Internet-Draft Abbreviated Title July 2022
Zhang, et al. Expires 4 January 2023 [Page]
Workgroup:
Internet Engineering Task Force
Internet-Draft:
draft-zwx-rift-leaf-ring-01
Published:
Intended Status:
Standards Track
Expires:
Authors:
Z. Zhang
ZTE Corporation
Y. Wei
ZTE Corporation
B. Xu
ZTE Corporation

Supporting leaves without northbound neighbors connecting to a fat-tree network using RIFT

Abstract

This document discusses the usage and solution for leaf nodes without northbound neighbors connecting to a fat-tree network by leaf nodes having direct northbound neighbors in RIFT.

Status of This Memo

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 4 January 2023.

Table of Contents

1. Introduction

[I-D.ietf-rift-rift] specifies a dynamic routing protocol for Clos and fat-tree network topology. It suits most of the deployments. In some situations, the leaves are connected by ring or chain topology, and some of them may have no northbound link, so these nodes may not be able to connecting the network. This document discusses the usage and proposes a solution.

1.1. Requirements Language

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119].

2. Problem Statement

[I-D.ietf-rift-rift] specifies a dynamic routing protocol for Clos and fat-tree network topology. The leaf part of a traditional Clos and fat-tree network topology is shown as:

      +                  +                  +
      | N1               | N2               | N3
      |                  |                  |
   +--+----+          +--+----+          +--+-----+
   |       |          |       |          |        |
   |  S01  +----------+  S02  +----------+  S03   | Level 1
   ++-+-+--+          ++--+--++          +---+-+-++
    | | |              |  |  |               | | |
    | | +----------------------------------+ | | |
    | |                |  |  |             | | | |
    | +-------------+  |  |  |  +--------------+ |
    |               |  |  |  |  |          | |   |
    | +----------------+  |  +-----------------+ |
    | |             |     |     |          | | | |
    | | +------------------------------------+ | |
    | | |           |     |     |          |   | |
   ++-+-+--+        | +---+---+ |        +-+---+-++
   |       |        +-+       +-+        |        |
   |  L01  |----------|  L02  |----------|  L03   | Level 0
   +-------+          +-------+          +--------+
Figure 1

In most of the cases, each leaf node has north connections with at least one spine, and there may be east-west links between leaf nodes. In case a leaf node lost all the north connections, it can still access the network through the east-west link between leaves. For example in Figure 1, there is an east-west link between L01 and L02, and there is an east-west link between L02 and L03. When the northbound connections for the leaves are all work well, L01, L02 and L03 may generate a Prefix South TIE with default route and advertise it through the east-west links according to the definition in section 4.2.3.4 in [I-D.ietf-rift-rift]. In case L01 lost all the northbound links with S01, S02 and S03, according to Northbound SPF algorithm defined in section 4.2.4.1 in [I-D.ietf-rift-rift], L01 can compute the next hop L02 for the default route. On the other hand, the prefix of L01 can be flood northbound by L02. Then L01 can still access the network through the east-west link between L01 and L02.

But in some deployments, the leaves may connect with each other by ring topology (sometimes, a chain topology), and not all of them have northbound connection with spine nodes.

For example, in the IP Radio Access Network (IP RAN, mobile backhaul network), the 4G eNB or the 5G gNB connect to an IP access network of a ring topology. The access network attaches to an aggregation network through two aggregation nodes. In 5G era, the aggregation network and the metro network is envolving to Spine-and-Leaf architecture to take the advantage of Spine-and-Leaf. Figure 2 depicts an digram of an IPRAN network with Spine-and-Leaf architecture. If the aggregation network runs RIFT, using the proposal of this draft, the access network ring does not need to deploy other IGP protocol to enable the routing.

                 +----+
                 |gNB |
                 +--+-+
                    |
                 +--+-+
           +-----+CSG4+---- --+
           |     +----+       |    Aggregation
+---+   +--+-+              +-+--+   Network +-----+     xxxxx
|eNB+---+CSG1|              |    +-----------+Spine+--xxxx   xxxx
+---+   +--+-+              |Leaf|           +----++            xxx
           |                |    |                |               xx
           |                +----+------------+   |                x
           | Access Network                   |   | Metro Network  x
           |                +----+------------+---+                x
           |                |    |            |                   xx
+---+   +--+-+              |Leaf|           ++----+            xxx
|gNB+---+CSG2|              |    +-----------+Spine+--xxxx   xxxx
+---+   +--+-+              +-+--+           +-----+     xxxxx
           |     +----+       |
           +-----+CSG3+---- --+
                 +--+-+
                    |
                 +--+-+
                 |eBN |
                 +----+
Figure 2: IPRAN network with Spine-and-Leaf architecture
        +                  +                  +
        | N1               | N2               | N3
        |                  |                  |
     +--+----+          +--+----+          +--+-----+
     |       |          |       |          |        |
     |  S01  +----------+  S02  +----------+  S03   | Level 1
     ++-+-+--+          ++--+--++          +---+-+-++
      | | |              |  |  |               | | |
      | | +----------------------------------+ | | |
      | |                |  |  |             | | | |
      | +-------------+  |  |  |  +--------------+ |
      |               |  |  |  |  |          | |   |
      | +----------------+  |  +-----------------+ |
      | |             |     |     |          | | | |
      | | +------------------------------------+ | |
      | | |           |     |     |          |   | |
     ++-+-+--+        | +---+---+ |        +-+---+-++
     |       |        +-+       +-+        |        |
  +--|  L01  |--------- |  L02  |----------|  L03   |--+
  |  +-------+          +-------+          +--------+  |
  |                                                    |
  |                                                    | Level 0
  |  ++-+-+--+         ++-+-+--+           ++-+-+--+   |
  |  |       |         |       |           |       |   |
  +--|  L04  |---------|  L05  |-----------|  L06  |---+
     +-------+         +-------+           +-------+
Figure 3

Figure 3 is an abstract of Figure 2, several leaves connect to the fat-tree network by ring topology, each leaf has two east-west connections with other leaves, but only L01, L02 and L03 have northbound connection with spine nodes. L01, L02 and L03 advertise a Prefix South TIE with default route through the east-west links. L04 and L06 can access the network through L01 and L03. But L04 and L06 do not generate or flood the Prefix South TIE with default route because they have no northbound link. So L05 cannot receive the Prefix South TIE with default route through the link between L04 and L05, or the link between L05 and L06. On the other hand, the prefix of L05 cannot be flooded east-west by L04 and L05. So L05 cannot send flow to other leaf, and other leaf cannot send flow to L05 also. L05 cannot access the network.

This document discuss the extension that can be used to solve the problem when some leaves without northbound neighbors connecting to a fat-tree network.

3. Solution

3.1. Capability advertisement

A new link capability which is named leaf-transport is set in LIE and node TIE. The new capability indicates that the leaf node can transfer the Prefix TIE received from the east-west link to the other leaf node.

The LIE FSM will not be affected if only one leaf supports the capability and the neighbor does not support.

The capability can be used for diagnosis in case there is something wrong when computing route.

3.2. Prefix transferring

When a leaf node which has no northbound connection receives Prefix TIE from a neighbor by an east-west link, it can transfer the Prefix TIE to the other neighbor by the other east-west link, with increased metric. The increased metric should be the sum of the received metric and the metric of the east-west link.

The Prefix TIE can be the default route and the prefix of the leaf node, other prefixes can also be transferred. The network administrator can control the prefix transferring by policy.

Prefix South TIE transferring
The leaf node without northbound neighbors which supports the leaf transfer capability, MUST transfer the Prefix South TIE received from an east-west neighbor to the other east-west neighbor which also has no northbound connection.
Prefix North TIE transferring
The leaf node without northbound neighbors which supports the leaf transfer capability, MUST transfer the Prefix North TIE received from an east-west neighbor which has no northbound connection to the other east-west neighbor.
External Prefix North TIE transferring
The leaf node without northbound neighbors which supports the leaf transfer capability, MUST transfer the External Prefix North TIE received from an east-west neighbor which has no northbound connection to the other east-west neighbor.

The leaf node without northbound neighbors which supports the leaf transfer capability, MAY transfer the Prefix TIE from an east-west neighbor to the other east-west neighbor which has northbound connection. According to Northbound SPF algorithm defined in section 4.2.4.1 in [I-D.ietf-rift-rift], the transferring does not affect the routing calculating result of the neighbors which has northbound connection.

3.3. Example

In figure 2, either L04 or L06, or both of them advertise the leaf-transport capability in LIE to L05 when they are forming an adjacency. And the leaf-transport capability is also set in node TIE when L04 or L06 advertise the TIE.

L04/L06 receives Prefix South TIE with default route from L01/L03, L04/L06 transfers the route through Prefix South TIE with increased metric to L05.

L04/L06 receives Prefix North TIE of L05 and transfers the route through Prefix North TIE with increased metric to L01/L03.

L05 receives the Prefix South TIE from L04/L06, after N-SPF calculation, L05 calculates the default route with next hop L04/L06. L01/L03 receives prefix of L05 from L04/L06, and L01/03 floods the Prefix North TIE northbound. Then L05 can send flow to other leaf through L04/L06. The flow sent by other leaf with destination set to the prefix of L05 can also be routed to L05. Then L05 can access the network.

4. IANA Considerations

A new type for 'leaf-transfer' is requested in link capability.

5. Security Considerations

When the node without northbound neighbors supports the function defined in this document, there may be unnecessary Prefix TIE advertisement, and unnecessary prefix may be leaked.

6. References

6.1. Normative References

[I-D.ietf-rift-rift]
Sharma, A., Thubert, P., Rijsman, B., Afanasiev, D., and A. Przygienda, "RIFT: Routing in Fat Trees", Work in Progress, Internet-Draft, draft-ietf-rift-rift-15, , <https://datatracker.ietf.org/doc/html/draft-ietf-rift-rift-15>.
[RFC2119]
Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, , <https://www.rfc-editor.org/info/rfc2119>.

Authors' Addresses

Zheng Zhang
ZTE Corporation
China
Yuehua Wei
ZTE Corporation
China
Benchong Xu
ZTE Corporation
China