Transcript pdpta99 - School of Computer Science

Dv: A toolkit for building
remote interactive visualization
services
David O’Hallaron
School of Computer Science
Carnegie Mellon University
Martin Aeschlimann, Peter Dinda, Loukas
Kallivokas, Julio Lopez, Bruce Lowekamp
Teora, Italy
1980
Jacobo Bielak and Omar Ghattas (CMU CE)
Thomas Gross (CMU CS and ETH Zurich)
David O’Hallaron (CMU CS and ECE)
Visualization of 1994
Northridge aftershock
Visualization of 1994
Northridge aftershock
Internet service models
request
server
client
response
• Traditional lightweight service model
– small to moderate amount of computation to satisfy requests
– e.g. serving web pages, stock quotes, online trading, search
engines
• Proposed heayweight service model
– massive amounts of computations to satisfy requests
– scientific visualization, data mining, medical imaging
Typical Quake visualization
pipeline
FEM solver materials
engine
database
remote
database
ROI
reading
resolution
interpolation
contours
isosurface
extraction
vtk library routines
scene
scene
synthesis
local
display
and
input
rendering
Heavyweight grid service model
WAN
Remote compute hosts
(allocated once per service
by the service provider)
Local compute hosts
(allocated once per request
by the service user)
Active frames
Active Frame Server
Input Active Frame
Frame
data
Frame
program
Output Active Frame
Active
frame
interpreter
Application
libraries
e.g, vtk
Host
Frame
data
Frame
program
Overview of a
Dv visualization service
User
inputs
Local
Dv Display
client
Request frame
Response
frames
Remote
dataset
Dv
Server
(request Response
server)
frames
Dv
Server
Remote DV Active Frame Servers
...
Response
frames
Dv
Server Response
frames
Dv
Server
Local DV Active Frame Servers
Grid-enabling vtk with Dv
request frame
[request server, scheduler, flowgraph, data reader ]
status
request server
reader
scheduler
local
Dv
client
result
...
...
response frames (to other Dv servers)
local
Dv
server
[native data, schedule, flowgraph,control ]
remote machine
(Dv request server)
local machine
(Dv client)
Scheduling Dv programs
• Scheduling at request frame creation time
– all response frames use same schedule
– performance portability (i.e. adjusting to heterogeneous
resources) is possible.
– no adaptivity (i.e., adjusting to dynamic resources)
• Scheduling at response frame creation time
– performance portability and limited adaptivity.
• Scheduling at response frame delivery time
– performance portability and greatest degree of adaptivity.
– per-frame scheduling overhead a potential disadvantage.
Scheduling scenarios
Ultrahigh
Bandwidth
Link
low-end
remote
server
powerful
local
server
Scheduling scenarios
High
Bandwidth
Link
high-end
remote
server
powerful
local
workstation
Scheduling scenarios
Low
Bandwidth
Link
high-end
remote
server
local PC
Scheduling scenarios
Low
Bw
Link
High
Bandwidth
Link
high-end
remote
server
powerful
local
proxy
server
low-end
local
PC or PDA
Summary
• Heavyweight grid service model
– service providers can constrain resources allocated to a
particular service
– service users can contribute resources to improve response
time of throughput
• Active frames
– general software framework for providing heavyweight
Internet services
– framework can be specialized for a particular service type
• Dv
– specialized version of active frame server for vizualization
– grid-enables existing vtk toolkit
– flexible framework for experimenting with scheduling algs