Response to CMU ITR Site Visit Team Questions

1
Response to CMU ITR Site Visit Team Questions

• May 26, 2005

2
Common Themes Among Responses

• First 18 months
• Each area used clean-slate approach and
  structured-network principle to achieve results
• Project Blueprint
• A blueprint is a set of design principles and set
  of compatible solutions for all major components
  of the design of a network.
• The blueprint will include an outline for the
  components of the target design, including a
  description of component synergies, interactions
  and integration activities.
• The blueprint will continually evolve over the
  duration of the project
• We have now completed the bootstrapping process
  for the components
• We now have the foundations to explore
  interactions, intersections, and synergies among
  components

3
Benefits of Synergy

• How will we create more value working together
  than we would by working independently?
• Ambitious goal and broader context help us to
  formulate bold and relevant research problems
• Breadth of disciplines required to realize
  100x100 vision is more than any one person or
  group captured by our emphasis on holistic
  design
• Many interdisciplinary research efforts

4
Example Synergistic Activities

• End-to-end lossless flow control
• Stanford/Fraser Research
• Network-wide control and management, security,
  RCP
• CMU/ATT/Stanford
• Integrated platform
• Economics-informed network design
• Berkeley/CMU/Rice
• Berkeley/Fraser/Stanford
• CMU/Fraser
• CMU/ATT/Utopia
• New default-disconnect security model
• CMU/Stanford
• Federated system for Dragnet network forensic
• CMU/Internet 2
• Many joint papers
• CMU/Rice, CMU/ATT, CMU/Stanford, Berkeley/Rice

5
Evolution of milestones - responses

• Can you go through milestones for years 1-3 and
  say where you are? Changed, who is doing it,
  status, OBE
• Wireless work retargeted to MIMO and multihop
• Backbone work on track
• Access on track
• Security on track
• Economics on track
• New work of 4D on Ethernet
• If altered (and this is not a bad thing), why?
• Several milestones on circuit switching not
  addressed. Give us more context for change.
• This line of research is on hold while we are
  pursueing VLB architecture
• Informally, tracking the evolution of milestones
  would be helpful, as in annual report.
• Excellent idea --- we will adopt it

6
Clean slate - responses

• What is success? What is failure? Deployment as
  a clean slate product or influence over the
  current world?
• Measure of success
• A coherent set of guidelines for a large scale
  network
• Fundamental understanding of network design as a
  science
• Impact on networking industry and research
  community
• We evaluate on
• 15-year retrospective shows recognizable impact
  in network
• Individual components succeeding
• Seem to be too many places where they act
  timidly.
• Our ambition is to be bold well take this
  feedback and be bolder
• Our definition of clean slate we hold nothing in
  the prior art sacred, but are not compelled to
  change everything

7
Clean slate responses -2

• Parts sometimes not well integrated
• We are actively driving to convergence on all
  parts of projects
• Security work did not seem well-integrated--not
  clean slate
• Well address this specifically on the security
  slide
• What basic principles are they applying to derive
  their clean slate.
• Our definition of clean slate we hold nothing in
  the prior art sacred, but are not compelled to
  change everything.
• Holistic design simultaneously embracing all
  aspects of network architecture.
• Structured-design principle exploitation of
  structures in networks
• We see a lack of a strawman blueprint as
  hindering collaboration and bold integrative
  thinking.
• See earlier discussion of blueprint

8
Management responses - 1

• Impression is communications and exchange of
  ideas at student and cross-institution seems to
  work, but we did not see consistent awareness of
  other work, even at one university.
• Question are management meetings meeting this
  goal?
• We strive for collaboration, and are innovating
  beyond management meetings to achieve it
• We recognize it is difficult in a large
  distributed,multi-disciplinary group and will
  continue to work to improve
• Examples
• Wiki for sharing of early results
• Student seminars started this spring
• Expanding to become project-wide technical
  meetings
• Poster sessions so everyones work is heard
• Formal student exchanges

9
Management responses - 2

• Even higher level comment/concern--even more
  coord will be needed to achieve integrative
  goals. How are you going to achieve integrated
  architecture.
• Example Management/Control and security
• Example Fiber and Radio
• Now that we have solid results in the individual
  areas, we are looking at opportunities for
  interactions and intersections between them
• We are also identifying requirements for
  integration
• On a continuing basis these will be incorporated
  into the blueprint ( described earlier)
• We will use the blueprint to guide project
  activities
• Do you have mechanisms to enforce to integration
  across projects?
• Integration achieved through encouraging
  consensus
• Multiple vehicles
• Cross project meetings
• Formal student exchanges
• Blueprint as means to capture consensus

10
Economics responses - 1

• Can you give example of how economic
  research/principles is having an influence on
  other parts of the research?
• Regional node and interconnection architectures
  to support backbone competition
• Regional node architecture ability to support
  multiple virtual access trees to facilitate
  access service competition
• Outside plant architecture to support different
  models of service level competition in fiber
  access network
• Incentives in wireless access networks location
  and fairness ownership models

11
Economics responses - 2

• Have you accepted a regulatory or industry model
  in your technical design.
• What regulatory intervention will be required in
  different scenarios?
• Example access networks
• Where feasible, prefer architecture that does not
  preclude competition
• For access network, regulatory principles of
  anti-trust and common carriage may be necessary
  to prevent domination of access service provision
  by network owner
• Opportunities identify points of leverage in
  design to drive toward preferred outcomes.
• Agree. For example regional node architecture
  and interconnection architecture to facilitate
  access service competition and backbone
  competition

12
Services and applications - responses

• Can you comment on your view of what future apps
  and services are?
• Applications cannot be foreseen
• Only known all applications will still be
  supported
• Designing a new network infrastructure having
  sufficient capacity and flexibility with the
  richness to support future applications
• What we are assuming
• General data service interfaces
• Could be different service classes (performance,
  security attributes)

13
Services and applications responses - 2

• Are you driving that future, assuming it?
• No, users and developers drive the applications
• Do you see a need for heterogeneity in access
  service?
• Yes, and our access design permits heterogeneity

14
Network control 1

• Seemed to be going back to past without
  recognition of strengths and weaknesses of that
  history.
• In the context as presented, what is new?
• Different approach from existing data network
• Centralized vs distributed
• Decision logic separation from switches vs. logic
  bundling with switches
• Decision logic separation from protocols vs.
  logic bundling with switches
• Different context from telephony networks
  (borrowed the area of separation)
• Forwarding paradigm and services (packets,
  datagrams, tunnels, VPLS, NAT, routing, ACLs)
• Different from centralized management existing
  control
• Refactoring simplies switches, control logic, and
  protocols
• 4D contributions
• Elucidated 3 principles
• Flexible decision logic with multiple objectives
• Seeking to unify control of multiple forwarding
  technologies, wire formats, and network contexts

15
Network control 2

• Confusion over role of control and
  management.
• Concern about service mgt.
• There is often confusion among these terms
• IP management/control architecture incrementally
  evolved, resulting in tremendous complexity
• We are defining a complete architecture
• 4D to date has focused on real-time network-wide
  decision making
• Provide primitives and abstractions needed to
  support management applications

16
Network control 3

• Fit this research into a holistic clean-slate
  architecture. Otherwise, cannot evaluate the
  utility.
• How does this relate to the principle of
  structured topologies?
• 4D responsible for converting physical topologies
  into structured logical topologies

17
Security - 1

• Issue of relevance and integration. Alignment
  with overall goals seemed weak.
• Achieved initial clean-slate results in security
  area
• Integration will tighten over time, following a
  staged approach to clean slate design and
  integration
• Narrow attention to individual fundamental
  capabilities
• Example Auditing forensics
• Assess their utility
• Example Dragnet investigation
• Generalize to accommodate influence network
  architecture, and to address tradeoffs
• Example Address distributed data collection,
  privacy,

18
Security - 2

• Will utilize similar methodology for new
  projects, e.g.,
• A default-disconnected network
• Support communities of connectivity
• Embed VPN-like services into 100x100 network,
  while augmenting with additional protections
• Control-plane and data-plane robustness
• Detection and resilience to communication
  interference
• Integrating cryptography into network fabric
• E.g., crypto (public) keys as addresses for
  better robustness
• Scale of 100x100 underlies challenges in each
  area
• Inspired approach in auditing forensics, for
  example

19
US centric responses

• What is US centric and what are general
  principles that will apply anywhere? Why is the
  project centered as it is?
• Projects goal is clean slate design for the
  network
• Elements of the technical design will be relevant
  in other larger scale networks.
• Economic/policy/regulatory results may not be
  relevant outside the US
• We are expending our finite energy on the US as a
  concrete example principles are likely to apply
  elsewhere

20
Wireless Response -1

• Testbed is great experiment.
• Thank you
• But we did not see research to get to 400 mb/s.
  Is there an overall view?
• What is new that this project is driving?
• Research to get to 400 Mb/sec at individual
  layers, cross-layer, and holistic design
• Physical layer MIMO algorithms to achieve
  unprecedented spectral efficiency of 20
  bits/sec/Hz (vs. 0.5 for 802.11b)
• Challenges (1) receiver-to-sender feedback based
  algorithms with limited and imperfect channel
  state information (2) current algorithms do not
  incorporate resource cost of feedback
• Medium access
• Challenges (1) exploit multi-hop tree structure
  for clean-slate MAC protocol design, (2) design
  opportunistic protocols that scale to high PHY
  rates
• Fairness Algorithm
• Without it, nodes can starve under any physical
  layer rate
• Challenge design a practical distributed
  algorithm with provable performance properties

21
Wireless Response -2

• Can you clarify interrelationship of wired infra
  and wireless?
• More integrated way of thinking about this?
• Ongoing efforts
• Development of engineering performance/cost
  models
• Wireless optimization given wire locations
• Fundamental limits of wireless, e.g., maximal
  path-length and aggregation level before a fiber
  is necessary
• Future work
• Strategic placement of wireline entry points
• Path diversity opportunities with multiple
  wireline entry points

22
Education and outreach - response

• Identify the student coauthors
• Stanford Rui Zhang, Nandita Dukkipati, Martin
  Casado, Pablo Molinero Fernandez
• CMU Aditya Ganjam, Andy Myers, Hong Yan,
  Yinglian Xie, Vyas Sekar, Debabrata Dash, Aditya
  Akella, David Maltz, Mike Collins, Anupam
  Banerjee
• Rice Joseph Camp, Arnab Chakrabarti, Violeta
  Gambiroza, Joshua Robinson, Jingpu Shi, Chris
  Steger
• Berkeley Nicolas Christin, Jens Grossklags,
  Michal Feldman, Ahsan Habib, Paul Laskowski,
  Sonja Buchegger, Pedro Ferreira, Rahul Jain
• What courses are new?
• All of them and all specifically because of this
  project
• Will software for teaching be widely available?
• Yes - it is, and is being used. Please use it in
  your classes