Broadcast Federation

1
Broadcast Federation

• An architecture for scalable
• inter-domain multicast/broadcast.

Mukund Seshadri mukunds_at_cs.berkeley.edu
with
Yatin Chawathe yatin_at_research.att.com
Fall 2002
http//www.cs.berkeley.edu/mukunds/ms/ms-report.p
s.gz
2
Motivation
Goal

• Design an architecture for the composition of
  different, non-interoperable multicast/broadcast
  domains to provide an end-to-end one-to-many
  packet delivery service.
• Design and implement a high performance
  (clustered) broadcast gateway for the above
  architecture

• No universally deployed multicast protocol.
• e.g. IP Multicast, SSM, Overlays, CDNs
• Typical problems
• Address-space scarcity (in IP Multicast)
• Limited scalability
• e.g. IP Multicast involves some form of flooding
• Need for administrative boundaries.

3
Architecture
Source

• Broadcast Network (BN) any multicast capable
  network/domain/CDN)
• Broadcast Gateway (BG)
• Bridges between 2 BNs
• Explicit BG peering
• Overlay of BGs
• Analogous to BGP routers.
• App-level
• For both app-layer and IP-layer protocols
• Less efficient link usage, and more delay
• Commodity hardware
• Easier customizability and deployability
• Inefficient hardware

Clients
BG
BN
Peering
Data
4
Naming

• A session is associated with a unique Owner BN
• For shared tree protocols
• Address space limited only by individual BNs
  naming protocols.
• URL style- bin//Owner_BN/native_session_name/flow
  _1_name/flow_2_name/?pmtr1value1...
• native_session_name - specific to the owner BN
• flow(s) related flows grouped under 1 session.
• pmtr(s) metrics, transport, single-source/multi-
  source.

Routing

• Path vector algorithm to propagate BN/BG
  reachability info
• Scope for forwarding policy hooks
• Different routes for different metrics/options
• e.g. BN-hop-count best-effortmulti-source,
  latencyreliable, etc.
• Session-agnostic, in order to avoid all BNs
  knowing about all sessions.

5
Distribution Trees

• One reverse shortest path-tree per session
• Tree rooted at owner BN.
• Soft BG tree state(Session Upstream node
  list of downstream nodes)
• Bi-directional.
• Fine-grained selectivity using SROUTE messages
  before JOIN phase.
• Built using JOIN messages, like PIM-SM.

B-Fed Mediator (an abstraction)

• Client in owner BN ? no interaction with the
  federation.
• Client not in owner BN ? needs to send JOIN to a
  BG in its BN
• So, BNs are required to implement the Mediator
  abstraction, for sending JOINs for sessions to
  BGs.
• e.g. a mediator IP multicast group
• e.g. routers or other BN-specific aggregators

6
Data Forwarding

• Decouples control from data
• i.e. control nodes from data nodes.
• TRANSLATION messages carry data path addresses
  per session
• e.g. TCP/UDP/IP Multicast addressport.
• e.g. a transit SSM network might require 2
  channels to be setup for one session.

NativeCast (interface)

• Encapsulates all BN-specific customization
• Interface to local broadcast capability
• Send and Receive broadcast data
• Allocate and reclaim local broadcast addresses
• Subscribe to and unsubscribe from local broadcast
  sessions
• Implement Mediator functionality intercept
  and reply to local JOINs
• Get SROUTE values.

7
Clustered BG design

• 1 Control noden Data nodes.
• Control node ?routing tree-building.
• Independent data paths ? flow directly through
  data nodes.
• TRANSLATION messages contain IP addresses of data
  nodes in the cluster.
• Throughput bottlenecked only by the IP
  router/NIC.
• Soft data-forwarding state at data nodes.

8
Implementation

• Linux/C event-driven program
• NativeCast for IP Multicast, simple HTTP-based
  CDN and SSM 700 lines of code each.
• Setup
• No. of sources no. of sinks no. of Dnodes.
  (so that sources/sinks dont become bottleneck).
• Each machine capable of driving a 350Mbps
  throughput TCP stream.
• 500MHz PIIIs 1 Gbps NICs.

Note maximum (TCP-based) throughput achievable
using different data message (framing) sizes is
shown above.
Variation of 1-Dnode throughput when no.
of sessions is increased to several sessions is
shown. The sources are rate-unlimited.
9
Summary

• 1 Gbps B/W with 5 Dnodes
• 2500 sessions with 5 Dnodes
• Throughput reduces with high number of sessions
• Higher framing sizes yield higher throughput.

Variation of 1-Dnode throughput when the frame
size is varied is shown. The sources are
rate-unlimited.
Future Work

• Use better I/O interfaces for higher throughput
  with large number of file descriptors (kernel
  patch or OS change required).
• What if a client has access to multiple BNs?
• Wide area deployment?