The Great Internet Deadlock Abhay Parekh
Dec 26, 2015
The Great Internet DeadlockThe Great Internet Deadlock
Abhay Parekh
The Decade of NetworkingThe Decade of Networking
Trillions of dollars spent on infrastructure The world changed The scalability of the internet enabled all of this
How did the architecture evolve? More importantly, how did it not evolve? What are the consequences?
The Network in 1998: Content Provider FrustrationThe Network in 1998: Content Provider Frustration
Huge potential of the Internet Delivery obvious World Wide Wait Cloud confusion: no one to ensure end-to-end performance
Conflicting service provider incentives: uneasy equilibrium New functions like multicast had limited traction Streaming audio and video exploding but no support within the
network (Broadcast.com, RealNetworks) Industry held view of the problem:
Application waits for Infrastructure
Infrastructure waits for Application
Router Lock-outRouter Lock-out
Integrating new functions into routers was hard Adding a feature required vendor buy-in Vendor buy-in required service provider buy-in Service provider buy-in required general availability
Federated nature of service providers made things even more difficult
How to satisfy content provider needs with this situation? Prevalent solution in the networking industry: build more
capacity into the network and build faster routers
Edge Networks at Colocation SitesEdge Networks at Colocation Sites
Networking products that could bypass the router by hijacking the addressing infrastructure Caches Firewalls Load Balancers
Level 4 switches
Work better in slow, unreliable networks In-network devices Stacked choke points
Content NetworkingContent Networking
Offer a service to content owners over the existing network based on application performance
Change the currency of the internet service from bits to information
Use co-location facilities at different svc providers to deploy an overlay network that overcomes cloud confusion
Enable new business applications to be deployed over the internet quickly by using the overlay network
Higher Performance allows allow media and other services
Content Internet Value ChainContent Internet Value Chain
Content Producer
Content Distribution
Network
Internet Service Provider
Consumer
$$
$$
$$
Content Producer
Media Clients
The Media Internet: 1999The Media Internet: 1999
Content Distribution Network
FastForward Networks Mission FastForward Networks Mission
To convert the Internet into the “next generation” broadcasting medium, allowing service providers and content distributors to broadcast hundreds of thousands of channels to millions of simultaneous
viewers
Why the Internet for Broadcast?Why the Internet for Broadcast?
Broadcast Today Small number of players control broadcasting Limited by geographical boundaries
Internet
Why the Internet for Broadcast?Why the Internet for Broadcast?
Anyone can be a broadcaster No geographical boundaries
Why the Internet for Broadcast?Why the Internet for Broadcast?
Audience Tracking and Management It’s not just radios and TVs!
Content Producer
Media Clients
The Evolving Broadcast InternetThe Evolving Broadcast Internet
Media Distribution Media Distribution NetworkNetwork
Content Producer
CDN
The Broadcast InternetThe Broadcast Internet
Content Producer
Content Producer
Man
agem
ent
Pla
tfo
rm
redirection management
load balancingsystem availability
network management
monitoring & provisioning
server management
viewer management
subscriptions, PPV,monitoring, Neilson ratings, targeted advertising
content management
injection & real-time control
Redirection
Media DeliverySystem
Broadcast Overlay ArchitectureBroadcast Overlay Architecture
Broadcast ManagementBroadcast Management
Scales to millions Application-level
information for management and tracking
Works across multiple networks
Content Producer event programming with ad-hoc query audience statistics
Broadcast Manager Broadcast Manager
Node Information
Stream Switchover
Policy ManagementPolicy Management
What happened: 1998-2000?What happened: 1998-2000?
Thousands of Content Providers signed up with CDNs
Even mainstream content producers got excited about internet broadcasting
Our stuff worked We did huge broadcasts e.g. Big Brother We improved lossy links and rerouted a lot of video We had Digital Island peering with AOL We made a multi-million dollar sale We merged with Inktomi for about $1.3B
What happened: 2000-2002?What happened: 2000-2002?
Most content distribution networks went bankrupt Those left are busy suing each other
The major caching companies were unable to build businesses
FastForward Products retargeted to the enterprise Many High speed access companies went bankrupt Streaming is still prevalent but has not taken off the
way we expected it to.
What went wrong?What went wrong?
Overlay Networks didn’t work as well as advertised in improving performance Always at the mercy of underlying facilities (bad links) Too Expensive to build a ubiquitous network Local caches are cheap and available Overlay Networks that did not depend on network
performance did fine Underlying access network not engineered to fulfill the Content
Networking vision DSL Access fraught with problems Cable Companies not interested in participating
Bottom Line: Can’t solve QoS problems just with Overlays!! Corollary: Can’t build the Broadcast Internet without quality
enhancing mechanisms in the network
Why is there no Network QoS?Why is there no Network QoS?
Since bandwidth is undifferentiated and is priced per bit: Vendors produce boxes for undifferentiated service
Focus on raw throughput not packet handling Boxes work to reduce the cost of transporting a bit
No reason to take QoS features seriously Standards efforts don’t take hold
Networks are engineered for better “web surfing” This applies to multicast as well…
When presented with a genuine revenue opportunity for a high speed/quality sensitive application: No way to deliver No way to peer: Content Peering is vague and QoS-based
Peering is not possible without network QoS mechanisms No way to charge: E.g.. either video is too expensive or
everything else is free Exercise one or more of the following objections:
“No Business Model” “QoS is harmful” “No congestion problem”– current quality is fine “QoS work is just incomprehensible research”
Why is there no Network QoS?Why is there no Network QoS?
Internet Infrastructure providesundifferentiated service
More capacity is thrown at the undifferentiated network, and emphasis continues on “speeding up the internet”, but this just speeds up existing applications
No future for internet media or other bandwidth intensive applications
No future for significant high speed access penetration These are huge lost opportunities!!
The Great Internet DeadlockThe Great Internet Deadlock
No BusinessModel Cop-out No way to charge, peer or deliver
high speed/quality sensitive applications
What to do?Rethink the nature of QoSWhat to do?Rethink the nature of QoS
Don’t try to come up with a “complete” definition of QoS Sometimes generality dilutes relevance Also causes confusion
Justify the extra mechanisms on the basis of tangible improved performance A “Best Effort” approach to QoS may not be effective Must involve routing
Spell out the tough choices: Carry voice and video as clearly differentiated session types
Would make peering easier Maybe this violates end-to-end arguments
Odlyzko: Have two networks one more expensive than the other. Add more mechanisms to the expensive one over time
What to do?Demonstrate feasibilityWhat to do?Demonstrate feasibility
Simulation and theory aren’t enough Need a software platform that enables a “test network” to be
built and an integrated voice/video network deployed Must not be unrelated to the routing infrastructure Must not try too hard to fit in seamlessly
Router vendors will only adopt once the demonstration is convincing
Applications will not be the bottleneck! Don’t worry too much about hardware scalability Don’t worry too much about “market timing”
User Heterogeneity- Secure and Robust - Application Heterogeneity
Work AheadWork Ahead
Can we really build an integrated services network which can support audio, video and multicast levels comparable to the alternatives? How adversely do new, in-the-network application level
gateways impact our ability to do this? How about non-responsive flows?
How do we tackle the problems of peering and pricing? These problems are hard and a “Re-thinking of The
Fundamentals” approach might work best Isolate the quality sensitive traffic to the largest possible
extent The greatest and most exciting phase of the internet will
begin when audio and video are a true part of it.