A Brief History of Multicast

1
A Brief History of Multicast

• Jon Crowcroft (University of Cambridge)

2
20 years in the doldrums

• IP Multicast arrived way way too soon, in 1988
• Remember, 1988 was the year TCP congestion
  control was first deployed
• The NSFNet had not been deconstructed, and was
  still the core - traffic between european
  countries went via it in the US
• Most non academic users were on 56kbps dial up at
  best
• There was no WWW. No URLs, No HTTP, no HTML, no
  XML
• There was no P2P, facebook, myspace, Halo,
  secondlife
• Steve Deering was just far too clever for us

3
The Killer App

• To make matters worse, we also built the killer
  app 20 years too soon
• Vic, Vat, IVS, RAT, Wb, Mbone family of
  real-time multimedia tools that worked rather
  well
• assumed all users on 10Mbps untrammelled internet
  access, and overprovisioned core
• Some ISPs immediately banned all non DNS UDP
  traffic
• Worse was to be discovered - unintended DDoS
• Anyone could join any/all group, despite not
  having enough downlink speed
• Anyone could setup groups
• Group management wasnt (managed)

4
Go Native? Leave comfort of home?

• First deployment of IP multicast was an overlay
  of IP tunnels (mrouted/dvmrp)
• Actually, predating this, we tried loose source
  routing, but broke all the cisco kit out there
  for unicast (documented in NANOG) - fixed by IP
  in IP tunnels, and throw away the MTU overhead
• Then when cisco deployed PIM, they broke their
  _own_ unicast (publically documented by berkeley
  at SIGCOMM a few years ago)
• And no-one could figure out inter-domain

5
Plan B Reliable Multicast?

• So in early 1990s, community tried 2nd wave
  killer app- Content Distribution via reliable
  multicast
• RMTP, PGM, Digital Fountain plethora of
  techniques
• All work pretty well in fact
• But all depend on IP multicast deployment which
  was stil waiting for Cisco to hire Isidor
  Kouvelas to fix Dinos code-)
• Meanwhile, web cache acceleration seemed to be
  the thing
• Akamai, Inktomi, and some small inkling of
  googles to come, altavista, scaled up data center
• Optical backbone meant capacity fine so repeated
  unicast not regarded as insane the way it was in
  1980s
• Where was a poor multicaster going to earn a
  crust?

6
Post hoc rationalisation of failure

• Diot et al wrote a long diatribe which was well
  received explaining why IP multicast would
  never be deployed
• Somehow, the reviewers failed to spot that the
  same explanations would have meant IP unicast was
  never deloyed-)
• Lack of security
• Lack of prevention of DDoS
• Lack of pricing model
• Lack of killer app
• .see where I am heading?

7
So now we have IP TV

• And we have DSL access (yes 6Mbpsgt original
  Ethernet)
• And we have tiny 100 machines that can do video
• And not even the massive data center and optical
  capacity can run 1M TV live channels
• So multicast to the rescue!!!
• Of course, we also have
• IGMPv3, source specific multicast, and a bunch of
  overlays that allowed us to evolve a business
  model for CDNs, to be fair-)