originated by Node 111 and Node 112 and each would be sent to Leaf 111 and Leaf 112. Leaf 111 and Leaf 112 would each send the S-TIE from Node 111 to Node 112 and the S-TIE from Node 112 to Node 111.
Similarly, an S Tie with a default IP prefix would be originated by Node 121 and Node 122 and each would be sent to Leaf 121 and Leaf 122. Leaf 121 and Leaf 122 would each send the S-TIE from Node 121 to Node 122 and the S-TIE from Node 122 to Node 121. At this point IP connectivity with maximum possible ECMP has been established between the leafs while constraining the amount of information held by each node to the minimum necessary for normal operation and dealing with failures.
. | | | |
.+-+---+-+ +-+---+-+
.| | | |
.|Node111| |Node112|
.+-+---+-+ ++----+-+
. | | | |
. | +---------------+ X
. | | | X Failure
. | +-------------+ | X
. | | | |
.+-+---+-+ +--+--+-+
.| | | |
.|Leaf111| |Leaf112|
.+-------+ +-------+
. + +
. Prefix111 Prefix112
Figure 11: Single Leaf link failure
In case of a failing leaf link between node 112 and leaf 112 the link-state information will cause re-computation of the necessary SPF and the higher levels will stop forwarding towards prefix 112 through node 112. Only nodes 111 and 112, as well as both spines will see control traffic. Leaf 111 will receive a new S-TIE from node 112 and reflect back to node 111. Node 111 will de-aggregate prefix 111 and prefix 112 but we will not describe it further here since de-aggregation is emphasized in the next example. It is worth observing however in this example that if leaf 111 would keep on forwarding traffic towards prefix 112 using the advertised south-bound default of node 112 the traffic would end up on spine 21 and spine 22 and cross back into pod 1 using node 111. This is arguably not as bad as black-holing present in the next example but clearly undesirable. Fortunately, de-aggregation prevents this type of behavior except for a transitory period of time.
. +--------+ +--------+ S-TIE of Spine21
. | | | | received by
. |Spine 21| |Spine 22| reflection of
. ++-+--+-++ ++-+--+-++ Nodes 112 and 111
. | | | | | | | |
. | | | | | | | 0/0
. | | | | | | | |
. | | | | | | | |
. +--------------+ | +--- XXXXXX + | | | +---------------+
. | | | | | | | |
. | +-----------------------------+ | | |
. 0/0 | | | | | | |
. | 0/0 0/0 +- XXXXXXXXXXXXXXXXXXXXXXXXX -+ |
. | 1.1/16 | | | | | |
. | | +-+ +-0/0-----------+ | |
. | | | 1.1./16 | | | |
.+-+----++ +-+-----+ ++-----0/0 ++----0/0
.| | | | | 1.1/16 | 1.1/16
.|Node111| |Node112| |Node121| |Node122|
.+-+---+-+ ++----+-+ +-+---+-+ ++---+--+
. | | | | | | | |
. | +---------------+ | | +----------------+ |
. | | | | | | | |
. | +-------------+ | | | +--------------+ | |
. | | | | | | | |
.+-+---+-+ +--+--+-+ +-+---+-+ +---+-+-+
.| | | | | | | |
.|Leaf111| |Leaf112| |Leaf121| |Leaf122|
.+-+-----+ ++------+ +-----+-+ +-+-----+
. + + + +
. Prefix111 Prefix112 Prefix121 Prefix122
. 1.1/16
Figure 12: Fabric partition
Figure 12 shows the arguably most catastrophic but also the most interesting case. Spine 21 is completely severed from access to Prefix 121 (we use in the figure 1.1/16 as example) by double link failure. However unlikely, if left unresolved, forwarding from leaf 111 and leaf 112 to prefix 121 would suffer 50% black-holing based on pure default route advertisements by spine 21 and spine 22.
The mechanism used to resolve this scenario is hinging on the distribution of southbound representation by spine 21 that is reflected by node 111 and node 112 to spine 22. Spine 22, having computed reachability to all prefixes in the network, advertises with the default route the ones that are reachable only via lower level neighbors that spine 21 does not show an adjacency to. That results in node 111 and node 112 obtaining a longest-prefix match to prefix 121 which leads through spine 22 and prevents black-holing through spine 21 still advertising the 0/0 aggregate only.
The prefix 121 advertised by spine 22 does not have to be propagated further towards leafs since they do no benefit from this information. Hence the amount of flooding is restricted to spine 21 reissuing its S-TIEs and reflection of those by node 111 and node 112. The resulting SPF in spine 22 issues a new prefix S-TIEs containing 1.1/16. None of the leafs become aware of the changes and the failure is constrained strictly to the level that became partitioned.
To finish with an example of the resulting sets computed using notation introduced in Section 4.2.8, spine 22 constructs the following sets:
With that and |H (for r=prefix 121) and |H (for r=prefix 122) being disjoint from |A (for spine 21), spine 22 will originate an S-TIE with prefix 121 and prefix 122, that is flooded to nodes 112, 112, 121 and 122.
. + + +
. X N1 | N2 | N3
. X | |
.+--+----+ +--+----+ +--+-----+
.| |0/0> <0/0| |0/0> <0/0| |
.| A01 +----------+ A02 +----------+ A03 | Level 1
.++-+-+--+ ++--+--++ +---+-+-++
. | | | | | | | | |
. | | +----------------------------------+ | | |
. | | | | | | | | |
. | +-------------+ | | | +--------------+ |
. | | | | | | | | |
. | +----------------+ | +-----------------+ |
. | | | | | | | | |
. | | +------------------------------------+ | |
. | | | | | | | | |
.++-+-+--+ | +---+---+ | +-+---+-++
.| | +-+ +-+ | |
.| L01 | | L02 | | L03 | Level 0
.+-------+ +-------+ +--------+
Figure 13: North Partitioned Router
Figure 13 shows a part of a fabric where level 1 is horizontally connected and A01 lost its only northbound adjacency. Based on N-SPF rules in Section 4.2.5.1 A01 will compute northbound reachability by using the link A01 to A02 (whereas A02 will NOT use this link during N-SPF). Hence A01 will still advertise the default towards level 0 and route unidirectionally using the horizontal link. Moreover, based on Section 4.3.10 it may advertise its loopback address as south PGP to remain reachable "from the south" for operational purposes. This is necessary since A02 will NOT route towards A01 using the E-W link (doing otherwise may form routing loops).
As further consideration, the moment A02 looses link N2 the situation evolves again. A01 will have no more northbound reachability while still seeing A03 advertising northbound adjacencies in its south node tie. With that it will stop advertising a default route due to Section 4.2.3.7. Moreover, A02 may now inject its loopback address as south PGP.
Ideally RIFT can be stretched out to the lowest level in the IP fabric to integrate ToRs or even servers. Since those entities would run as leafs only, it is worth to observe that a leaf only version is significantly simpler to implement and requires much less resources:
. +-----+ +-----+
. | | | |
.+-+ S0 | | S1 |
.| ++---++ ++---++
.| | | | |
.| | +------------+ |
.| | | +------------+ |
.| | | | |
.| ++-+--+ +--+-++
.| | | | |
.| | A0 | | A1 |
.| +-+--++ ++---++
.| | | | |
.| | +------------+ |
.| | +-----------+ | |
.| | | | |
.| +-+-+-+ +--+-++
.+-+ | | |
. | L0 | | L1 |
. +-----+ +-----+
Figure 14: Level Shortcut
Strictly speaking, RIFT is not limited to Clos variations only. The protocol preconditions only a sense of 'compass rose direction' achieved by configuration (or derivation) of levels and other topologies are possible within this framework. So, conceptually, one could include leaf to leaf links and even shortcut between layers but certain requirements in Section 3 will not be met anymore. As an example, shortcutting levels illustrated in Figure 14 will lead either to suboptimal routing when L0 sends traffic to L1 (since using S0's default route will lead to the traffic being sent back to A0 or A1) or the leafs need each other's routes installed to understand that only A0 and A1 should be used to talk to each other.
Whether such modifications of topology constraints make sense is dependent on many technology variables and the exhausting treatment of the topic is definitely outside the scope of this document.
Obviously, an implementation may choose to originate southbound instead of a strict default route (as described in Section 4.2.3.7) a shorter prefix P' but in such a scenario all addresses carried within the RIFT domain must be contained within P'.
The protocol has provisions for nonces and can include authentication mechanisms in the future comparable to [RFC5709] and [RFC7987].
One can consider additionally attack vectors where a router may reboot many times while changing its system ID and pollute the network with many stale TIEs or TIEs are sent with very long lifetimes and not cleaned up when the routes vanishes. Those attack vectors are not unique to RIFT. Given large memory footprints available today those attacks should be relatively benign. Otherwise a node can implement a strategy of e.g. discarding contents of all TIEs of nodes that were not present in the SPF tree over a certain period of time. Since the protocol, like all modern link-state protocols, is self-stabilizing and will advertise the presence of such TIEs to its neighbors, they can be re-requested again if a computation finds that it sees an adjacency formed towards the system ID of the discarded TIEs.
Section 4.2.9 presents many attack vectors in untrusted environments, starting with nodes that oscillate their level offers to the possiblity of a node offering a three way adjacency with the highest possible level value with a very long holdtime trying to put itself "on top of the lattice" and with that gaining access to the whole southbound topology. Session authentication mechanisms are necessary in environments where this is possible.
This section introduces the schema for information elements.
On schema changes that
major version of the schema MUST increase. All other changes MUST increase minor version within the same major.
Thrift serializer/deserializer MUST not discard optional, unknown fields but preserve and serialize them again when re-flooding whereas missing optional fields MAY be replaced with according default values if present.
All signed integer as forced by Thrift support must be cast for internal purposes to equivalent unsigned values without discarding the signedness bit. An implementation SHOULD try to avoid using the signedness bit when generating values.
The schema is normative.
/**
Thrift file with common definitions for RIFT
*/
namespace * common
/** @note MUST be interpreted in implementation as unsigned 64 bits.
* The implementation SHOULD NOT use the MSB.
*/
typedef i64 SystemIDType
typedef i32 IPv4Address
/** this has to be of length long enough to accomodate prefix */
typedef binary IPv6Address
/** @note MUST be interpreted in implementation as unsigned 16 bits */
typedef i16 UDPPortType
/** @note MUST be interpreted in implementation as unsigned 32 bits */
typedef i32 TIENrType
/** @note MUST be interpreted in implementation as unsigned 32 bits */
typedef i32 MTUSizeType
/** @note MUST be interpreted in implementation as unsigned 32 bits */
typedef i32 SeqNrType
/** @note MUST be interpreted in implementation as unsigned 32 bits */
typedef i32 LifeTimeInSecType
/** @note MUST be interpreted in implementation as unsigned 16 bits */
typedef i16 LevelType
/** @note MUST be interpreted in implementation as unsigned 32 bits */
typedef i32 PodType
/** @note MUST be interpreted in implementation as unsigned 16 bits */
typedef i16 VersionType
/** @note MUST be interpreted in implementation as unsigned 32 bits */
typedef i32 MetricType
/** @note MUST be interpreted in implementation as unsigned 32 bits */
typedef i32 BandwithInMegaBitsType
typedef string KeyIDType
/** node local, unique identification for a link (interface/tunnel
* etc. Basically anything RIFT runs on). This is kept
* at 32 bits so it aligns with BFD [RFC5880] discriminator size.
*/
typedef i32 LinkIDType
typedef string KeyNameType
typedef i8 PrefixLenType
/** timestamp in seconds since the epoch */
typedef i64 TimestampInSecsType
/** security nonce */
typedef i64 NonceType
/** adjacency holdtime */
typedef i16 HoldTimeInSecType
/** Flags indicating nodes behavior in case of ZTP and support
for special optimization procedures. It will force level to `leaf_level`
*/
enum LeafIndications {
leaf_only =0,
leaf_only_and_leaf_2_leaf_procedures =1,
}
/** default bandwidth on a link */
const BandwithInMegaBitsType default_bandwidth = 10
/** fixed leaf level when ZTP is not used */
const LevelType leaf_level = 0
const LevelType default_level = leaf_level
/** This MUST be used when node is configured as superspine in ZTP.
This is kept reasonably low to alow for fast ZTP convergence on
failures. */
const LevelType default_superspine_level = 24
const PodType default_pod = 0
const LinkIDType undefined_linkid = 0
/** default distance used */
const MetricType default_distance = 1
/** any distance larger than this will be considered infinity */
const MetricType infinite_distance = 0x70000000
/** any element with 0 distance will be ignored,
* missing metrics will be replaced with default_distance
*/
const MetricType invalid_distance = 0
const bool overload_default = false
const bool flood_reduction_default = true
const HoldTimeInSecType default_holdtime = 3
/** by default LIE levels are ZTP offers */
const bool default_not_a_ztp_offer = false
/** 0 is illegal for SystemID */
const SystemIDType IllegalSystemID = 0
/** empty set of nodes */
const set<SystemIDType> empty_set_of_nodeids = {}
/** normalized bandwidth metric maximum, i.e. node with lowest northbound bandwidth
* at its level uses this metric to advertise its default route */
const MetricType normalized_bw_metric_max = 0x1fff
/** normalized bandwidth metric minimum, i.e. node with highest northbound bandwidth
* at its level uses this metric to advertise its default route */
const MetricType normalized_bw_metric_min = 0x00ff
/** default UDP port to run LIEs on */
const UDPPortType default_lie_udp_port = 6949
const UDPPortType default_tie_udp_flood_port = 6950
/** default MTU size to use */
const MTUSizeType default_mtu_size = 1400
/** default mcast is v4 224.0.1.150, we make it i64 to
* help languages struggling with highest bit */
const i64 default_lie_v4_mcast_group = 3758096790
/** indicates whether the direction is northbound/east-west
* or southbound */
enum TieDirectionType {
Illegal = 0,
South = 1,
North = 2,
DirectionMaxValue = 3,
}
enum AddressFamilyType {
Illegal = 0,
AddressFamilyMinValue = 1,
IPv4 = 2,
IPv6 = 3,
AddressFamilyMaxValue = 4,
}
struct IPv4PrefixType {
1: required IPv4Address address;
2: required PrefixLenType prefixlen;
}
struct IPv6PrefixType {
1: required IPv6Address address;
2: required PrefixLenType prefixlen;
}
union IPAddressType {
1: optional IPv4Address ipv4address;
2: optional IPv6Address ipv6address;
}
union IPPrefixType {
1: optional IPv4PrefixType ipv4prefix;
2: optional IPv6PrefixType ipv6prefix;
}
enum TIETypeType {
Illegal = 0,
TIETypeMinValue = 1,
/** first legal value */
NodeTIEType = 2,
PrefixTIEType = 3,
TransitivePrefixTIEType = 4,
PGPrefixTIEType = 5,
KeyValueTIEType = 6,
TIETypeMaxValue = 7,
}
/** @note: route types which MUST be ordered on their preference
* PGP prefixes are most preferred attracting
* traffic north (towards spine) and then south
* normal prefixes are attracting traffic south (towards leafs),
* i.e. prefix in NORTH PREFIX TIE is preferred over SOUTH PREFIX TIE
*/
enum RouteType {
Illegal = 0,
RouteTypeMinValue = 1,
/** First legal value. */
/** Discard routes are most prefered */
Discard = 2,
/** Local prefixes are directly attached prefixes on the
* system such as e.g. interface routes.
*/
LocalPrefix = 3,
/** advertised in S-TIEs */
SouthPGPPrefix = 4,
/** advertised in N-TIEs */
NorthPGPPrefix = 5,
/** advertised in N-TIEs */
NorthPrefix = 6,
/** advertised in S-TIEs */
SouthPrefix = 7,
/** transitive southbound are least preferred */
TransitiveSouthPrefix = 8,
RouteTypeMaxValue = 9
}
/**
Thrift file for packet encodings for RIFT
*/
include "common.thrift"
/** represents protocol encoding schema major version */
const i32 protocol_major_version = 8
/** represents protocol encoding schema minor version */
const i32 protocol_minor_version = 0
/** common RIFT packet header */
struct PacketHeader {
1: required common.VersionType major_version = protocol_major_version;
2: required common.VersionType minor_version = protocol_minor_version;
/** this is the node sending the packet, in case of LIE/TIRE/TIDE
also the originator of it */
3: required common.SystemIDType sender;
/** level of the node sending the packet, required on everything except
* LIEs. Lack of presence on LIEs indicates UNDEFINED_LEVEL and is used
* in ZTP procedures.
*/
4: optional common.LevelType level;
}
/** Community serves as community for PGP purposes */
struct Community {
1: required i32 top;
2: required i32 bottom;
}
/** Neighbor structure */
struct Neighbor {
1: required common.SystemIDType originator;
2: required common.LinkIDType remote_id;
}
/** Capabilities the node supports */
struct NodeCapabilities {
/** can this node participate in flood reduction,
only relevant at level > 0 */
1: optional bool flood_reduction =
common.flood_reduction_default;
/** does this node restrict itself to be leaf only (in ZTP) and
does it support leaf-2-leaf procedures */
2: optional common.LeafIndications leaf_indications;
}
/** RIFT LIE packet
@note this node's level is already included on the packet header */
struct LIEPacket {
/** optional node or adjacency name */
1: optional string name;
/** local link ID */
2: required common.LinkIDType local_id;
/** UDP port to which we can receive flooded TIEs */
3: required common.UDPPortType flood_port =
common.default_tie_udp_flood_port;
/** layer 3 MTU */
4: optional common.MTUSizeType link_mtu_size =
common.default_mtu_size;
/** this will reflect the neighbor once received to provid
3-way connectivity */
5: optional Neighbor neighbor;
6: optional common.PodType pod = common.default_pod;
/** optional nonce used for security computations */
7: optional common.NonceType nonce;
/** optional node capabilities shown in the LIE. The capabilies
MUST match the capabilities shown in the Node TIEs, otherwise
the behavior is unspecified. A node detecting the mismatch
SHOULD generate according error.
*/
8: optional NodeCapabilities capabilities;
/** required holdtime of the adjacency, i.e. how much time
MUST expire without LIE for the adjacency to drop
*/
9: required common.HoldTimeInSecType holdtime =
common.default_holdtime;
/** indicates that the level on the LIE MUST NOT be used
to derive a ZTP level by the receiving node. */
10: optional bool not_a_ztp_offer =
common.default_not_a_ztp_offer;
}
/** LinkID pair describes one of parallel links between two nodes */
struct LinkIDPair {
/** node-wide unique value for the local link */
1: required common.LinkIDType local_id;
/** received remote link ID for this link */
2: required common.LinkIDType remote_id;
/** more properties of the link can go in here */
}
/** ID of a TIE
@note: TIEID space is a total order achieved by comparing the elements
in sequence defined and comparing each value as an
unsigned integer of according length
*/
struct TIEID {
/** indicates direction of the TIE */
1: required common.TieDirectionType direction;
/** indicates originator of the TIE */
2: required common.SystemIDType originator;
3: required common.TIETypeType tietype;
4: required common.TIENrType tie_nr;
}
/** Header of a TIE */
struct TIEHeader {
2: required TIEID tieid;
3: required common.SeqNrType seq_nr;
/** lifetime expires down to 0 just like in ISIS */
4: required common.LifeTimeInSecType lifetime;
}
/** A sorted TIDE packet, if unsorted, behavior is undefined */
struct TIDEPacket {
/** all 00s marks starts */
1: required TIEID start_range;
/** all FFs mark end */
2: required TIEID end_range;
/** _sorted_ list of headers */
3: required list<TIEHeader> headers;
}
/** A TIRE packet */
struct TIREPacket {
1: required set<TIEHeader> headers;
}
/** Neighbor of a node */
struct NodeNeighborsTIEElement {
2: required common.LevelType level;
/** Cost to neighbor.
@note: All parallel links to same node
incur same cost, in case the neighbor has multiple
parallel links at different cost, the largest distance
(highest numerical value) MUST be advertised
@note: any neighbor with cost <= 0 MUST be ignored in computations */
3: optional common.MetricType cost = common.default_distance;
/** can carry description of multiple parallel links in a TIE */
4: optional set<LinkIDPair> link_ids;
/** total bandwith to neighbor, this will be normally sum of the
* bandwidths of all the parallel links.
**/
5: optional common.BandwithInMegaBitsType bandwidth =
common.default_bandwidth;
}
/** Flags the node sets */
struct NodeFlags {
/** node is in overload, do not transit traffic through it */
1: optional bool overload = common.overload_default;
}
/** Description of a node.
It may occur multiple times in different TIEs but if either
* capabilities values do not match or
* flags values do not match or
* neighbors repeat with different values or
* visible in same level/having partition upper do not match
the behavior is undefined and a warning SHOULD be generated.
Neighbors can be distributed across multiple TIEs however if
the sets are disjoint.
@note: observe that absence of fields implies defined defaults
*/
struct NodeTIEElement {
1: required common.LevelType level;
/** if neighbor systemID repeats in other node TIEs of same node
the behavior is undefined. Equivalent to |A_(n,s)(N) in spec. */
2: required map<common.SystemIDType,
NodeNeighborsTIEElement> neighbors;
3: optional NodeCapabilities capabilities;
4: optional NodeFlags flags;
/** optional node name for easier operations */
5: optional string name;
/** Nodes seen an the same level through reflection through nodes
having backlink to both nodes. They are equivalent to |V(N) in
future specifications. Ignored in Node S-TIEs if present.
*/
6: optional set<common.SystemIDType> visible_in_same_level
= common.empty_set_of_nodeids;
/** Non-overloaded nodes in |V seen as attached to another north
* level partition due to the fact that some nodes in its |V have
* adjacencies to higher level nodes that this node doesn't see.
* This may be used in the computation at higher levels to prevent
* blackholing. Ignored in Node S-TIEs if present.
* Equivalent to |PUL(N) in spec. */
7: optional set<common.SystemIDType> same_level_unknown_north_partitions
= common.empty_set_of_nodeids;
}
struct PrefixAttributes {
/** Observe that in default metric case the node is supposed to advertise
* metric calculated from comparison of bandwidths at all nodes at its
* level. **/
2: required common.MetricType metric = common.default_distance;
}
/** multiple prefixes */
struct PrefixTIEElement {
/** prefixes with the associated attributes.
if the same prefix repeats in multiple TIEs of same node
behavior is unspecified */
1: required map<common.IPPrefixType, PrefixAttributes> prefixes;
}
/** keys with their values */
struct KeyValueTIEElement {
/** if the same key repeats in multiple TIEs of same node
or with different values, behavior is unspecified */
1: required map<common.KeyIDType,string> keyvalues;
}
/** single element in a TIE. enum common.TIETypeType
in TIEID indicates which elements MUST be present
in the TIEElement. In case of mismatch the unexpected
elements MUST be ignored.
*/
union TIEElement {
/** in case of enum common.TIETypeType.NodeTIEType */
1: optional NodeTIEElement node;
/** in case of enum common.TIETypeType.PrefixTIEType */
2: optional PrefixTIEElement prefixes;
/** transitive prefixes (always southbound) which SHOULD be propagated
* southwards towards lower levels to heal
* pathological upper level partitioning, otherwise
* blackholes may occur. MUST NOT be advertised within a North TIE.
*/
3: optional PrefixTIEElement transitive_prefixes;
4: optional KeyValueTIEElement keyvalues;
/** @todo: policy guided prefixes */
}
/** @todo: flood header separately in UDP to allow caching to TIEs
while changing lifetime?
*/
struct TIEPacket {
1: required TIEHeader header;
2: required TIEElement element;
}
union PacketContent {
1: optional LIEPacket lie;
2: optional TIDEPacket tide;
3: optional TIREPacket tire;
4: optional TIEPacket tie;
}
/** protocol packet structure */
struct ProtocolPacket {
1: required PacketHeader header;
2: required PacketContent content;
}
This specification will request at an opportune time multiple registry points to exchange protocol packets in a standardized way, amongst them multicast address assignments and standard port numbers. The schema itself defines many values and codepoints which can be considered registries themselves.
Many thanks to Naiming Shen for some of the early discussions around the topic of using IGPs for routing in topologies related to Clos. Russ White to be especially acknowledged for the key conversation on epistomology that allowed to tie current asynchronous distributed systems theory results to a modern protocol design presented here. Adrian Farrel, Joel Halpern and Jeffrey Zhang provided thoughtful comments that improved the readability of the document and found good amount of corners where the light failed to shine. Kris Price was first to mention single router, single arm default considerations. Jeff Tantsura helped out with some initial thoughts on BFD interactions while Jeff Haas corrected several misconceptions about BFD's finer points. Artur Makutunowicz pointed out many possible improvements and acted as sounding board in regard to modern protocol implementation techniques RIFT is exploring. Barak Gafni formalized first time clearly the problem of partitioned spine on a (clean) napkin in Singapore.