The Case For Prediction-based Best-effort Real-time - PowerPoint PPT Presentation

About This Presentation
Title:

The Case For Prediction-based Best-effort Real-time

Description:

The Case For Prediction-based Best-effort Real-time Peter A. Dinda Bruce Lowekamp Loukas F. Kallivokas David R. O Hallaron Carnegie Mellon University – PowerPoint PPT presentation

Number of Views:133
Avg rating:3.0/5.0
Slides: 28
Provided by: pete77
Category:
Tags: based | best | case | effort | prediction | real | snmp | time

less

Transcript and Presenter's Notes

Title: The Case For Prediction-based Best-effort Real-time


1
The Case For Prediction-based Best-effort
Real-time
  • Peter A. Dinda
  • Bruce Lowekamp
  • Loukas F. Kallivokas
  • David R. OHallaron
  • Carnegie Mellon University

2
Overview
  • Distributed interactive applications
  • Could benefit from best-effort real-time
  • Example QuakeViz (Earthquake Visualization) and
    the DV (Distributed Visualization) framework
  • Evidence for feasibility of prediction-based
    best-effort RT service for these applications
  • Mapping algorithms
  • Execution time model
  • Host load prediction

3
Application Characteristics
  • Interactivity
  • Users initiate tasks with deadlines
  • Timely, consistent, and predictable feedback
  • Resilience
  • Missed deadlines are acceptable
  • Distributability
  • Tasks can be initiated on any host
  • Adaptability
  • Task computation and communication can be
    adjusted

Shared, unreserved computing environments
4
Motivation for QuakeViz
Teora, Italy 1980
5
Northridge Earthquake Simulation
Real Event
High Perf. Simulation
40 seconds of an aftershock of Jan 17, 1994
Northridge quake in San Fernando Valley of
Southern California
16,666 40M x 40M SMVPs 15 GBytes of RAM 6.5 hours
on 256 T3D PEs 80 trillion (1012) FLOPs 3.5
sustained GFLOP/s 1.4 peak GB/s
Huge Model
50 x 50 x 10 km region 13,422,563
nodes 76,778,630 tetrahedrons 1 Hz frequency
resolution 20 meter spatial resolution
HUGE OUTPUT
16,666 time steps 13,422,563 3-tuples per step 6
Terabytes
6
Must Visualize Massive Remote Datasets
Datasets must be kept at remote supercomputing
site due to their sheer size
Visualization is inherently distributed
Problem
One Month Turnaround Time
7
QuakeViz Distributed Interactive
Visualizationof Massive Remote Earthquake
Datasets
Sample 2 host visualization of Northridge
Earthquake
Goal
Interactive manipulation of massive remote
datasets from arbitrary clients
8
DV A Framework For Building Distributed
Interactive Visualizations of Massive Remote
Datasets
  • Logical View Distributed pipelines of vtk
    modules
  • Example

local display and user
User feedback and quality settings
resolution
contours
ROI
interpolation
isosurface extraction
Dataset
reading
scene synthesis
rendering
interpolation
morphology reconstruction
Display update latency
deadline
Visualization Toolkit, open source C library
9
DV A Framework For Building Distributed
Interactive Visualizations of Massive Remote
Datasets
  • Logical View Distributed pipelines of vtk
    modules
  • Example

local display and user
User feedback and quality settings
resolution
contours
ROI
interpolation
isosurface extraction
Dataset
reading
scene synthesis
rendering
interpolation
morphology reconstruction
Display update latency
deadline
Visualization Toolkit, open source C library
10
Active Frames
Physical View of Example Pipeline
interpolation
isosurface extraction
scene synthesis
deadline
deadline
deadline
Active Frame n2
?
Active Frame n1
?
Active Frame n
?
  • Encapsulates data, computation, and path through
    pipeline
  • Launched from server by user interaction
  • Dynamically chose on which host each pipeline
    stage will execute and what quality settings to
    use

11
Active Frames
Physical View of Example Pipeline
interpolation
isosurface extraction
scene synthesis
deadline
deadline
deadline
Active Frame n2
?
Active Frame n1
?
Active Frame n
?
  • Encapsulates data, computation, and path through
    pipeline
  • Launched from server by user interaction
  • Dynamically chose on which host each pipeline
    stage will execute and what quality settings to
    use

12
Active Frame Execution Model
deadline
Active Frame
  • pipeline stage
  • quality params

Resource Predictions
Mapping Algorithm
Exec Time Model
CMU Remos API
Prediction
Prediction
Host Load Measurement
Network Measurement
Remos Measurement Infrastructure
13
Active Frame Execution Model
deadline
Active Frame
  • pipeline stage
  • quality params

Resource Predictions
Mapping Algorithm
Exec Time Model
CMU Remos API
Prediction
Prediction
Host Load Measurement
Network Measurement
Remos Measurement Infrastructure
14
Active Frame Execution Model
deadline
Active Frame
  • pipeline stage
  • quality params

Resource Predictions
Mapping Algorithm
Exec Time Model
CMU Remos API
Prediction
Prediction
Host Load Measurement
Network Measurement
Remos Measurement Infrastructure
15
Feasibility of Best-effort Mapping Algorithms
16
Active Frame Execution Model
deadline
Active Frame
  • pipeline stage
  • quality params

Resource Predictions
Mapping Algorithm
Exec Time Model
CMU Remos API
Prediction
Prediction
Host Load Measurement
Network Measurement
Remos Measurement Infrastructure
17
Feasibility of Execution Time Models
18
Active Frame Execution Model
deadline
Active Frame
  • pipeline stage
  • quality params

Resource Predictions
Mapping Algorithm
Exec Time Model
CMU Remos API
Prediction
Prediction
Host Load Measurement
Network Measurement
Remos Measurement Infrastructure
19
Why Is Prediction Important?
Bad Prediction No obvious choice
Good Prediction Two good choices
Predicted Exec Time
Predicted Exec Time
deadline
Good predictions result in smaller confidence
intervals Smaller confidence intervals simplify
mapping decision
20
Feasibility of Host Load Prediction
21
Comparing Prediction Models
Run 1000s of randomized testcases, measure
prediction error for each, datamine results
Inconsistent low error
Consistent high error
97.5
Mean Squared Error
75
Consistent low error
Mean
50
25
Model A
Model B
Model C
2.5
Good models achieve consistently low error
22
Comparing Linear Models for Host Load Prediction
15 second predictions for one host
97.5
75
Mean
50
25
2.5
Raw
Very
Cheap
Expensive
23
Conclusions
  • Identified and described class of applications
    that benefit from best-effort real-time
  • Distributed interactive applications
  • Example QuakeViz / DV
  • Showed feasibility of prediction-based
    best-effort real-time systems
  • Mapping algorithms, execution time model, host
    load prediction

24
Status - http//www.cs.cmu.edu/cmcl
  • QuakeViz / DV
  • Overview PDPTA'99, Aeschlimann, et al
  • http//www.cs.cmu.edu/quake
  • Currently under construction
  • Remos
  • Overview HPDC98, DeWitt, et al
  • Available from http//www.cs.cmu.edu/cmcl/remulac
    /remos.html
  • Integrating prediction services
  • Network measurement and analysis
  • HPDC98, DeWitt, et al HPDC99, Lowekamp, et al
  • Currently studying network prediction
  • Host load measurement and analysis
  • LCR98, Dinda SciProg99, Dinda
  • Host load prediction
  • HPDC99, Dinda, et al

25
Feasibility of Best-effort Mapping Algorithms
26
Feasibility of Host Load Prediction
27
Comparing Linear Models for Host Load Prediction
15 second predictions aggregated over 38 hosts
97.5
75
Mean
50
25
2.5
Raw
Very
Cheap
Expensive
Write a Comment
User Comments (0)
About PowerShow.com