BIER Use Cases
Cisco
7200 Kit Creek Road
Research Triangle Park
NC
27709
US
naikumar@cisco.com
Cisco
7200 Kit Creek Road
Research Triangle Park
NC
27709
US
rajiva@cisco.com
Huawei
mach.chen@huawei.com
Huawei
xuxiaohu@huawei.com
Alcatel-Lucent
600 March Road
Ottawa
ON
K2K2E6
Canada
andrew.dolganow@alcatel-lucent.com
Ericsson
300 Holger Way
San Jose
CA
95134
USA
antoni.przygienda@ericsson.com
Thomson Reuters
195 Broadway
New York
NY 10007
USA
arkadiy.gulko@thomsonreuters.com
id3as-company Ltd
UK
Dom@id3as.co.uk
DirecTV Inc
2230 E Imperial Hwy
CA
90245
USA
varya@directv.com
Nexenta Systems
451 El Camino Real
Santa Clara
CA
US
caitlin.bestler@nexenta.com
Bit Index Explicit Replication (BIER) is an architecture that provides optimal multicast forwarding through a "BIER domain" without requiring intermediate routers to maintain any multicast related per-flow state. BIER also does not require any explicit tree-building protocol for its operation. A multicast data packet enters a BIER domain at a "Bit-Forwarding Ingress Router" (BFIR), and leaves the BIER domain at one or more "Bit-Forwarding Egress Routers" (BFERs). The BFIR router adds a BIER header to the packet. The BIER header contains a bit-string in which each bit represents exactly one BFER to forward the packet to. The set of BFERs to which the multicast packet needs to be forwarded is expressed by setting the bits that correspond to those routers in the BIER header.
This document describes some of the use-cases for BIER.
Bit Index Explicit Replication (BIER) is an architecture that provides optimal multicast forwarding through a "BIER domain" without requiring intermediate routers to maintain any multicast related per-flow state. BIER also does not require any explicit tree-building protocol for its operation. A multicast data packet enters a BIER domain at a "Bit-Forwarding Ingress Router" (BFIR), and leaves the BIER domain at one or more "Bit-Forwarding Egress Routers" (BFERs). The BFIR router adds a BIER header to the packet. The BIER header contains a bit-string in which each bit represents exactly one BFER to forward the packet to. The set of BFERs to which the multicast packet needs to be forwarded is expressed by setting the bits that correspond to those routers in the BIER header.
The obvious advantage of BIER is that there is no per flow multicast state in the core of the network and there is no tree building protocol that sets up tree on demand based on users joining a multicast flow. In that sense, BIER is potentially applicable to many services where Multicast is used and not limited to the examples described in this draft. In this document we are describing a few use-cases where BIER could provide benefit over using existing mechanisms.
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 .
The Multicast L3VPN architecture describes many different profiles in order to transport L3 Multicast across a providers network. Each profile has its own different tradeoffs (see section 2.1 ). When using "Multidirectional Inclusive" “Provider Multicast Service Interface” (MI-PMSI) an efficient tree is build per VPN, but causes flooding of egress PE’s that are part of the VPN, but have not joined a particular C-multicast flow. This problem can be solved with the “Selective” PMSI to build a special tree for only those PE’s that have joined the C-multicast flow for that specific VPN. The more S-PMSI’s, the less bandwidth is wasted due to flooding, but causes more state to be created in the providers network. This is a typical problem network operators are faced with by finding the right balance between the amount of state carried in the network and how much flooding (waste of bandwidth) is acceptable. Some of the complexity with L3VPN’s comes due to providing different profiles to accommodate these trade-offs.
With BIER there is no trade-off between State and Flooding. Since the receiver information is explicitly carried within the packet, there is no need to build S-PMSI’s to deliver multicast to a sub-set of the VPN egress PE’s. Due to that behaviour, there is no need for S-PMSI’s.
Mi-PMSI’s and S-PMSI’s are also used to provide the VPN context to the Egress PE router that receives the multicast packet. Also, in some MVPN profiles it is also required to know which Ingress PE forwarded the packet. Based on the PMSI the packet is received from, the target VPN is determined. This also means there is a requirement to have a least a PMSI per VPN or per VPN/Ingress PE. This means the amount of state created in the network is proportional to the VPN and ingress PE’s. Creating PMSI state per VPN can be prevented by applying the procedures as documented in . This however has not been very much adopted/implemented due to the excessive flooding it would cause to Egress PE’s since *all* VPN multicast packets are forwarded to *all* PE’s that have one or more VPN’s attached to it.
With BIER, the destination PE’s are identified in the multicast packet, so there is no flooding concern when implementing . For that reason there is no need to create multiple BIER domain’s per VPN, the VPN context can be carry in the multicast packet using the procedures as defined in . Also see for more information.
With BIER only a few MVPN profiles will remain relevant, simplifying the operational cost and making it easier to be interoperable among different vendors.
The current widespread adoption of L2VPN services , especially
the upcoming
EVPN solution which transgresses many limitations of VPLS,
introduces the need for an efficient
mechanism to replicate broadcast, unknown and multicast (BUM) traffic towards the PEs
that participate in the same EVPN instances (EVIs). As simplest deployable mechanism,
ingress replication is used but poses accordingly a high burden on the ingress
node as well as saturating the underlying links with many copies of the same frame headed to
different PEs.
Fortunately enough, EVPN signals internally P-Multicast Service
Interface (PMSI) attribute to establish transport for BUM frames
and with that allows to deploy a plethora of multicast replication services
that the underlying network
layer can provide.
It is therefore relatively simple to deploy BIER P-Tunnels for EVPN and with that
distribute BUM traffic without building of P-router state in the core
required by PIM, mLDP or comparable solutions.
Specifically, the same I-PMSI attribute suggested for mVPN can be used easily in
EVPN and given EVPN can multiplex and disassociate BUM frames on p2mp and mp2mp
trees using upstream assigned labels, BIER P-Tunnel will support
BUM flooding for any number of EVIs
over a single sub-domain for maximum scalability
but allow at the other extreme of the spectrum to use a single BIER sub-domain
per EVI if such a deployment is necessary.
Multiplexing EVIs onto the same PMSI forces the PMSI to span more than
the necessary number of PEs normally, i.e. the union of all PEs participating in the EVIs
multiplexed on the PMSI. Given the properties of BIER it is however possible to
encode in the receiver bitmask only the PEs that participate in the EVI the
BUM frame targets. In a sense BIER is an inclusive as well as a selective tree
and can allow to deliver the frame to only the set of receivers interested
in a frame even though many others participate in the same PMSI.
As another significant advantage, it is imaginable that the same BIER tunnel needed for
BUM frames can optimize the delivery of the multicast frames though the signaling of group
memberships for the PEs involved has not been specified as of date.
IPTV is a service, well known for its characteristics of allowing both live and on-demand delivery of
media traffic over end-to-end Managed IP network.
Over The Top (OTT) is a similar service, well known for its characteristics of allowing live and on-demand delivery of media traffic between IP domains, where the source is often on an external network relative to the receivers.
Content Delivery Networks (CDN) operators provide layer 4 applications, and often some degree of managed layer 3 IP network, that enable media to be securely and reliably delivered to many receivers. In some models they may place applications within third party networks, or they may place those applications at the edges of their own managed network peerings and similar inter-domain connections. CDNs provide capabilities to help publishers scale to meet large audience demand. Their applications are not limited to audio and video delivery, but may include static and dynamic web content, or optimized delivery for Massive Multiplayer Gaming and similar. Most publishers will use a CDN for public Internet delivery, and some publishers will use a CDN internally within their IPTV networks to resolve layer 4 complexity.
In a typical IPTV environment the egress routers
connecting to the receivers will build the tree towards the ingress router connecting to the IPTV servers.
The egress routers would rely on IGMP/MLD (static or dynamic) to learn about the receiver's interest in
one or more multicast group/channels. Interestingly, BIER could allows provisioning any new multicast
group/channel by only modifying the channel mapping on ingress routers. This is deemed beneficial for
the linear IPTV video broadcasting in which every receivers behind every egress PE routers would receive
the IPTV video traffic.
With BIER in IPTV environment, there is no need of tree building from egress to ingress. Further, any addition of new channel or new egress routers can be directly controlled from ingress router. When a new channel is included, the multicast group is mapped to Bit string that includes all egress routers. Ingress router would start sending the new channel and deliver it to all egress routers. As it can be observed, there is no need for static IGMP provisioning in each egress routers whenever a new channel/stream is added. Instead, it can be controlled from ingress router itself by configuring the new group to Bit Mask mapping on ingress router.
With BIER in OTT environment, these edge routers in CDN domain terminating the OTT user session connect
to the Ingress BIER routers connecting content provider domains or a local cache server
and leverage the scalability benefit that BIER could provide. This may rely on MBGP
interoperation (or similar) between the egress of one domain and the ingress of the next
domain, or some other SDN control plane may prove a more effective and simpler way to deploy
BIER. For a single CDN operator this could be well managed in the Layer 4 applications that
they provide and it may be that the initial receiver in a remote domain is actually an
application operated by the CDN which in turn acts as a source for the Ingress BIER router
in that remote domain, and by doing so keeps the BIER more descrete on a domain by domain
basis.
Increasingly operators deploy single networks for multiple-services. For example a single Metro Core network could be deployed to provide Residential IPTV retail service, residential IPTV wholesale service, and business L3VPN service with multicast. It may often be desired by an operator to use a single architecture to deliver multicast for all of those services. In some cases, governing regulations may additionally require same service capabilities for both wholesale and retail multicast services. To meet those requirements, some operators use multicast architecture as defined in . However, the need to support many L3VPNs, with some of those L3VPNs scaling to hundreds of egress PE’s and thousands of C-multicast flows, make scaling/efficiency issues defined in earlier sections of this document even more prevalent. Additionally support for ten’s of millions of BGP multicast A-D and join routes alone could be required in such networks with all consequences such a scale brings.
With BIER, again there is no need of tree building from egress to ingress for each L3VPN or individual or group of c-multicast flows. As described earlier on, any addition of a new IPTV channel or new egress router can be directly controlled from ingress router and there is no flooding concern when implementing .
With the advent of Software Defined Networking, some operators are looking at various ways to reduce the overall cost of providing networking services including multicast delivery. Some of the alternatives being consider include minimizing capex cost through deployment of network-elements with simplified control plane function, minimizing operational cost by reducing control protocols required to achieve a particular service, etc. Segment routing as described in provides a solution that could be used to provide simplified control-plane architecture for unicast traffic. With Segment routing deployed for unicast, a solution that simplifies control-plane for multicast would thus also be required, or operational and capex cost reductions will not be achieved to their full potential.
With BIER, there is no longer a need to run control protocols required to build a distribution tree. If L3VPN with multicast, for example, is deployed using with MPLS in P-instance, the MPLS control plane would no longer be required. BIER also allows migration of C-multicast flows from non-BIER to BIER-based architecture, which makes transition to control-plane simplified network simpler to operationalize. Finally, for operators, who would desire centralized, offloaded control plane, multicast overlay as well as BIER forwarding could migrate to controller-based programming.
Virtual eXtensible Local Area Network (VXLAN) is a kind of network virtualization overlay technology which is intended for multi-tenancy data center networks. To emulate a layer2 flooding domain across the layer3 underlay, it requires to have a mapping between the VXLAN Virtual Network Instance (VNI) and the IP multicast group in a ratio of 1:1 or n:1. In other words, it requires to enable the multicast capability in the underlay. For instance, it requires to enable PIM-SM or PIM-BIDIR multicast routing protocol in the underlay. VXLAN is designed to support 16M VNIs at maximum. In the mapping ratio of 1:1, it would require 16M multicast groups in the underlay which would become a significant challenge to both the control plane and the data plane of the data center switches. In the mapping ratio of n:1, it would result in inefficiency bandwidth utilization which is not optimal in data center networks. More importantly, it is recognized by many data center operators as a unaffordable burden to run multicast in data center networks from network operation and maintenance perspectives. As a result, many VXLAN implementations are claimed to support the ingress replication capability since ingress replication eliminates the burden of running multicast in the underlay. Ingress replication is an acceptable choice in small-sized networks where the average number of receivers per multicast flow is not too large. However, in multi-tenant data center networks, especially those in which the NVE functionality is enabled on a high amount of physical servers, the average number of NVEs per VN instance would be very large. As a result, the ingress replication scheme would result in a serious bandwidth waste in the underlay and a significant replication burden on ingress NVEs.
With BIER, there is no need for maintaining that huge amount of multicast states in the underlay anymore while the delivery efficiency of overlay BUM traffic is the same as if any kind of stateful multicast protocols such as PIM-SM or PIM-BIDIR is enabled in the underlay.
Financial services extensively rely on IP Multicast to deliver stock market
data and its derivatives, and critically require optimal latency path (from
publisher to subscribers), deterministic convergence (so as to deliver
market data derivatives fairly to each client) and secured delivery.
Current multicast solutions e.g. PIM, mLDP etc., however, don't
sufficiently address the above requirements. The reason is that the current
solutions are primarily subscriber driven i.e. multicast tree is setup
using reverse path forwarding techniques, and as a result, the chosen path for
market data may not be latency optimal from publisher to the (market data)
subscribers.
As the number of multicast flows grows, the convergence time might
increase and make it somewhat nondeterministic from the first to the last
flow depending on platforms/implementations. Also, by having more protocols
in the network, the variability to ensure secured delivery of multicast data
increases, thereby undermining the overall security aspect.
BIER enables setting up the most optimal path from publisher to
subscribers by leveraging unicast routing relevant for the subscribers.
With BIER, the multicast convergence is as fast as unicast, uniform and
deterministic regardless of number of multicast flows. This makes BIER a
perfect multicast technology to achieve fairness for market derivatives per
each subscriber.
In a broadcast network environment, the media content is sourced from various
content providers across different locations. The 4k broadcast video is an evolving service
with enormous demand on network infrastructure in terms of Low latency, faster convergence,
high throughput, and high bandwidth.
In a typical broadcast satellite network environment, the receivers are the satellite
Terminal nodes which will receive the content from various sources and feed the data
to the satellite. Typically a multicast group address is assigned for each source.
Currently the receivers can join the sources using either
PIM-SM or PIM-SSM .
In such network scenarios, normally PIM will be the multicast routing protocol used to
establish the tree between Ingress connecting the content media sources to egress routers
connecting the receivers. In PIM-SM mode, the receivers relies on shared tree to learn the
source address and build source tree while in PIM-SSM mode, IGMPv3 is used by
receiver to signal the source address to the egress router. In either case, as the number of sources
increases, the number of multicast trees in the core also increases resulting with
more multicast state entries in the core and increasing the convergence time.
With BIER in 4k broadcast satellite network environment, there is no need to run
PIM in the core and no need to maintain any multicast state. The obvious advantage with
BIER is the low multicast state maintained in the core and the faster convergence (which
is typically at par with the unicast convergence). The edge router at the content source
facility can act as BIFR router and the edge router at the receiver facility can act as
BFER routers. Any addition of a new content source or new satellite Terminal nodes can
be added seamlessly in to the BEIR domain. The group membership from the receivers to
the sources can be provisioned either by BGP or SDN controller.
Distributed Storage Clusters can benefit from dynamically targeted multicast
messaging both for dynamic load-balancing negotiations and efficient concurrent
replication of content to multiple targets.
For example, in the NexentaEdge storage cluster (by Nexenta Systems) a Chunk Put
transaction is accomplished with the following steps:
The Client multicast a Chunk Put Request to a multicast group known as a
Negotiating Group. This group holds a small number of storage targets
that are collectively responsible for providing storage for a stable
subset of the chunks to be stored. In NexentaEdge this is based upon
a cryptographic hash of the Object Name or the Chunk payload.
Each recipient of the Chunk Put Request unicast a Chunk Put Response to
the Client indicating when it could accept a transfer of the Chunk.
The Client selects a different multicast group (a Rendezvous Group) which
will target the set storage targets selected to hold the Chunk. This is a
subset of the Negotiation Group, presumably selected so as to complete
the transfer as early as possible.
>The Client multicast a Chunk Put Accept message to inform the Negotiation
Group of what storage targets have been selected, when the transfer will
occur and over what multicast group.
The client performs the multicast transfer over the Rendezvous Group at
the agreed upon time.
Each recipient sends a Chunk Put Ack to positively or negatively
acknowledge the chunk transfer.
The client will retry the entire transaction as needed if there are not
yet sufficient replicas of the Chunk.
Chunks are retrieved by multicasting a Chunk Get Request to the same Negotiating
Group, collecting Chunk Get Responses, picking one source from those responses,
sending a Chunk Get Accept message to identify the selected source and having the
selected storage server unicast the chunk to the source.
Chunks are found by the Object Name or by having the payload cryptographic hash
of payload chunks be recorded in a "chunk reference" in a metadata chunk.
The metadata chunks are found using the Object Name.
The general pattern in use here, which should apply to other cluster applications,
is that multicast messages are sent amongst a dynamically selected subset of the
entire cluster, which may result in exchanging further messages over a smaller
subset even more dynamically selected.
Currently the distributed storage application discussed use of MLD managed IPV6
multicast groups. This in turn requires either a push-based mechanism for
dynamically configuring Rendezvous Groups or pre-provisioning a very large
number of potential Rendezvous Groups and dynamically selecting the multicast
group that will deliver to the selected set of storage targets.
BIER would eliminate the need for a vast number of multicast groups.
The entire cluster can be represented as a single BIER domain using only
the default sub-domain.
Each Negotiating Group is simply a subset of the whole that is deterministically
selected by the Cryptographic Hash of the Object Name or Chunk Payload.
Each Rendezvous Group is a further subset of the Negotiating Group.
In a simple mapping of the MLD managed multicast groups, each Negotiating Group
could be represented by a short Bitstring selected by a Set Identifier. The Set
Indentier effectively becomes the Negotiating Group. To address the entire
Negotiating Group you set the Bitstring to all ones. To later address a subset
of the group a subset Bitstring is used.
This allows a short fixed size BIER header to multicast to a very large
storage cluster.
Scenarios where a number of HTTP-level clients are quasi-synchronously
accessing the same HTTP-level resource can benefit from the the dynamic
multicast group formation enabled by BIER.
For example, in the FLIPS (Flexible IP Services) solution by InterDigital,
network attachment points (NAPs) provide a protocol mapping from HTTP to
an efficient BIER-compliant transfer along a bit-indexed path between an
ingress (here the NAP to which the clients connect) and an egress (here
the NAP to which the HTTP-level server connects). This is accomplished
with the following steps:
at the client NAP, the HTTP request is terminated at the HTTP level at
a local HTTP proxy.
the HTTP request is published by the client NAP towards the FQDN of the
server defined in the HTTP request
if no local BIER forwarding information exists to the server (NAP),
a path computation entity (PCE) is consulted, which calculates a
unicast path to the egress NAP (here the server NAP). The PCE
provides the forwarding information to the client NAP,
which in turn caches the result.
if the local BIER forwarding information exists in the NAP-local
cache, it is used instead.
Upon arrival of a client NAP request at the server NAP, the server NAP
proxy forwards the HTTP request as a well-formed HTTP request locally to the
server.
If no client NAP forwarding information exists for the reverse
direction, this information is requested from the PCE. Upon arrival
of such reverse direction forwarding information, it is stored in a
local table for future use.
Upon arrival of any further client NAP request at the server NAP to an
HTTP request whose response is still outstanding, the client NAP is added
to an internal request table and the request is suppressed from being
sent to the server.
If no client NAP forwarding information exists for the reverse
direction, this information is requested from the PCE. Upon arrival
of such reverse direction forwarding information, it is stored in a
local table for future use.
Upon arrival of an HTTP response at the server NAP, the server NAP
consults its internal request table for any outstanding HTTP requests
to the same request
the server NAP retrieves the stored BIER forwarding information for
the reverse direction for all outstanding HTTP requests found above
and determines the path information to all client NAPs through a binary
OR over all BIER forwarding identifiers with the same SI field.
This newly formed joint BIER multicast response identifier is used
to send the HTTP response across the network, while the procedure
is executed until all requests have been served.
Upon arrival of the HTTP response at a client NAP, it will be sent by the
client NAP proxy to the locally connected client.
A number of solutions exist to manage necessary updates in locally stored
BIER forwarding information for cases of client/server mobility as well as
for resilience purposes.
Applications for HTTP-level multicast are manifold. Examples are HTTP-level
streaming (HLS) services, provided as an OTT offering, either at the level
of end user clients (connected to BIER-enabled NAPs) or site-level clients.
Others are corporate intranet storage cluster solutions that utilize HTTP-
level synchronization. In multi-tenant data centre scenarios such as outlined
in Section 3.6., the aforementioned solution can satisfy HTTP-level requests
to popular services and content in a multicast delivery manner.
BIER enables such solution through the bitfield representation of forwarding
information, which is in turn used for ad-hoc multicast group formation at
the HTTP request level. While such solution works well in SDN-enabled intra-
domain scenarios, BIER would enable the realization of such scenarios in
multi-domain scenarios over legacy transport networks without relying on
SDN-controlled infrastructure.
There are no security issues introduced by this draft.
There are no IANA consideration introduced by this draft.
The authors would like to thank IJsbrand Wijnands, Greg Shepherd and Christian Martin for their contribution.
Dirk Trossen
InterDigital Inc
Email: dirk.trossen@interdigital.com