Overcoming the Internet Impasse through Virtualization

1
Overcoming the Internet Impasse through
Virtualization
Larry Peterson, Scott Shenker, Jonathan Turner

• Presented by
• Aaron Ballew
• Sagar Vemuri

2
Motivation

• Impasse in the current Internet
• There is interest in studying new architectures
• New architectures cannot be evaluated
• Even if they could, cannot be deployed

3
The Idea of Architecture

• Gives system parameters
• Fewer variables to deal with
• Can be violated if it benefits you
• Promotes interoperability
• Does not mandate it
• No need to recreate everything from the bottom up
  for every new idea

4
Our Argument

• Live experimentation with new architectures is
  problematic
• Traditional testbeds have severe limitations
• Ability to evaluate and deploy new architectures
  is quite limited.
• Call to return to the roots of roots of applied
  architectural research instead of being satisfied
  with paper designs
• We will provide you the means to test your new
  architecture

5
Difficulties in adopting a new architecture

• Requires changes in routers and host software
• Requires ISPs to jointly agree to support the
  architecture
• Ad hoc work-arounds serve a valuable short-term
  purpose without offering long-term flexibility
• Live experimentation is problematic

6
Overcoming the Impasse

• Ability to easily experiment with new
  architecture on live traffic
• Availability of deployment path for
  architectural ideas to be put into practice
• Address a broad range of problems rather than
  focussing on a single narrow problem.

7
Means of testing new architectures

• Simulation / Emulation
• Physical Testbeds
• Production-oriented
• Research-oriented
• Overlays

8
Simulation

• Only as good as the model

9
Physical Testbeds

• Leased lines connecting a limited set of
  locations
• Utilize dedicated transmission links
• Small geographic extent
• Expensive to operate at a very large scale
• Expensive and time consuming to create for each
  experiment

10
Production Testbeds

• Support real traffic from real users
• Provide valuable information about the
  operational behavior of an architecture
• Users have no choice on participation
• Extremely conservative in their experimentation

11
Research Testbeds

• Driven by synthetically generated traffic and/or
  a small collection of users
• Do not carry traffic from a wide variety of real
  users
• Results much less indicative due to lack of real
  operational viability

12
Overlays

• Not limited geographically
• Users can access from anywhere
• Usage can be voluntary
• Users can decide whether or not to participate in
  an overlay
• No significant expenditure

13
Overlays 2

• Underutilized tool due to high barrier to entry
• Largely seen as a way of deploying narrow fixes
  to specific problems
• Little thought devoted to deal with the big
  picture
• Architecturally tame
• Most assume IP as the architecture inside the
  overlay itself
• Need a philosophical change rather than a
  technical change in how they are used

14
Virtual Testbed

• Proposed to overcome the problems of Testbeds and
  Overlays
• Support multiple simultaneous architectures
• Reduce the barrier to entry for new architectural
  ideas
• Provides a clean path for radical new
  architectures to be globally deployed

15
Virtual Testbed 2

• Two basic components
• Overlay substrate
• A set of dedicated but multiplexed overlay nodes
• Allows multiple experiments to be run
  simultaneously
• Proxy mechanism
• Allows any host to opt-in to a particular
  experiment
• Treats a nearby overlay node as the hosts first
  hop router

16
The Substrate

• Could use the existing PlanetLabs infrastructure
• Consists of a set of virtual routers connected
  into whatever topology the experimenter selects
• Runs whatever the experimenter designs
• Does not have to be IP!

17
Opting In

• Can take advantage of existing DNS system, to
  direct users into or not into the VT.
• The VT is then free to do whatever it wants with
  the packets, using whatever IP or non-IP
  protocols are appropriate to service the packet,
  and tunneling over protocols it hopes to replace

18
Considerations

• Security
• By using non-IP inside the VT, perhaps interfere
  with IP-based security
• Interesting problem, and worth looking at. Enjoy
  yourself.
• QoS
• Nothing is perfect. Intra-VT QoS is still
  present. The underlying variations can be
  treated as Noise.
• No reason to abandon the study of noisy signals.
  
• QoS may not be pertinent to a particular
  experiment anyway.

19
Deployment

• Leverage the strength of overlays
• Can occur independently without any coordination
  between various deployments
• No mandate for a particular architecture. Market
  forces will address, just as it does today.

20
Purists

• Believe that IP is the single universal protocol
• Overlays are necessary evils that are reluctantly
  tolerated
• Virtualization is only a means by which new
  architectures are installed, not a fundamental
  aspect of the architecture itself
• Aim for flexibility of an architecture giving
  importance to long term goals

21
Pluralists

• View IP as only one of the several different
  components of the Internet
• Overlays are just one more way to deliver
  services to the users
• Architecture evolves dynamically and is the union
  of existing overlays and protocols
• More emphasis of short-term improvements

22
Virtualization

• Virtual Testbed uses virtualization in two
  crucial ways
• Used in a typical overlay sense
• The client proxy plus the virtual links allow the
  overlay to be qualitatively equivalent to a
  native network
• Multiplexing allows many virtual testbeds to
  operate simultaneously
• Greatly reduces the barrier-to-entry

23
Virtualization

• Virtualization Techniques are not tied to the
  architecture being tested.
• If architectural changes are rare
• Virtualization is only a means to accomplish the
  architectural shifts
• If Internet is in a constant change of flux
• Virtualization plays the central role

24
To Reiterate

• We recognize the interest and value in
  architectural research
• We want researchers to have a place to work and
  test their ideas
• We are not interested in mandating a particular
  solution to anyone

25
Thank You

• Questions?