Discussion

Home
Syllabus
Papers
Presentations
Discussion
Handouts
Projects

These are a few of the topics that were discussed during class.

09/12/02: Towards an Active Network Architecture

bulletWhat are the characteristics of an active network?
bulletWhat is "technology push" and "application pull"? What does this mean in the context of active networks?
bulletWhat does an active network do in the context of a video compression application?
bulletWhat levels of programming abstraction are there?
bulletHow is the "program" distributed? What are "foundation components"?
bulletWhat is "sandboxing" and why is it necessary?
bulletHow could an active network be deployed?

09/17/02: Smart Packets: Applying Active Networks to Network Management

bulletWhat does Moore's Law say? How is it relevant for active networks and network management?
bulletWhat does the ANEP header do?
bulletWhat management applications could use active networks?
bulletHow does the traceroute application work in normal and in active networks?
bulletWhat does the data path in an active router look like? Where are scalability problems?
bulletHow is security achieved?
bulletWhat does an active network testbed look like?

09/19/02: A Scalable High-Performance Active Network Node

bulletWhy is the proposed architecture scalable?
bulletWhat does the data path look like?
bulletWhy does this architecture promise higher performance than the others we have discussed?
bulletWhere does other active network software (e.g., ANTS) fit into this architecture?
bulletWhy is the use of selectors similar to MPLS?
bulletWhere does the packet processing code come from?
bulletHow is "safe" execution achieved?

09/24/02: Active Reliable Multicast

bulletWhat are problems with non-active multicast? What is NACK implosion?
bulletHow do the three components of ARM (data caching, NACK suppression, scoped retransmission) solve these problems?
bulletWhat is soft state?
bulletHow does the sender-receiver RTT play a role in multicast?
bulletHow does a partially active network help for ARM?

09/26/02: An Active Router Architecture for Multicast Video Distribution

bulletHow can wave video be scaled? Why does this not work with MPEG? What are the benefits of MPEG in general?
bulletHow is this system different from reliable multicast?
bulletHow does the active router make the decision to forward or drop a packet?
bulletHow realistic is the test network? What is good about such a setup?
bulletWhy would a complete network simulation give more meaningful results?
bulletWhat does the evaluation of bursts of cross traffic show?

10/01/02: NodeOS and ABONE

bulletWhat are the functions of NodeOS, EEs, and AAs?
bulletWhat are the design goals of the NodeOS?
bulletHow is memory and processing time managed and allocated to AAs?
bulletHow does the ABONE work?

10/03/02: PLAN and SNAP

bulletWhat is the difference between safety and security?
bulletHow do PLAN and SNAP achieve safety and security?
bulletHow is resource consumption (CPU, memory, bandwidth) bound in PLAN and SNAP?
bulletWhat guarantees can be made by forcing the execution time to be proportional to the packet size?
bulletWhy can applications not be checked for safety and security before installation?
bulletWhat do the measurement results show?

10/08/02: Strong Security for Active Networks

bulletWhat is integrity, authenticity, privacy?
bulletHow impact does hop-by-hop and end-to-end integrity have on active networking? What about privacy?
bulletWhat are the trust issues in active networks?
bulletWhy do the authors propose to split the packet payload into fixed and modifiable content?

10/10/02: Darwin: Customizable Resource Management for Value-Added Network Services

bulletWhy do we need resource reservation?
bulletWhy should resources be allocated hierarchically?
bulletHow can computation and bandwidth be traded off? When is such a tradeoff useful?
bulletWhat are the functions of Xena, delegates, schedulers, and Beagle?

10/17/02: Enabling Conferencing Applications on the Internet using an Overlay Multicast Architecture

bulletHow is overlay multicast different from IP multicast and active multicast?
bulletWhat are the benefits and drawbacks of overlay networks?
bulletWhy is a mesh used instead of just a spanning tree?
bulletHow are bandwidth and latency used to determine the connections? How does this influence the performance of the overlay network?
bulletWhat do the measurements in the paper show? What is the rank order and why is it used?
bulletWhat kind of overhead is associated with overlay networks?

11/12/02: Network Processors: IBM PowerNP and Intel IXP

bulletHow is the parallelism in networking workloads reflected in network processor architectures?
bulletWhat are the key components of these network processors?
bulletWhat are the differences between the PowerNP and the IXP?
bulletWhat is the application domain for these network processors?
bulletWhat kind of functions are implemented on co-processors? Why is it worthwhile doing that?

11/14/02: CommBench - A Telecommunications Benchmark for Network Processors

bulletWhy do traditional processor benchmarks (e.g., SPEC) not capture networking characteristics?
bulletWhat applications are suitable for NP benchmarks? How can different levels of processing be distinguished?
bulletWhat benchmark metrics are relevant to network processors?
bulletHow is computational complexity defined in the context of CommBench?
bulletWhat design implications can be drawn from the benchmark results?

11/19/02: Scheduling Computations on a Software-Based Router

bulletHow is scheduling done on a workstation processor with a typical operating systems? Why does a network processor require a different approach?
bulletWhat is live-lock and what causes it? How can it be prevented?
bulletWhy should input processing, forwarding, and link scheduling be handled by independent processes/components?
bulletWhat is batching and how does it help improve performance? What is the draw-back of batching?

11/21/02: Dynamic Hardware Plugins

bulletWhy can/should we use FPGAs on programmable routers?
bulletHow do FPGAs compare to general purpose processors and ASICs in terms of flexibility, performance, and development time?
bulletWhat are the benefits of using hardware plugins instead of freely programming the FPGA?
bulletWhat does the data flow look like in the proposed architecture?

11/26/02: Methodology and Simulator for the Study of Network Processors

bulletWhy do we want to simulate network processor?
bulletWhat components of the network processor should be simulated?
bulletWhat is the basic architecture of Cisco's Toaster?
bulletWhat traces are used and how were they obtained?
bulletWhat do the simulation results yield?

12/03/02: Design Space Exploration

bulletWhy do analytic design space exploration?
bulletWhat components of a network processor should be modeled?
bulletWhat are possible optimization criteria?
bulletWhat are arrival and service curves?
bulletWhat is Pareto-optimal?

12/05/02: Commercial Network Processors

bulletWhat characteristics are similar in most commercial NPs?
bulletWhy is it important to have a sophisticated software development kit for NPs?
bulletWhat are typical configurations for number of processors, on-chip cache, clock rates, power consumption?
bulletWhat trends can be observed for newer NP generations?
  2002 by Tilman Wolf