Transcript Goel-Group2

ToNC workshop
Next generation architecture
H. Balakrishnan, A. Goel, D. Johnson, S.
Muthukrishnan, S.Tekinay, T. Wolf
DAY 2, Feb 17 2005
Agenda
• What is architecture?
• Broad outlines?
– Focused problems
• Outreach
• Mechanisms and logistics
What is Architecture
(Systems)
• What are the optimum design decisions
• Functionality: where and why?
– Example: How much application-specific
processing should be done at the core?
• State -- allocation of state among
components; by functionality
– What scales better?
– Cross-layering vs re-layering
– Example: autonomous system structure?
• Do not take TCP, IP addresses as given
What is Arch. (theory)
• Also study optimality, algorithmic
details, lower and upper bounds
• Also, topology, network design
Why study architecture?
• Systems issues -- Current Internet is
broken.
• Theory point -- also ask, are there cliffs
we are approaching?
Broad topics
• Naming and Addressing
• Communication abstractions and primitives
– Overlays
•
•
•
•
Measurement and Diagnostics
Other networks -- sensor, adhoc
Autonomy
Interaction of economics with network
architecture
• Topology and network design decisions
Naming and Addressing
• Motivation: Mobility and multi-homing
– interesting parallels to phone networks
• What should be permanent/transient?
– Should we route based on names? Should we
have topological identifiers?
– Identifiers serve two purposes: Routing and
identification
• Algorithmic and implementation questions are
key in making the decision
– Routing efficiency; Lookup efficiency
• Interactions with security?
– Capabilities and address based access control
Primitives and abstractions?
• Is Layering Fundamental? What is the
optimum layering structure?
– What state should be exposed (if any) to end
nodes?
• Example of a different architecture
– Put/get in addition to and in place of send/receive (eg. web,
distributed hash tables)
– Initial design of the Internet was motivated by telnet
– Content/Attribute based addressing
– Implications for load balancing problems
– Implications for routing and distributed data structures
Primitives and abstractions
• What are the right primitives for network
tasks?
– Characterize capabilities/functionality?
– How should they be implemented?
– What primitives should the network provide to
facilitate peer-to-peer applications?
• Automated mapping of tasks
– Given abstract primitives, what is the best
mapping of task on to architecture/primitives
• NSF program for DDDAS (Distributed data driven
application systems)
– Covers natural, social, biological, and communication
networks
Measurement and Diagnostics
• Measurement based decision making
– Quantifiable benefits of measurement
– Massive data set analysis
– Optimum response to measurements/interpreting
measurements?
– Forensic measurements? What are the right
queries to issue?
• How do you architect networks to facilitate
measurement?
• Self-administering and the role of
measurements
Autonomous systems,
architecture, and economics
• Do autonomous systems need to exist? Is the
current structure optimal?
– Degree of autonomy/centralization?
– Are there more monolithic models that still allow
for individual latitude?
– Connections to the phone network
– Interaction of network organization and network
economics
• Volume based charging and its implication for
architecture?
Other broad areas
• Wireless and sensor net architecture
• Topology issues
– Network design
– Optical networks
• Application level networks (P2P)
Outreach
• CS looks at the world through discrete lens.
EE reduces everything to convex optimization
– More serious discrete modeling
– More serious versions of continuous algorithms
• Outreach goal: Not be isolationist
– More proposals to Nets
– Already, good collaborations with network systems
researchers. Build on that. More co-teaching etc.
– More integration with EE theory