Related Work

From Bits

<-- ECE1373

Contents 1 Competition! 2 Simulation 3 Emulation 3.1 Emulation vs. Simulation? 3.2 Papers 4 Traffic Generation 5 Priority Queue 6 References

[edit] Competition!

1. Internet In A Box, Network emulation on FPGA.
   • A project report suggests current work on a threaded processor (64 threads) for use in emulating many nodes.
   • i3 on Internet in a Box (PPT)
   • Experiencing MicroBlaze Hardware and Software (PPT). Lots of complaints :S

[edit] Simulation

1. A Reconfigurable Hardware Approach to Network Simulation, 1997 project using FLEX FPGAs to build an ATM-network simulator
   • Interface to get external traffic into testbed
2. Simulation Software for Communications Networks: The State of the Art.
3. Immersive Real-Time Large-Scale Network Simulation: A Research Summary. Jason Liu. IPDPS '08.
   • real-time hybrid emulation
4. A Primer for Real-Time Simulation of Large-Scale Network. Jason Liu. Simulation Symposium (ANSS) '08.
   • PRIME: a parallel real-time network simulator
5. PlanetLab
   • Overlay on real-network
6. Overcoming the Internet Impasse through Virtualization. Peterson, Shenker, Turner. Computer, 38(4):34–41, 2005.
   • Motivation for realistic network test infrastructure (they propose virtualization on overlay)
7. Large-Scale Network Simulation: How Big? How Fast?
   • ns2 on P-III 800: 94k packets/s, 538 simulated nodes
   • PDNS on 1536 x Alpha 21264 1GHz: 106M packets/s, ~4M simulated nodes
8. GENI - Global Environment for Network Innovations
9. A Simulation Study of Data Distribution Strategies for Large-scale Scientific Data Collaborations
   • A simulator for P2P that models at the file "block" granularity, but does model underlying topology.
   • Their numbers seem odd: Distributing a 1GB file takes ~20 seconds for 30 nodes.
   • Simulator runtime scales ~E³ in the number of End nodes. ~1100min simulation time for distributing a 1GB file to 200 nodes.
10. Advanced Concepts in Large-scale Network Simulation.

To Read

1. Towards Composable and Extensible Network Simulation
2. Sluggish Calendar Queues for Network Simulation
3. LARGE-SCALE NETWORK SIMULATIONS WITH GTNETS
4. Exploiting the Predictability of TCP’s Steady-state Behavior to Speed Up Network Simulation
5. Report of NSF Workshop on Network Research Testbeds

Packet-Level Simulations

1. ns2
2. GloMoSim, Qualnet (commercial)
3. Opnet

Criteria for Network Simulators (see Immersive by Liu)

1. Accuracy
2. Realism
3. Efficiency
4. Scalability
5. Repeatability (Overlay models don't have this)
6. Controllability
7. Flexibility

[edit] Emulation

[edit] Emulation vs. Simulation?

• Emulation is non-deterministic. Somehow this is ok with the users...
• Emulation has no concept of a "simulation time". Real-time is assumed.
  • Performance of emulation seems to be measured in the amount of traffic or nodes able to be emulated before emulator's CPU becomes busy. Simulation is in packets/time.
• Simulation often has unrealistic traffic generation, while emulation usually uses the real protocol processing?^[1]
• Intuitively, emulation should be faster. But comparing Simulation->6 (94k P/s) with Emulation->1 (10-15k P/s), it appears NS-2 is faster than Crystal, despite P3-800 vs. 2xP4 3.6GHz. (Scaled performance: ns-2 ~36x faster than emulation?)

[edit] Papers

1. Crystal: An Emulation Framework for Practical Peer-to-Peer Multimedia Streaming Systems
   • An emulation framework. Uses underlying OS's TCP/IP implementation.
   • Event driven, two threads per emulated peer. User protocol code needs to be written in event-driven style.
   • "Network" thread emulates bandwidth and delay, by delaying packets. No provision is made for negative delays if processing time exceeds the delay specified by the network model. "Emulation" implies real-time, and no "simulation" clock. Performance evaluation measures aggregate performance until CPU usage limits runtime, at which point the emulation is no longer real-time. It looks like they got ~10-15K packets/s out of one dual-processor server (P4 3.6GHz or K8 2.4) using some basic P2P protocol.
   • Modeling P2P requires supporting dynamically adding/dropping nodes and random dynamically-generated topologies.
2. ModelNet and paper
   • Another emulator. Maps virtual "edge nodes" onto some cluster machines, and uses other cluster machines as the internet "core". Aims to model internet-like networks.
     
3. vBET: a VM-Based Emulation Testbed
   • Network emulation using virtual machines. Runs native network applications on real OS.

[edit] Traffic Generation

• Leave good traffic generation for future work?

[edit] Priority Queue

1. Scalable hardware priority queue architectures for high-speed packet switches
   • Compares priority queue implementations based on Binary Tree of Comparators, FIFO, Shift Register, and Systolic Arrays. Also proposes two hybrid priority queues for improved speed.
2. An Efficient O(1) Priority Queue for Large FPGA-Based Discrete Event Simulations of Molecular Dynamics
   • Hide off-chip memory access latency by storing a small sorted tree on-chip and coarsely ordered segments off-chip

[edit] References

1. ↑ DummyNet: a simple approach to the evaluation of network protocols