RFC 5640 PROPOSED STANDARD BGP · TUNNELING · LOAD BALANCING

Softwire mesh LB: Load-Balancing for Mesh Softwires

A small BGP extension that lets core routers spread tunneled traffic over equal-cost paths.

On rfc-editor.org ↗
PUBLISHED
August 2009
AUTHORS
C. Filsfils, P. Mohapatra, C. Pignataro
STREAM
IETF
UPDATED BY
RFC 9012 — BGP Tunnel Encap
SUB-TLV TYPE
5
SUB-TLV LENGTH
2 octets
VALUE
Block length in bits
TUNNEL TYPES COVERED
1 (L2TPv3 over IP) and 2 (GRE)
EXAMPLE BLOCK SIZE
24 (Session ID 0x1234ABCD)
DOI
10.17487/RFC5640

RFC text

Contents
Original text · RFC Editor View on rfc-editor.org ↗
Network Working Group                                        C. Filsfils
Request for Comments: 5640                                  P. Mohapatra
Category: Standards Track                                   C. Pignataro
                                                           Cisco Systems
                                                             August 2009


                   Load-Balancing for Mesh Softwires

Abstract

   Payloads transported over a Softwire mesh service (as defined by BGP
   Encapsulation Subsequent Address Family Identifier (SAFI) information
   exchange) often carry a number of identifiable, distinct flows.  It
   can, in some circumstances, be desirable to distribute these flows
   over the equal cost multiple paths (ECMPs) that exist in the packet
   switched network.  Currently, the payload of a packet entering the
   Softwire can only be interpreted by the ingress and egress routers.
   Thus, the load-balancing decision of a core router is only based on
   the encapsulating header, presenting much less entropy than available
   in the payload or the encapsulated header since the Softwire
   encapsulation acts in a tunneling fashion.  This document describes a
   method for achieving comparable load-balancing efficiency in a
   network carrying Softwire mesh service over Layer Two Tunneling
   Protocol - Version 3 (L2TPv3) over IP or Generic Routing
   Encapsulation (GRE) encapsulation to what would be achieved without
   such encapsulation.

Status of This Memo

   This document specifies an Internet standards track protocol for the
   Internet community, and requests discussion and suggestions for
   improvements.  Please refer to the current edition of the "Internet
   Official Protocol Standards" (STD 1) for the standardization state
   and status of this protocol.  Distribution of this memo is unlimited.

Copyright Notice

   Copyright (c) 2009 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 in effect on the date of
   publication of this document (http://trustee.ietf.org/license-info).
   Please review these documents carefully, as they describe your rights
   and restrictions with respect to this document.

Table of Contents

   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . . . 2
     1.1.  Requirements Language . . . . . . . . . . . . . . . . . . . 2
   2.  Load-Balancing Block sub-TLV  . . . . . . . . . . . . . . . . . 2
     2.1.  Applicability to Tunnel Types . . . . . . . . . . . . . . . 3
     2.2.  Encapsulation Considerations  . . . . . . . . . . . . . . . 4
   3.  IANA Considerations . . . . . . . . . . . . . . . . . . . . . . 4
   4.  Security Considerations . . . . . . . . . . . . . . . . . . . . 4
   5.  Acknowledgements  . . . . . . . . . . . . . . . . . . . . . . . 4
   6.  Normative References  . . . . . . . . . . . . . . . . . . . . . 5

1.  Introduction

   Consider the case of a router R1 that encapsulates a packet P into a
   Softwire bound to router R3.  R2 is a router on the shortest path
   from R1 to R3.  R2's shortest path to R3 involves equal cost multiple
   paths (ECMPs).  The goal is for R2 to be able to choose which path to
   use on the basis of the full entropy of packet P.

   This is achieved by carrying in the encapsulation header a signature
   of the inner header, hence enhancing the entropy of the flows as seen
   by the core routers.  The signature is carried as part of one of the
   fields of the encapsulation header.  To aid with better description
   in the document, we define the generic term "load-balancing field" to
   mean such a value that is specific to an encapsulation type.  For
   example, for L2TPv3-over-IP [RFC3931] encapsulation, the load-
   balancing field is the Session Identifier (Session ID).  For GRE
   [RFC2784] encapsulation, the Key field [RFC2890], if present,
   represents the load-balancing field.  This mechanism assumes that
   core routers base their load-balancing decisions on a flow definition
   that includes the load-balancing field.  This is an obvious and
   generic functionality as, for example, for L2TPv3-over-IP tunnels,
   the Session ID is at the same well-known constant offset as the TCP/
   UDP ports in the encapsulating header.

   The Encapsulation SAFI [RFC5512] is extended such that a contiguous
   block of the load-balancing field is bound to the Softwire advertised
   by a BGP next-hop.  On a per-inner-flow basis, the ingress Provider
   Edge (PE) selects one value of the load-balancing field from the
   block to preserve per-flow ordering and, at the same time, to enhance
   the entropy across flows.

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.  Load-Balancing Block sub-TLV

   This document defines a new sub-TLV for use with the Tunnel
   Encapsulation Attribute defined in [RFC5512].  The new sub-TLV is
   referred to as the "Load-Balancing Block sub-TLV" and MAY be included
   in any Encapsulation SAFI UPDATE message where load-balancing is
   desired.

   The sub-TLV type of the Load-Balancing Block sub-TLV is 5.  The sub-
   TLV length is 2 octets.  The value represents the length of the block
   in bits and MUST NOT exceed the size of the load-balancing field.
   This format is very similar to the variable-length subnet masking
   (VLSM) used in IP addresses to allow arbitrary length prefixes.  The
   block is determined by extracting the initial sequence of 'block
   size' bits from the load-balancing field.

   If a load-balancing field is not signaled (e.g., if the encapsulation
   sub-TLV is not included in an advertisement as in the case of GRE
   without a Key), then the Load-Balancing Block sub-TLV MUST NOT be
   included.

   The smaller the value field of the Load-Balancing Block sub-TLV, the
   larger the space for per-flow identification, and hence the better
   entropy for potential load-balancing in the core, as well as, the
   lower the polarization when mapping flows to ECMP paths.  However,
   reducing the load-balancing block size consumes more L2TPv3 Session
   IDs or GRE Keys, resulting in potentially less numbers of supported
   services.  A typical deployment would need to arbitrate between this
   trade-off.

   As an example, assume that there is a Softwire set up between R1 and
   R3 with L2TPv3-over-IP tunnel type.  Assume that R3 encodes the
   Session ID with value 0x1234ABCD in the encapsulation sub-TLV.  It
   also includes the Load-Balancing Block sub-TLV and encodes the value
   24.  This should be interpreted as follows:

   o  If an ingress router does not understand the Load-Balancing Block
      sub-TLV, it continues to use the Session ID 0x1234ABCD and
      encapsulates all packets with that Session ID.

   o  If an ingress router understands the Load-Balancing Block sub-TLV,
      it picks the first 24 bits out of the Session ID (0x1234AB) to be
      used as the block and fills in the lower-order 8 bits with a per-
      flow identifier (e.g., it can be determined based on the inner
      packet's source, destination addresses, and TCP/UDP ports).  This
      selection preserves the per-flow ordering of packets.

   This requirement and solution applies equally to GRE where the Key
   plays the same role as the Session ID in L2TPv3.

   Needless to say, if an egress router does not support the Load-
   Balancing Block sub-TLV, the Softwire continues to operate with a
   single load-balancing field with which all ingress routers
   encapsulate.

2.1.  Applicability to Tunnel Types

   The Load-Balancing Block sub-TLV is applicable to tunnel types that
   define a load-balancing field.  This document defines load-balancing
   fields for tunnel types 1 (L2TPv3 over IP) and 2 (GRE) as follows:

   o  L2TPv3 over IP - Session ID.  Special care needs to be taken to
      always create a non-zero Session ID.  When an egress router
      includes a Load-Balancing Block sub-TLV, it MUST encode the
      Session ID field of the encapsulation sub-TLV in a way that
      ensures that the most significant bits of the Session ID, after
      extracting the block, are non-zero.

   o  GRE - GRE Key

   This document does not define a load-balancing field for the IP-in-IP
   tunnel type (tunnel types 7).  Future tunnel types that desire to use
   the Load-Balancing Block sub-TLV MUST define a load-balancing field
   that is part of the encapsulating header.

2.2.  Encapsulation Considerations

   Fields included in the encapsulation header besides the load-
   balancing field are not affected by the Load-Balancing Block sub-TLV.
   All other encapsulation fields are shared between variations of the
   load-balancing field.  For example, for the L2TPv3-over-IP tunnel
   type, if the optional cookie is included in the encapsulation sub-TLV
   by the egress router during Softwire signaling, it applies to all the
   "Session ID" values derived at the ingress router after applying the
   load-balancing block as described in this document.

3.  IANA Considerations

   IANA has assigned the value 5 for the Load-Balancing Block sub-TLV,
   in the BGP Tunnel Encapsulation Attribute Sub-TLVs registry (number
   space created as part of the publication of [RFC5512]):

       Sub-TLV name                            Value
       -------------                           -----
       Load-Balancing Block                      5

4.  Security Considerations

   This document defines a new sub-TLV for the BGP Tunnel Encapsulation
   Attribute.  Security considerations for the BGP Encapsulation SAFI
   and the BGP Tunnel Encapsulation Attribute are covered in [RFC5512].
   There are no additional security risks introduced by this design.

5.  Acknowledgements

   The authors would like to thank Stewart Bryant, Mark Townsley, Rajiv
   Asati, Kireeti Kompella, and Robert Raszuk for their review and
   comments.

6.  Normative References

   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119, March 1997.

   [RFC2784]  Farinacci, D., Li, T., Hanks, S., Meyer, D., and P.
              Traina, "Generic Routing Encapsulation (GRE)", RFC 2784,
              March 2000.

   [RFC2890]  Dommety, G., "Key and Sequence Number Extensions to GRE",
              RFC 2890, September 2000.

   [RFC3931]  Lau, J., Townsley, M., and I. Goyret, "Layer Two Tunneling
              Protocol - Version 3 (L2TPv3)", RFC 3931, March 2005.

   [RFC5512]  Mohapatra, P. and E. Rosen, "The BGP Encapsulation
              Subsequent Address Family Identifier (SAFI) and the BGP
              Tunnel Encapsulation Attribute", RFC 5512, April 2009.

Authors' Addresses

   Clarence Filsfils
   Cisco Systems
   Brussels,
   Belgium

   EMail: cfilsfil@cisco.com


   Pradosh Mohapatra
   Cisco Systems
   170 W. Tasman Drive
   San Jose, CA  95134
   USA

   EMail: pmohapat@cisco.com


   Carlos Pignataro
   Cisco Systems
   7200 Kit Creek Road, PO Box 14987
   Research Triangle Park, NC  27709
   USA

   EMail: cpignata@cisco.com

AI ELI5

On this page
PROBLEMPackets inside a softwire tunnel look identical to core routers, because only the outer header is visible. Many separate flows therefore get pinned to a single equal-cost path.
MECHANISMA new Load-Balancing Block sub-TLV tells ingress routers how many leading bits of the tunnel's load-balancing field (L2TPv3 Session ID or GRE Key) are fixed. The ingress fills the remaining bits with a per-flow value, so core routers see more variety.
WHY IT MATTERSSoftwire mesh traffic can be balanced across equal-cost paths about as well as it would be without encapsulation. The RFC is a Proposed Standard and is updated by RFC 9012.

§1The Problem: Hidden Flows in a Tunnel

A Softwire mesh service carries payloads between routers using tunnels. The RFC's abstract points out that these payloads often carry a number of identifiable, distinct flows, and that it can be desirable to spread those flows over the equal cost multiple paths (ECMPs) in the network.

The catch is that only the ingress and egress routers can interpret the payload. A core router choosing among equal-cost paths looks only at the encapsulating header, which offers much less entropy than the payload or the inner header would.

The RFC's setup is a router R1 that encapsulates a packet P into a softwire bound for R3, with R2 on the shortest path between them. R2 has equal-cost paths to R3, and the goal is for R2 to choose among them using the full entropy of packet P.

Everything in the tunnel wears one outer label, so the road-sorting routers cannot tell the cars apart and send them all the same way.

§2The Load-Balancing Field

The idea is to carry a signature of the inner header in the encapsulation header, so core routers see more entropy. The signature rides in one field of the encapsulation header, which the RFC calls the load-balancing field.

The field depends on the tunnel type. For L2TPv3 over IP (RFC 3931) it is the Session Identifier (Session ID). For GRE (RFC 2784) it is the Key field from RFC 2890, if present.

The mechanism assumes core routers include the load-balancing field in their flow definition. The RFC calls this an obvious and generic capability: for L2TPv3-over-IP, the Session ID sits at the same well-known constant offset as the TCP/UDP ports in the encapsulating header.

The Encapsulation SAFI from RFC 5512 is extended so that a contiguous block of the load-balancing field is bound to the softwire advertised by a BGP next-hop. For each inner flow, the ingress Provider Edge (PE) selects one value from the block. That keeps packets of a flow in order and still adds entropy across flows.

The tunnel already has a number on its outside. Instead of using one number for everything, the sender picks from a small set of numbers, using a different one for each conversation.

§3The Load-Balancing Block Sub-TLV

The RFC defines a new sub-TLV for the Tunnel Encapsulation Attribute of RFC 5512. It is called the Load-Balancing Block sub-TLV, and it MAY be included in any Encapsulation SAFI UPDATE message where load-balancing is desired.

Its sub-TLV type is 5 and its length is 2 octets. The value is the length of the block in bits, and it MUST NOT exceed the size of the load-balancing field. The RFC compares this to variable-length subnet masking: the block is the initial sequence of that many bits taken from the load-balancing field.

If no load-balancing field is signaled, for example GRE without a Key, the sub-TLV MUST NOT be included.

There is a trade-off. A smaller value leaves more room for per-flow identification, which gives better entropy and less polarization when mapping flows to ECMP paths. But it consumes more Session IDs or GRE Keys, which can mean fewer supported services. A typical deployment has to arbitrate between the two.

The number says how much of the label is fixed and how much the sender may change. Changing more of it spreads cars better, but uses up more labels.

§4Worked Example: Session ID 0x1234ABCD

Suppose a softwire is set up between R1 and R3 using the L2TPv3-over-IP tunnel type. R3 advertises Session ID 0x1234ABCD in the encapsulation sub-TLV and also includes the Load-Balancing Block sub-TLV with the value 24.

  • An ingress router that does not understand the sub-TLV keeps using Session ID 0x1234ABCD for all packets.
  • An ingress router that does understand it takes the first 24 bits (0x1234AB) as the block and fills the lower-order 8 bits with a per-flow identifier. The RFC suggests deriving it from the inner packet's source and destination addresses and TCP/UDP ports. This preserves per-flow ordering.

The same approach applies to GRE, where the Key plays the role of the Session ID. If an egress router does not support the sub-TLV, the softwire keeps working with a single load-balancing field that all ingress routers use.

If the label is 1234ABCD, the sender keeps the first part fixed and swaps the last bit of it for a different value for each conversation. Old equipment simply ignores all this and uses the plain label.

§5Tunnel Types and Encapsulation

The sub-TLV applies to tunnel types that define a load-balancing field. The RFC defines one for tunnel type 1 (L2TPv3 over IP), where it is the Session ID, and for tunnel type 2 (GRE), where it is the GRE Key. It does not define one for the IP-in-IP tunnel type (tunnel type 7). Future tunnel types that want to use the sub-TLV MUST define a load-balancing field that is part of the encapsulating header.

For L2TPv3 there is an extra rule. An egress router that includes the sub-TLV MUST encode the Session ID so that its most significant bits, after extracting the block, are non-zero, so a non-zero Session ID is always created.

Other encapsulation fields are not affected. If an optional L2TPv3 cookie is in the encapsulation sub-TLV, it applies to all the Session ID values the ingress derives from the block.

Only tunnel types that have a suitable label field can play this game, and everything else about the tunnel stays as it was.

LOAD-BALANCING FIELD BY TUNNEL TYPE
Tunnel typeEncapsulationLoad-balancing field
1 THIS RFCL2TPv3 over IPSession ID
2 THIS RFCGREGRE Key
7IP-in-IPNone defined

§6IANA Registration

IANA assigned the value 5 to the Load-Balancing Block sub-TLV in the BGP Tunnel Encapsulation Attribute Sub-TLVs registry, a number space created with the publication of RFC 5512.

The official list of sub-label types got a new entry, number 5, for this feature.

§7Security Considerations

The RFC adds only a new sub-TLV to the BGP Tunnel Encapsulation Attribute. Security considerations for the BGP Encapsulation SAFI and that attribute are covered in RFC 5512, and the RFC states that there are no additional security risks introduced by this design.

This is a small add-on to an existing system, so the safety rules of that system already cover it.

§8References