SAHARAI3 First Summer Retreat Six Month Planning Session

1
SAHARA/I3 First Summer RetreatSix Month Planning
Session

• Randy H. Katz, Anthony Joseph, Ion Stoica
• Computer Science Division
• Electrical Engineering and Computer Science
  Department
• University of California, Berkeley
• Berkeley, CA 94720-1776

2
Retreat Goals Technology Transfer
People Project Status Work in Progress Prototype
Technology
Early Access to Technology Promising Directions
Reality Check Feedback
3
Six-Month Plan

• Public statement of our research plans for the
  next six months
• Goals
• Feedback/comments
• Opportunities for additional industrial
  collaboration

4
Sahara

• Goal evolve (mobile) Internet architecture to
  better support multi-network/multi-service
  provider model
• Research new mechanisms, techniques for
  end-to-end services w/ desirable, predictable,
  enforceable properties spanning potentially
  distrusting service providers

5
Sahara Mechanisms for Service Composition

• Trust Mgmt/Verification of Service Usage
• Authentication, Authorization, Accounting
  Services
• Service Level Agreement Verification
• Policy Management
• Visibility into local policies to better
  coordinate global policies among (cooperating)
  service providers
• Developing inter-AS architecture for load
  balancing, performance and failure mode policies
  to be applied throughout the network
• Interoperability through Transformation
• Interoperability of data, protocols, policies
  among composed service providers
• E.g., Broadcast federatio, protocol
  transformation gateways between admin domains
  employing non-interoperable multicast protocol
  implementations
• Utility-based Resource Allocation Mechanisms
• Auctions to dynamically allocate resources
• Congestion pricing influence user behavior to
  better utilize scarce resources
• Use overlay networks for enhancing end-to-end
  services

6
Verifiable BGP Routing

• Implement the probing algorithm for verifying
  route updates on top of a Click router
• Perform microbenchmarks
• Study topology characteristics to evaluate the
  vulnerability to bogus route advertisements
• Study the implications of implement BGP as a
  link-state protocol
• Can we make the verification problem easier to
  solve?
• Target SIGCOMM submission

7
BGP Overlay Control Plane

• Quantify potential impact of our proposed system
  on a large ISPs backbone traffic
• Finish protocol specification between components
  of our architecture
• Complete analytical evaluation to calculate
  potential benefits of the approach
• Examine diversity of paths
• Map onto AS structure
• Illustrate necessary distribution of protocol
  agents
• Examine scalability of protocol

8
Wide-Area Service Composition

• Algorithms to route and forward flows over a
  mixed network of dedicated, predictable intranet
  links and VPN tunnels over the Internet (joined
  work with HP Labs)
• Consider policy-based routing that takes into
  account dynamicity and multi-party issues
• wide-area policy routing across different ISPs
• Challenges
• coordination between ingress-egress points to
  achieve policy goal
• control ingress/egress behavior

9
Wide-Area Service Composition (contd)

• Take advantage of service replicas to deal with
  Internet path failures
• Architecture design, initial evaluation done
• Wide-area experiments on service composition
• Complete setup of wide-area testbed
• More results to demonstrate usefulness of failure
  detection
• Implement and evaluate recovery algorithms
• Collect more data to better characterize the
  availability improvement due to the recovery
  algorithms

10
Wide-Area Service Composition (contd)

• Evaluation of different load-balancing algorithms
  for distributing load across replicas
• Study interaction with link-state update
  algorithm (updating the load)
• Study effects of Internet path failures on load
  variation in the system during recovery
• Study effects of scale on overlay network

11
Broadcast Federations

• Explain and solve connection scaling problem
• Involves examination and possible
  re-implementation of event dispatch mechanisms
  used
• (joined work with Yatin Chawathe, ATT Labs)

12
Co-existent overlay networks

• Analyze overheads (network and cpu) when multiple
  overlay networks of the same type co-exist
• Study interactions between overlay networks of
  different types co-existing

13
Action-based Resource Allocation

• Continue/extend work on auction-based access
  control in wireless networks
• Implement and evaluate the auctioneer adaptation
  algorithm and the bidder prediction algorithm
• Implement and evaluate the distributed auctioneer
  election protocol and the membership management
• Compare our approach to contention-based access
  control method in terms of resource utilization,
  average response time, and fairness
• Build and analyze priority-based scheduling
  algorithm for bandwidth allocations (joined work
  with HP Labs)

14
Congestion Controlled Pricing

• Complete development of WLAN testbed.
• Conduct another round of user experiments to
  verify the results of using congestion pricing to
  allocate bandwidth.
• Target Infocom paper about using congestion
  pricing to allocate bandwidth.
• Write up thesis on applying congestion on data
  and voice traffic (Jimmy)

15
Robust Routing

• Extend fault-tolerant routing infrastructure by
  designing overlay topologies able to achieve
  statistical QoS guarantees
• Develop generalized reaction mechanism for
  failures in networks based on estimating network
  failure model parameters, investigate performance
  in ad-hoc wireless networks and wide area Internet

16
OverQoS

• Finish OverQoS implementation and deploy it in a
  wide-area testbed
• Port a streaming media application on top of
  OverQoS
• More simulation studies to better understand
  OverQoS behavior
• E.g., multiple overlays competing on a link
• Study how to support applications with different
  requirements
• Goal SIGCOMM submission

17
Multipath Routing to Improve TCP Throughput

• Investigate the effect of competing TCP flows on
  multipath policies
• Infer path properties dynamically
• Investigate approaches to deployment
• receiver side modification
• use mobility servers such as home agents
• Study the effectiveness of multipath routing for
  other transport protocols
• Implementation

18
Active Flow Management in Optical Networks

• Use L7 in conjunction with Active Services for
  flexible and efficient flow management
• Intercept first packets (L7) of a flow and pass
  them to application that make policy decisions
  and reconfigure optical switches forward
  subsequent packets on the fast path
• Develop API for applications using JVM
• Develop new applications which can benefit from
  this active component ideas

19
I3 and P2P

• Applications
• Mobility
• Multicast
• Load balancing
• Enhance I3 infrastructure
• Management
• Support QoS
• Security privacy and protection to DDoS
• P2P
• Routing, DHTs in sensornets
• Study end-system behavior in todays P2P networks

20
I3 Multicast

• Implement reliability (currently, only simulation
  results)
• Experiments with other tree constructions
  algorithms
• E.g., maximize throughput
• Add accounting, access control
• Perform more measurements and evaluate the
  algorithms in a in the wide area test-bed
• Study providing multicast and reliability service
  by a 3rd party entity

21
I3 Mobility (ROAM)

• Integrate with the distributed I3 prototype (now
  I3 consists of only one server)
• Better algorithms to select the IDs that maps on
  nearby servers
• Deploy and evaluate ROAM in wide-area

22
I3 General

• Security
• Privacy and protection against DDoS
• Performance measurements of I3 infrastructure
  estimate latencies and throughputs between a
  given sets of Ids
• QoS support and load balancing
• Applications
• VoIP (leverage mobility and multicast
  functionalities)
• Server selection
• Identify relationship with Tapestry
• Study the role and the tradeoffs of having the
  infrastructure provide locality

23
Routing, Naming and Data Storage in Sensor
Networks

• Design a DHT-type algorithm for sensor networks
  that combines routing, object location and naming
  in sensor networks
• Key questions
• How useful is geographical information in solving
  this problem?
• Do such algorithms also extend to other wireless
  peer-to-peer networks?
• ACIR internship
• Prepare for prelims (Karthik and AP)

24
Load Balancing in P2P Networks

• Dynamic load balancing that adapts to current
  load patterns
• Preliminary results are encouraging
• Work on the complete details of the protocol and
  implement it in the context of one particular p2p
  data location protocol
• Work on theoretical guarantees that the protocol
  can provide

25
Tapestry Deployment and Implementation

• Focus on implementing and doing rigorous tests on
  a full implementation of Tapestry including
• full parallel insertion algorithm (as per the
  SPAA paper)
• mobililty (possibly with rapid node integration)
• fault-tolerance (fault-detection and FRLS fault
  route around algorithm)

26
BGP Awarness Routing in Tapestry

• BGP awareness Brocade
• Provide scalable, fault-tolerable and fast
  fail-over inter-domain routing
• Support application level policy requirement for
  wide-area routing
• Regression Test for Tapestry
• Design a framework and develop a set of tools to
  test Tapestry after modification or adding new
  functionalities