1 / 43

Wireless Broadcast-based Communications Systems: Overview of Current and Future Research of C.A.C.LAB, A.U.Th.

Wireless Broadcast-based Communications Systems: Overview of Current and Future Research of C.A.C.LAB, A.U.Th. . Prof. Georgios Papadimitriou, Liaskos Christos. Aristotle University of Thessaloniki Department of Informatics Email: {gp, cliaskos}@csd.auth.gr. 2. Contents. Ι. Introduction

tassos
Download Presentation

Wireless Broadcast-based Communications Systems: Overview of Current and Future Research of C.A.C.LAB, A.U.Th.

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Wireless Broadcast-based Communications Systems:Overview of Current and Future Research ofC.A.C.LAB, A.U.Th. Prof. Georgios Papadimitriou, Liaskos Christos Aristotle University of Thessaloniki Department of Informatics Email: {gp, cliaskos}@csd.auth.gr

  2. 2 Contents Ι. Introduction Past Research Objectives - Motivation ΙΙ. Theoretical Background  Push vs Pull, The Broadcast Problem Client Models  Related Work ΙΙΙ. Research of the CACLAB team  Published, Submitted, Completed  Current ΙV. Future Research  New Research Objectives - Justification  Time Schedule

  3. Προηγούμενη - Περιεχόμενα- Επόμενη 3 Ι. Introduction

  4. Past Research Objectives Προηγούμενη - Περιεχόμενα- Επόμενη Definition of the specifications of the optimal , cost-efficient Wireless Broadcast System #1 #2 Analytical Minimization of the mean client serving timne Cost Minimization (CPU, RAM) U D CONVERGENCE Impact DEPRECATED!! Perf Cost Perform. Cost Perf. Cost

  5. Motives Προηγούμενη - Περιεχόμενα- Επόμενη • Inexpensive and Efficient Communications System for: • The continued education of borderland doctors. • The exchange of hospital patients between neighborly states. Honorable Praise, 5thScientific Conference of the Department of Medicine, A.U.Th.  INTERREG IIIA GREECE-BULGARIA , Decision300531/ΥΔ4388 01/11/2005, CrossBorderHealth project  INTERREG IIIA/ARCHIMED grand, C.N. Α.1.087, IntraMed Project

  6. Προηγούμενη - Περιεχόμενα- Επόμενη 6 ΙΙ. THEORETICAL BACKGROUND

  7. Προηγούμενη - Περιεχόμενα- Επόμενη Push? Pull? Broadcast? OnQuery Server Server Client Client Send Answer ? Usefull Drop • Pros: • Best Performance • Good Channel Utilization • Application Range • Cons: • -Server Load (Scalability)? • -Cost? • -Client –side Hardware? • -Security? (DoS - Impersonation) • Pros: • Minimal Cost • Architecture Simplicity • Lightweight Clients • Absolute Server Security • Cons: • -Non-optimal performance • -Maximum Channel Utilization(++) • -Application Range (Adaptivity?)

  8. Προηγούμενη - Περιεχόμενα- Επόμενη The Broadcast Problem • Data item = page • Uniform or not pages • Page <-> Broadcast “Cost” • Define the Broadcast Schedule that minimizes the mean client serving time AND the mean broadcast cost. ++ For Adaptive Push Systems --Minimal Schedule Size --QoS: Guaranteed Serving Time

  9. Προηγούμενη - Περιεχόμενα- Επόμενη Client Probabilistic Model Common but not confining!! • Each client accesses a portion (Range) of the available pages. • Pages in Rangeare divided in sets of pages called Region, with common access probability. • TheRegions follow thezipf p.d.f : Each combination of the Range, Regionand θ parameters, defines ap.d.f. and therefore a distinct Client Case

  10. Προηγούμενη - Περιεχόμενα- Επόμενη Related Work

  11. Προηγούμενη - Περιεχόμενα- Επόμενη Broadcast Disks Defines a scheduling framework, ensuring the periodicity and proportionality characteristics of the final broadcast program. • A feedback mechanism provides a metric of the pages’ access probabilities. • A Clustering Algorithm groups the pages by access probability, into teams called “Disks” • The Virtual Disks rotate around a common axis. • Appropriate rotating speeds are chosen. • Pages are serially extracted • The Broadcast Schedule is produced. The mean client serving time depends on all of the above steps.

  12. Προηγούμενη - Περιεχόμενα- Επόμενη -LCM calculation??? -Integer division? -Zero padding extend? BDISKS: Schedule construction algorithm • Calculate a max_chunks parameter as the LCM • of the virtual disks’ speeds • Divide each disks into a number of chunks: • Serialize the chunks: • fori = 0..(max_chunks - 1) • forj = 1..NoD • Broadcast chunk Cji(i mod num_chunks(j)) • end • end

  13. Προηγούμενη - Περιεχόμενα- Επόμενη Mathematical Analysis of the no-BC problem • Let L be the number of pages that constitute the Broadcast Sequence. • For each page Pi: • Should Pi be requested by a client, the mean delay time D(Pi) will be: 13

  14. Προηγούμενη - Περιεχόμενα- Επόμενη Mathematical Analysis of the no-BC problem Lagrange Parameters Method (CACLAB Extended Analysis)

  15. Προηγούμενη - Περιεχόμενα- Επόμενη 15 ΙΙΙ. Research of the CACLAB team

  16. Προηγούμενη - Περιεχόμενα- Επόμενη Overview (part 1)

  17. Προηγούμενη - Περιεχόμενα- Επόμενη Overview(part 2)

  18. Προηγούμενη - Περιεχόμενα- Επόμενη A. TheCWDB scheme • Presentation of a complete, BDISKS based, push system that incorporates: • i. A new feedback mechanism • ii. A K-means based Virtual Disk formation approach • iii. A new Disk speed definition algorithm • Comparison with other related, popular schemes • More than 50 client cases examined for the first time. Clustering-driven Wireless Data Broadcasting (CWDB) GREEDY-based Broadcasting Procedure (GBBP)

  19. Προηγούμενη - Περιεχόμενα- Επόμενη A. Feedback Scheme Votes Registry Snapshots 1,000 queries 35,000 queries 10,000 queries

  20. Προηγούμενη - Περιεχόμενα- Επόμενη A. Clustering-driven Wireless Data Broadcasting-

  21. Προηγούμενη - Περιεχόμενα- Επόμενη A. GREEDY Based Broadcasting Procedure- GBBP

  22. Προηγούμενη - Περιεχόμενα- Επόμενη A. CWDB andGBBP Comparison

  23. Προηγούμενη - Περιεχόμενα- Επόμενη A. Client Case Definition … compare the best results achieved by each algorithm For each client case … 54 Client cases

  24. Προηγούμενη - Περιεχόμενα- Επόμενη A. Results(Ι) Range=1000, Region=30 Range=30, Region=5

  25. Προηγούμενη - Περιεχόμενα- Επόμενη A. Results(ΙΙ) Range=2000, Region=100 Range=1000, Region=50

  26. Προηγούμενη - Περιεχόμενα- Επόμενη B. Optimization Based Scheduling Don’t split equiprobable pages to different disks 1 Definition of optimal model parameters Broadcast Disks Model Υπόθεση #1 Mathematical Analysis Υπόθεση #2 Projectionofm.c.r.t D Optimal Disk Sizes Optimal Ui Same Delay contribution for all Disks. 2 Validation of Analysis’ Conclusion ► Performance Estimates VsSimulation Results ► Comparison with other heuristic methods 26

  27. Προηγούμενη - Περιεχόμενα- Επόμενη B. Mathematical Analysis • Consider N consecutive client queries… • Out of them approximately: • refer to pages belonging to Disk #i, and correspond to an aggregate delay of • and thus the approximate mean delay time will be Positioning of the di is in accordance with assumption #1 27

  28. Προηγούμενη - Περιεχόμενα- Επόμενη B. Mathematical Analysis (ΙΙ) • Due to Assumption #2: • Which is expressed in matrix form as: • And usually, thus 28

  29. Προηγούμενη - Περιεχόμενα- Επόμενη B. Optimized Broadcast Scheduling Procedure-OBSP • Define Value Sets for Δ and ΝοD  SΔ, SNoD • For each pair (SΔx SNoD ) calculate the approximate mean delay • (CAUTION! L varies!) • Keep the pair (Δoptimal, NoDoptimal) that achieves the minimum • Calculate corresponding di values  Thus [Disk Sizes]optimalis also defined. • Objective #1 Completed • Optimal Parameter values are set (Δoptimal, NoDoptimal, [Disk Sizes]optimal). • An approximation of the mean response time is calculated ( D ) 29

  30. Προηγούμενη - Περιεχόμενα- Επόμενη 30 B. Client Cases Definition Run ANALYSISfor parameters’ value sets: For each case …. VALIDATION Keep Optimal NoDo, ΔoDt SIMULATION Dt(NoDo, Δo) Ds(NoDo, Δo) Equal ? Compare ? Keep Optimal NoD, ΔDs For each case …. 54 client cases Run SIMULATIONfor parameters’ value sets: 16

  31. Προηγούμενη - Περιεχόμενα- Επόμενη B. Converge of Analysis and Simulation (O.B.P.) Typically, 31

  32. Προηγούμενη - Περιεχόμενα- Επόμενη B. Comparison with G.B.P. 32 Equally distributed Workloads Not Working!

  33. Προηγούμενη - Περιεχόμενα- Επόμενη B. Conclusion • Analytical means of research on the Broadcast Disks Method were introduced. • A new and very efficient broadcast scheduling scheme was presented, based on workload distribution assumptions. • Was found to be dominant over traditional choices in the vast majority of test cases. • Other Workload distributions need to be tested. 33

  34. Προηγούμενη - Περιεχόμενα- Επόμενη C. Extending OBSP beyond uniform Disk Workload Distribution Generalize the DWD distribution  No whole-Region-grouping  No deterministic grouping of zero-voted pages • Non-Deterministic, Page-Grouping or NDPG-Variant. • Non-Deterministic, Region-Grouping or NDRG-Variant • Deterministic, Page-Grouping or DPG-Variant • Deterministic, Region-Grouping or DRG-Variant 34

  35. Προηγούμενη - Περιεχόμενα- Επόμενη C. Results Range 500 1500 3000 2000 4500 3500 out of 5000. Region=50

  36. Προηγούμενη - Περιεχόμενα- Επόμενη D. AcceleratingUOBP.. requires 4Ν +2 computations.. Thomas Method (better alternative): 8N-4 Thus was proven that the new method requires 50% the computational power in any case.

  37. Προηγούμενη - Περιεχόμενα- Επόμενη D. Reducing the Required Memory

  38. Προηγούμενη - Περιεχόμενα- Επόμενη CACLAB Research Overview(part 2)

  39. Προηγούμενη - Περιεχόμενα- Επόμενη 39 ΙV. Future Research

  40. Προηγούμενη - Περιεχόμενα- Επόμενη 40 • #1 • Extension for Huge Databases and umber • Client model extension (drop single equivalent client approach) ΙV. New Research Objectives #2  Optimize feedback mechanisms  Security Issues with respect to adaptivity Current State: Analysis covering serving time, BS Length, QoS, optimization  Cost effective, Data scrambling, adaptivity aware algorithms (cSOA) • #3 • Noise and Error tolerance. • Caching Policies • Client queries correlation • #4 • Hybrid Pull-Push Balancing System

  41. Προηγούμενη - Περιεχόμενα- Επόμενη 41 Time Schedule

  42. Προηγούμενη - Περιεχόμενα- Επόμενη 42 Why Hybrids?

  43. Προηγούμενη - Περιεχόμενα- Επόμενη 43 Questions??

More Related