Rajive Bagrodia

1. Rajive Bagrodia Simulation of Scaleable Systems Computer Science Department UCLA Partial support from the TRAVLER and DOMAINS contracts from DARPA ITO R. Bagrodia, Parsec Workshop ‘98

2. 8:00 am - 8:30 am Check-In / Breakfast 8:30 am - 10:30 am PARSEC Simulation Language Rajive Bagrodia Introduction: Simulation of Complex Scaleable Systems Rich Meyer Tutorial: Language Constructs and Migrating from Maisie 10:30 am - 10:45 amBreak 10:45 am – 12:15 am Advanced PARSEC Xiang Zeng Case Study: Smple-Net Simulator Mineo Takai Scalable & Parallel Simulation in Parsec 12:15 - 1:00 pm Lunch 1:00 pm – 2:45 pm Modeling and Network Applications L. Bajaj & R. Ahuja Case Study: Network Applications/GloMoSim Cong Duc Pham Interoperability: PARSEC and HLA 2:45 pm - 3:00 pm Break 3:00 pm – 4:30 pm Other Applications Jay Martin Compose/Database Adam Rosenstein Web Caching Gary Pei Multicast Protocols Lan Wang TCP and RSVP Renato Lo Cigno ATM Networking L. Bajaj & R. Ahuja Replicated File Systems 4:30 pm - 5:30 pm Demos Rich Meyer PAVE Adam Rosenstein Web Caching Gary Pei Multicast/Multihop Mineo Takai GloMoSim 9:00 am - 9:30 am Check-In / Breakfast 9:30 am - Noon Optional Session Mineo Takai & PCL Application Design Discussion Agenda R. Bagrodia, Parsec Workshop ‘98

3. UAV Network Tactical Internet OSPF, MMWN, or WRP routing? How does TCP perform in multi-hop wireless networks? R. Bagrodia, Parsec Workshop ‘98

4. Nomadic Computing Systems Sample Query: Estimate QoS of the Situation Awareness (SA) data for a given battle scenario with hundreds of replicas Graphic for a TOC R. Bagrodia, Parsec Workshop ‘98

5. c2center weapon dmaker threat War Game Model sensor R. Bagrodia, Parsec Workshop ‘98

6. Scalable Sensor Network At what density should the sensors be spread to ensure a given probability of detection over a specified active lifetime? Distributed, multihop communications,Multiple consumers/information sinks,Dynamic circuit establishment/teardown R. Bagrodia, Parsec Workshop ‘98

7. Desired Simulator Characteristics • Scalability: ability to simulate very large systems • multi-granularity models • parallel model execution • Migration of simulator to operational prototype • automated code generation capability • common language for simulation/operational software • Integration of operational components • multi-paradigm models • Domain-specific model libraries for large scale simulations • Visual, hierarchical model design R. Bagrodia, Parsec Workshop ‘98

8. PARSEC Simulation Environment Next generation of the Maisie simulation language • Scalable model design & execution • Visual model design • Support for hierarchical, multiple granularity models • Parallel execution using diverse algorithms & architectures • Commodity Platforms • Implemented on heterogeneous platforms and OS from commodity PCs to state of the art supercomputers • Interoperability • Provides HLA-compatibility for interoperability with stimulators (e.g.., MODSAF) and simulators R. Bagrodia, Parsec Workshop ‘98

9. PARSEC Simulation Environment • World Wide Web enabled • users need not download PARSEC — models can be submitted via a Java web interface, executed remotely, and results displayed at local client • Established user base • widespread use at 300+ academic (e.g., Illinois, Columbia, Caltech, Technion Israel, University College London, …) and industrial sites (e.g., IBM, BBN, Sun, Rockwell, Northrop, ... ) • use referenced in publications at ICC, Mobicom, Usenix, etc. • User Support • mailing list, web bulletin board, … R. Bagrodia, Parsec Workshop ‘98

10. HLA-based Interoperability PAVE (Visual) Front-End Java-Based WWW Interface C++ Library (COMPOSE) PARSEC (C-Based) Front-End Virtual Time Synchronization (Simulation) Algorithms Global Event List Hybrid Conservative Optimistic ISP Portable Multi-threaded Communication Library (MPC) Linux, Windows NT, Unix MPI/AIX Pthreads on Windows NT, Linux, Solaris, IRIX MPI CH/ BSD Unix Sun Sparc 1000/ SGI Origin 2000 Uniprocessor Machine IBM SP PC Network Simulation Environment R. Bagrodia, Parsec Workshop ‘98

11. (5,p) (8,p1) (5,p) e1: clock (e1)=0 clock (e1)=5 (8,p1) (9,p2) (12,p3) DES Execution • simulation-clock:=0; • event-list:={initial set of events}; • while (termination-criteria-is-false) do { remove event with earliest timestamp from event-list: e1::(t1,p1,...); deliver event to LP p1; simulation-clock:=t1; execute code to simulate event e1; insert new events (if any) in event-list; } R. Bagrodia, Parsec Workshop ‘98

12. P2 current event 15 P1 15 13 9 9 3 Network 9 P3 P2 P2 P3 P3 12 12 event queue Parallel Simulation • Execution of a discrete-event simulation on parallel computers by distributing the global event list among processors • Solutions • Blocking synchronization (conservative algorithm) • Potentially process events out of order (optimistic) • Adaptive algorithms R. Bagrodia, Parsec Workshop ‘98

13. DESIGN IMPLEMENTATION Rapid Prototyping SIMULATION R. Bagrodia, Parsec Workshop ‘98

14. DESIGN IMPLEMENTATION Rapid Prototyping SIMULATION R. Bagrodia, Parsec Workshop ‘98

15. Rapid Prototyping • Asynchronous message passing kernel • thread create/terminate; message send & recv • MPC: messages processed in physical arrival order 4 P1 M1’ P0 P2 M1,7 R. Bagrodia, Parsec Workshop ‘98

16. 4 Rapid Prototyping • Asynchronous message passing kernel • thread create/terminate; message send & recv • PARSEC: messages processed in global order of message (event) timestamps P1 P0 M1,7 P2 R. Bagrodia, Parsec Workshop ‘98

17. 4 M’ Rapid Prototyping • Asynchronous message passing kernel • thread create/terminate; message send & recv • PARSEC: messages processed in global order of message (event) timestamps P1 P0 M1,7 P2 R. Bagrodia, Parsec Workshop ‘98

18. Implementation Laptop NOS (C Code) Net. Ctrl. Algs. Net. Ctrl. Algs. Net. Ctrl. Algs. Porting Commun. Hardware Autoporting Network Ctrl Algs (Joel Short) Simulation Maisie Network Level Maisie Node Level NOS (Maisie) Commun. Hardware Channel Model R. Bagrodia, Parsec Workshop ‘98

19. LPA Logical Model PPD PPA LPDE LPB PPE PPB PPC PPF LPC LPF Physical System PPA LPDE PPB PPC LPF Hybrid Model HybridSimulation R. Bagrodia, Parsec Workshop ‘98

20. PPA An operational process LPDE PPB A simulated process LPF PPC HybridSimulation • More useful than pure Simulation • No need to model complex operational components • Speed • Accuracy R. Bagrodia, Parsec Workshop ‘98

21. e1 e2 e3 e1 (t1) e2 (t2) e3 (t3) e1 e2 e3 t1 t2 t3 Real Time Synchronization (S to O) In a completely operational system e1, e2, e3 would arrive at t1, t2, t3, respectively In Hybrid Simulation, event e(t) may not arrive exactly at real time t • Simulator may be faster/slower than operational comp. • Non-deterministic comm. latency between S and O R. Bagrodia, Parsec Workshop ‘98

22. Design (cont’d) OtoS Operational (O) Simulated (S) StoO R. Bagrodia, Parsec Workshop ‘98

23. Existing Applications • Wargame simulations • Wireless and wired network protocols • Replicated file systems • Web caching protocol models • Database models • ATM, electronic LAN models, fiber optic network, ... • Queueing networks • Transistor level VLSI designs • MEMS sensor networks • Parallel program (MPI) simulations • Parallel IO system simulation R. Bagrodia, Parsec Workshop ‘98