Transcript Multicast, etc

Network Architecture (R02) #3
Multicast and Deployment
Jon Crowcroft,
http://www.cl.cam.ac.uk/~jac22
http://www.cl.cam.ac.uk/teaching/1011/R02
+ve multicast for publisher

Capacity win at sender



For very very popular live content



Some capacity win in network
Latency lower potentially
Might be only technique
Might steal broadcast tv/radio customers into ISP
customer base
Ease of writing applications


Especially collaborative tools (wb/nte etc) for
ASM
Except many-to-many reliability very hard:(
-ve for IP Multicast
Problems abounded early on with ASM
Only tunnel based implementation

1.
1.
2.
Fanout limit 6! (ethernet:-(
DVMRP limits
First native code broke unicast
2.
1.
2.
3.
Berkeley report
Igmp v1 timers
PIM route loop!
Time to market

Slowed by bugs


New Needs




Since then, timeshifting for media (tivo)
Share trading still big customer
Urgent (zero day defense) software update
See vigilante for example
Market needs



ISP loses money from multicast:)
Big publisher wins
ISP loses money from CDNs too….
Architectural list of problems

Some are same as for CDN


Even Inter-domain


Except who manages things
E.g. neutrality debate is basically about
CDN v. ISP and impact on peering
Security perhaps main pb


Esp. potential DDoS leverage…
multiplier bad
State complexity v. packet
forwarding

PGM has same state complexity as SSM



Digital Fountain allows some timeshift


Latches rx nack state on fwd filter
Can have local rtx too
Complex to manage
All single ISP/AS, not interdomain



Interdomain addr allocation not solves really
Traffic breaks peering rules (reverse path:(
So does interdomain P2P and CDN traffic, though
Privacy, Integrity, DDoS prev

Need key distribution




key graph for ASM
Re-key cost (for fwd/reverse secrecy)
expensive
Need source management
Need receiver control


Police IGMP join to group per group
Same as CDN, but free there already

Auth is from subscriber to cloud
ASM, SSM, CDN, pub/sub, p2p

Traffic Engineering?


Energy reduction?



Per source? Per Content?
With timeshifting?
V. p2p?
Hybrids?



Server assist P2P?
Multicast assist CDN?
In Net record/timeshift?
general deployment tricks…




Overlay
Underlay
Tunnel
Map/encap
Overlays and Underlays

IP was once an overlay




On x.25 and on PSTN
Then moved down to bare metal
ip on photons too
Overlays are a deployment/evolutionary
route

see national academy of science on “looking
over the fence at network research)
Other overlay schemes

Mbone, 6bone, abone





Typically tunnel…but


RON
Application Layer Multicast
IP on SMS and on Fb!
IP on DNS
IP in IP overhead
IP on VC - complexity



Circuit setup/teardown
#circuits, qos
routing
Underlays


more revolutionary than overlays
Not many examples 
MPLS, gMPLS


3gpp/cellular


Finesse management/routing problem
Finesse mobility&billing problem
Other?
Multicast, mobile, multihome

Similar



More than 1 address (at time, in space)
Dynamic - add receiver/delete
Different



Group dynamics possibly fast
Mobility limited by physics
Multihoming slow
Next talk for 25/10/10
The Location/Identity split has been pushed since
8+8 as the solution to mobility (and multihoming)
Discuss the various approaches (e.g. 8+8 and LISP)
and their various merits and demerits