Network Working Group Yiqun. Cai
Internet-Draft Heidi. Ou
Intended status: Standards Track Alibaba Group
Expires: December 21, 2018 Sri. Vallepalli
Mankamana. Mishra
Stig. Venaas
Cisco Systems
Andy. Green
British Telecom
June 19, 2018

PIM Designated Router Load Balancing
draft-ietf-pim-drlb-08

Abstract

On a multi-access network, one of the PIM routers is elected as a Designated Router (DR). On the last hop LAN, the PIM DR is responsible for tracking local multicast listeners and forwarding traffic to these listeners if the group is operating in PIM-SM. In this document, we propose a modification to the PIM-SM protocol that allows more than one of these last hop routers to be selected so that the forwarding load can be distributed among these routers.

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 December 21, 2018.

Copyright Notice

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.


Table of Contents

1. Introduction

On a multi-access LAN such as an Ethernet, one of the PIM routers is elected as a DR. The PIM DR has two roles in the PIM-SM protocol. On the first hop network, the PIM DR is responsible for registering an active source with the Rendezvous Point (RP) if the group is operating in PIM-SM. On the last hop LAN, the PIM DR is responsible for tracking local multicast listeners and forwarding to these listeners if the group is operating in PIM-SM.

Consider the following last hop LAN in Figure 1:

            
                         ( core networks )
                           |     |     |
                           |     |     |
                          R1    R2     R3
                           |     |     |
                        --(last hop LAN)--
                                 |
                                 |
                         (many receivers)

                    Figure 1: Last Hop LAN
                  

Assume R1 is elected as the Designated Router. According to [RFC4601], R1 will be responsible for forwarding traffic to that LAN on behalf of any local members. In addition to keeping track of IGMP and MLD membership reports, R1 is also responsible for initiating the creation of source and/or shared trees towards the senders or the RPs.

Forcing sole data plane forwarding responsibility on the PIM DR uncovers a limitation in the protocol. In comparison, even though an OSPF DR or an IS-IS DIS handles additional duties while running the OSPF or IS-IS protocols, they are not required to be solely responsible for forwarding packets for the network. On the other hand, on a last hop LAN, only the PIM DR is asked to forward packets while the other routers handle only control traffic (and perhaps drop packets due to RPF failures). Hence the forwarding load of a last hop LAN is concentrated on a single router.

This leads to several issues. One of the issues is that the aggregated bandwidth will be limited to what R1 can handle towards this particular interface. It is very common that the last hop LAN usually consists of switches that run IGMP/MLD or PIM snooping. This allows the forwarding of multicast packets to be restricted only to segments leading to receivers who have indicated their interest in multicast groups using either IGMP or MLD. The emergence of the switched Ethernet allows the aggregated bandwidth to exceed, sometimes by a large number, that of a single link. For example, let us modify Figure 1 and introduce an Ethernet switch in Figure 2.

                              

                        ( core networks )
                          |     |     |
                          |     |     |
                         R1    R2     R3
                          |     |     |
                       +=gi0===gi1===gi2=+
                       +                 +
                       +      switch     +
                       +                 +
                       +=gi4===gi5===gi6=+
                          |     |     |
                         H1    H2     H3


            Figure 2: Last Hop Network with Ethernet Switch


                  
              

Let us assume that each individual link is a Gigabit Ethernet. Each router, R1, R2 and R3, and the switch have enough forwarding capacity to handle hundreds of Gigabits of data.

Let us further assume that each of the hosts requests 500 Mbps of unique multicast data. This totals to 1.5 Gbps of data, which is less than what each switch or the combined uplink bandwidth across the routers can handle, even under failure of a single router.

On the other hand, the link between R1 and switch, via port gi0, can only handle a throughput of 1Gbps. And if R1 is the only DR (the PIM DR elected using the procedure defined by [RFC4601]) at least 500 Mbps worth of data will be lost because the only link that can be used to draw the traffic from the routers to the switch is via gi0. In other words, the entire network's throughput is limited by the single connection between the PIM DR and the switch (or the last hop LAN as in Figure 1).

The problem may also manifest itself in a different way. For example, R1 happens to forward 500 Mbps worth of unicast data to H1, and at the same time, H2 and H3 each request 300 Mbps of different multicast data. R1 experiences packet drop once again. while, in the meantime, there is sufficient forwarding capacity left on R2 and R3 and unused link capacity between the switch and R2/R3.

Another important issue is related to failover. If R1 is the only forwarder on the last hop router for shared LAN, when R1 goes out of service, multicast forwarding for the entire LAN has to be rebuilt by the newly elected PIM DR. However, if there was a way that allowed multiple routers to forward to the LAN for different groups, failure of one of the routers would only lead to disruption to a subset of the flows, therefore improving the overall resilience of the network.

There is limitation in the hash algorithm used in this document, but this draft provides the option to have different and more consistent hash algorithms in the future.

In this document, we propose a modification to the PIM-SM protocol that allows more than one of these routers, called Group Designated Routers (GDR) to be selected so that the forwarding load can be distributed among a number of routers.

2. Terminology

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 [RFC2119].

With respect to PIM, this document follows the terminology that has been defined in [RFC4601].

This document also introduces the following new acronyms:

3. Applicability

The proposed change described in this specification applies to PIM-SM last hop routers only.

It does not alter the behavior of a PIM DR on the first hop network. This is because the source tree is built using the IP address of the sender, not the IP address of the PIM DR that sends the registers towards the RP. The load balancing between first hop routers can be achieved naturally if an IGP provides equal cost multiple paths (which it usually does in practice). Also distributing the load to do registering does not justify the additional complexity required to support it.

4. Functional Overview

In the existing PIM DR election, when multiple last hop routers are connected to a multi-access LAN (for example, an Ethernet), one of them is selected to act as PIM DR. The PIM DR is responsible for sending local Join/Prune messages towards the RP or source. In order to elect the PIM DR, each PIM router on the LAN examines the received PIM Hello messages and compares its DR priority and IP address with those of its neighbors. The router with the highest DR priority is the PIM DR. If there are multiple such routers, their IP addresses are used as the tie-breaker, as described in [RFC4601].

In order to share forwarding load among last hop routers, besides the normal PIM DR election, the GDR is also elected on the last hop multi-access LAN. There is only one PIM DR on the multi-access LAN, but there might be multiple GDR Candidates.

For each multicast flow, that is, (*,G) for ASM and (S,G) for SSM, a hash algorithm is used to select one of the routers to be the GDR. A new DR Load Balancing Capability (DRLBC) PIM Hello Option, which contains hash algorithm type, is announced by routers on interfaces where this specification is enabled. Last hop routers with the new DRLBC Option advertised in its Hello, and using the same GDR election hash algorithm and the same DR priority as the PIM DR, are considered as GDR Candidates.

Hash Masks are defined for Source, Group and RP separately, in order to handle PIM ASM/SSM. The masks, as well as a sorted list of GDR Candidates' Addresses, are announced by DR in a new DR Load Balancing GDR (DRLBGDR) PIM Hello Option.

A hash algorithm based on the announced Source, Group, or RP masks allows one GDR to be assigned to a corresponding multicast state. And that GDR is responsible for initiating the creation of the multicast forwarding tree for multicast traffic.

4.1. GDR Candidates

GDR is the new concept introduced by this specification. GDR Candidates are routers eligible for GDR election on the LAN. To become a GDR Candidate, a router MUST support this specification, have the same DR priority and run the same GDR election hash algorithm as the DR on the LAN.

For example, assume there are 4 routers on the LAN: R1, R2, R3 and R4, which all support this specification. R1, R2 and R3 have the same DR priority while R4's DR priority is less preferred. In this example, R4 will not be eligible for GDR election, because R4 will not become a PIM DR unless all of R1, R2 and R3 go out of service.

Furthermore, assume router R1 wins the PIM DR election, R1 and R2 run the same hash algorithm for GDR election, while R3 runs a different one. In this case, only R1 and R2 will be eligible for GDR election, while R3 will not.

As a DR, R1 will include its own Load Balancing Hash Masks and the identity of R1 and R2 (the GDR Candidates) in its DRLBGDR Hello Option.

4.2. Hash Mask and Hash Algorithm

A Hash Mask is used to extract a number of bits from the corresponding IP address field (32 for v4, 128 for v6) and calculate a hash value. A hash value is used to select a GDR from GDR Candidates advertised by PIM DR. For example, 0.0.255.0 defines a Hash Mask for an IPv4 address that masks the first, the second, and the fourth octets.

There are three Hash Masks defined,

The hash masks need to be configured on the PIM routers that can potentially become a PIM DR, unless the implementation provides default Hash Mask. An implementation SHOULD provide masks with default values 255.255.255.255 (IPv4) and FFFF:FFFF:FFFF:FFFF:FFFFF:FFFF:FFFF:FFFF (IPv6).

A simple Modulo hash algorithm will be discussed in this document. However, to allow another hash algorithms to be used, a 4-bytes "Hash Algorithm Type" field is included in DRLBC Hello Option to specify the hash algorithm used by a last hop router.

If different hash algorithm types are advertised among last hop routers, only last hop routers running the same hash algorithm as the DR (and having the same DR priority as the DR) are eligible for GDR election.

4.3. Modulo Hash Algorithm

Modulo hash algorithm is discussed here with a detailed description on hashvalue_RP. The same algorithm is described in brief for hashvalue_Group using the group address instead of the RP address for an ASM group with RP_hashmask==0, and also with hashvalue_SG for a the source address of an (S,G), instead of the RP address,

4.4. PIM Hello Options

When a last hop PIM router sends a PIM Hello from an interface with this specification enabled, it includes a new option, called "Load Balancing Capability (DRLBC)".

Besides this DRLBC Hello Option, the elected PIM DR also includes a new "DR Load Balancing GDR (DRLBGDR) Hello Option". The DRLBGDR Hello Option consists of three Hash Masks as defined above and also the sorted list of all GDR Candidates' Address on the last hop LAN.

The elected PIM DR uses DRLBC Hello Option advertised by all routers on the last hop LAN to compose its DRLBGDR. The GDR Candidates use DRLBGDR Hello Option advertised by PIM DR to calculate hash value.

5. Hello Option Formats

5.1. PIM DR Load Balancing Capability (DRLBC) Hello Option

            
0                   1                   2                   3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|           Type = TBD          |         Length = 4            |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                     Hash Algorithm Type                       |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+


     Figure 3: Capability Hello Option
                      

This DRLBC Hello Option SHOULD be advertised by last hop routers from interfaces with this specification enabled.

5.2. PIM DR Load Balancing GDR (DRLBGDR) Hello Option

0                   1                   2                   3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|           Type = TBD          |         Length                |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                            Group Mask                         |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                            Source Mask                        |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                            RP Mask                            |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                    GDR Candidate Address(es)                  |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+


     Figure 4: GDR Hello Option

                      

6. Protocol Specification

6.1. PIM DR Operation

The DR election process is still the same as defined in [RFC4601]. A DR that has this specification enabled on the interface advertises the new DRLBGDR Hello Option, which contains value of masks from user configuration, followed by a sorted list of all GDR Candidates' Addresses, from the highest value to the lowest value. Moreover, same as non-DR routers, DR also advertises DRLBC Hello Option to indicate its capability of supporting this specification and the type of its GDR election hash algorithm.

If a PIM DR receives a PIM Hello with DRLBGDR Option, the PIM DR SHOULD ignore the TLV.

If a PIM DR receives a neighbor DRLBC Hello Option, which contains the same hash algorithm type as the DR, and the neighbor has the same DR priority as the DR, PIM DR SHOULD consider the neighbor as a GDR Candidate and insert the GDR Candidate's Address into the sorted list of DRLBGDR Option.

6.2. PIM GDR Candidate Operation

When an IGMP/MLD join is received, without this specification, only PIM DR will handle the join and potentially run into the issues described earlier. Using this specification, a hash algorithm is used on GDR Candidate to determine which router is going to be responsible for building forwarding trees on behalf of the host.

If a router supports this specification then each of the interfaces where multicast protocol is enabled, it MUST advertise DRLBC Hello Option in its PIM Hello. Though DRLBC option in PIM hello does not guarantee that this router would be considered as a GDR candidate. For example, this router may have lower priority configured on shared LAN compare to other PIM routers. Once DR election is done, DRLBGDR Hello option would be received from the current PIM DR on the link which would contain list of GDR.

A GDR Candidate may receive a DRLBGDR Hello Option from PIM DR with different Hash Masks from those configured on it. The GDR Candidate must use the Hash Masks advertised by the PIM DR to calculate the hash value.

A GDR Candidate may receive a DRLBGDR Hello Option from a PIM router which is not DR. The GDR Candidate MUST ignore such DRLBGDR Hello Option.

A GDR Candidate may receive a Hello from the elected PIM DR, and the PIM DR does not support this specification. The GDR election described by this specification will not take place, that is only the PIM DR joins the multicast tree.

A router only acts as GDR if it is included in the GDR list of DRLBGDR Hello Option

6.2.1. Router Receives New DRLBGDR

When a router receives a new DRLBGDR from the current PIM DR, it need to process and check if router is in list of of GDR

  1. If a router is not listed as a GDR candidate in DRLBGDR, no action is needed.
  2. If a router is listed as a GDR candidate in DRLBGDR, then it need to process each of the groups in the IGMP/MLD reports. The masks are announced in the PIM Hello by DR as DRLBGDR Hello option. For each of groups in the reports it (PIM Router) needs to run hash algorithm (described in section 4.3) based on the announced Source, Group or RP masks to determine if it is GDR for specified group. If the hash result is to be the GDR for the multicast flow, it does build the multicast forwarding tree. If it is not the GDR for the multicast flow, no action is needed.

6.2.2. Router Receives Updated DRLBGDR

If a router (GDR or non GDR) receives an unchanged DRLBGDR from the current PIM DR, no action is needed.

If a router (GDR or non GDR) receives a new or modified DRLBGDR from the current PIM DR. It requires processing as described below:

  1. If it was GDR and still included in current GDR list: it needs to process each of the groups and run the hash algorithm to check if it is still the GDR for the given group.

  2. If it was not the GDR , and updated DRLBGDR from current PIM DR contains this router as one of the GDR. In this case this router being new GDR candidate MUST run hash algorithm for each of the groups (multicast flows) and for given group,

6.3. PIM Assert Modification

It is possible that the identity of the GDR might change in the middle of an active flow. Examples this could happen include:

When the GDR changes, existing traffic might be disrupted. Duplicates or packet losses might be observed. To illustrate the case, consider the following scenario where there are two streams G1 and G2. R1 is the GDR for G1, and R2 is the GDR for G2. When R3 comes up online, it is possible that R3 becomes GDR for both G1 and G2, hence R3 starts to build the forwarding tree for G1 and G2. If R1 and R2 stop forwarding before R3 completes the process, packet loss might occur. On the other hand, if R1 and R2 continue forwarding while R3 is building the forwarding trees, duplicates might occur.

This is not a typical deployment scenario but might still happen. Here we describe a mechanism to minimize the impact. We essentially want to minimize packet loss. Therefore, we would allow a small amount of duplicates and depend on PIM Assert to minimize the duplication.

When the role of GDR changes as above, instead of immediately stopping forwarding, R1 and R2 continue forwarding to G1 and G2 respectively, while, at the same time, R3 build forwarding trees for G1 and G2. This will lead to PIM Asserts.

With the introduction of GDR, the following modification to the Assert packet MUST be done: if a router enables this specification on its downstream interface, but it is not a GDR (before network event it was GDR), it would adjust its Assert metric to (PIM_ASSERT_INFINITY - 1).

Using the above example, for G1, assume R1 and R3 agree on the new GDR, which is R3. R1 will set its Assert metric as (PIM_ASSERT_INFINITY - 1). That will make R3, which has normal metric in its Assert as the Assert winner.

For G2, assume it takes a slightly longer time for R2 to find out that R3 is the new GDR and still considers itself being the GDR while R3 already has assumed the role of GDR. Since both R2 and R3 think they are GDRs, they further compare the metric and IP address. If R3 has the better routing metric, or the same metric but a better tie-breaker, the result will be consistent during GDR selection. If unfortunately, R2 has the better metric or the same metric but a better tie-breaker, R2 will become the Assert winner and continues to forward traffic. This will continue until:

The next PIM Hello option from DR selects R3 as the GDR. R3 will then build the forwarding tree and send an Assert.

The process continues until R2 agrees to the selection of R3 as the GDR, and set its own Assert metric to (PIM_ASSERT_INFINITY - 1), which will make R3 the Assert winner. During the process, we will see intermittent duplication of traffic but packet loss will be minimized. In the unlikely case that R2 never relinquishes its role as GDR (while every other router thinks otherwise), the proposed mechanism also helps to keep the duplication to a minimum until manual intervention takes place to remedy the situation.

7. Compatibility

In case of the hybrid Ethernet shared LAN ( where some PIM router enables specification defined in this draft and some do not enable)

8. Manageability Considerations

9. IANA Considerations

IANA has temporarily assigned type 34 for the PIM DR Load Balancing Capability (DRLBC) Hello Option, and type 35 for the PIM DR Load Balancing GDR (DRLBGDR) Hello Option. IANA is requested to make these assignments permanent when this document is published as an RFC. The string TBD should be replaced by the assigned values accordingly.

10. Security Considerations

Security of the new DR Load Balancing PIM Hello Options is only guaranteed by the security of PIM Hello message, so the security considerations for PIM Hello messages as described in PIM-SM [RFC4601] apply here.

11. Acknowledgement

The authors would like to thank Steve Simlo, Taki Millonis for helping with the original idea, Bill Atwood, Bharat Joshi for review comments, Toerless Eckert and Rishabh Parekh for helpful conversation on the document.

Special thanks to Anish Kachinthaya, Anvitha Kachinthaya and Jake Holland for reviewing the document and providing comments.

12. References

12.1. Normative References

[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997.
[RFC4291] Hinden, R. and S. Deering, "IP Version 6 Addressing Architecture", RFC 4291, DOI 10.17487/RFC4291, February 2006.
[RFC4601] Fenner, B., Handley, M., Holbrook, H. and I. Kouvelas, "Protocol Independent Multicast - Sparse Mode (PIM-SM): Protocol Specification (Revised)", RFC 4601, DOI 10.17487/RFC4601, August 2006.
[RFC6395] Gulrajani, S. and S. Venaas, "An Interface Identifier (ID) Hello Option for PIM", RFC 6395, DOI 10.17487/RFC6395, October 2011.

12.2. Informative References

[HELLO-OPT] IANA, "PIM Hello Options", IANA PIM-HELLO-OPTIONS, March 2007.

Authors' Addresses

Yiqun Cai Alibaba Group EMail: yiqun.cai@alibaba-inc.com
Heidi Ou Alibaba Group
Sri Vallepalli Cisco Systems 3625 Cisco Way, Sanjose, CALIFORNIA 95134 UNITED STATES EMail: svallepa@cisco.com
Mankamana Mishra Cisco Systems 821 Alder Drive, MILPITAS, CALIFORNIA 95035 UNITED STATES EMail: mankamis@cisco.com
Stig Venaas Cisco Systems 821 Alder Drive, MILPITAS, CALIFORNIA 95035 UNITED STATES EMail: stig@cisco.com
Andy Green British Telecom Adastral Park Ipswich IP5 2RE United Kingdom EMail: andy.da.green@bt.com