Professional Documents
Culture Documents
Cosc6321 f14 l07
Cosc6321 f14 l07
Methods
in
computer
science
Fall
2014
Lecture
7
Omprakash
Gnawali
September
17,
2014
Agenda
Research
Conference
Updates
Experiment
and
metric
design
exercise
Experiments
in
uncontrolled
environment
DescripJon
of
experiment
setup
HW
Metric
Design
Exercise
Project:
InstrucJon
system
(HCI)
Goal:
Make
furniture
assembly
faster
Possible
metrics
and
experiments?
Deployment
Experiments
RealisJc
setup
and
inputs
Uncontrolled
environment
Spans
the
parameter
space?
Wireless
Experiments
Today
Protocol
Comparison
Experiments
Run
the
new
protocol
Run
best-‐known
prior
work
Compare
SimulaJons
+
Testbed
experiments
7
Serial
Experiments
Run
one
protocol
at
a
Jme
Compare
the
results
A
B
t
8
RepeaJng
Experiments
Enough?
1
Delivery
raJo
0.8
0.6
!
0.4
0.2
0
5
10
15
20
25
30
35
Tutornet
Time
(hrs)
t
Advantages
Consistent
environment
for
both
the
protocols
Concerns
ContenJon
of
different
types
10
EvaluaJon
Strategy
Serial
Results
Experiment
==
Protocols
?
Concurrent
Experiment
Results
Ideally
same
conclusions
from
both
methods
EvaluaJng
methodologies
not
protocols
Experiments
on
Tutornet
testbed
11
Protocols
CollecJon
CTP
[Gnawali
2009]
MulJhopLQI
[TinyOS
2007]
(LQI)
12
Results
from
Serial
CTP
vs
LQI
Experiment
on
Tutornet
8
7
6
5
4
CTP
3
LQI
2
1
0
Delivery
Cost
Path
Length
Churn/node-‐hr
13
Results
from
Concurrent
CTP
vs
LQI
Experiment
on
Tutornet
8
7
6
5
4
CTP
3
LQI
2
1
0
Delivery
Cost
Path
Length
Churn/node-‐hr
14
Pueng
Concurrent
Methodology
to
Use:
Expts.
with
External
Interference
Engineered
Scenario
15
Pueng
Concurrent
Methodology
to
Use:
Experiments
in
a
Dynamic
Network
Hours
CTP
and
LQI
react
differently
to
dynamics.
16
HW
Make
a
list
of
metrics
from
the
ten
research
papers
you
selected
Most
common
metric,
most
uncommon
metric
Most
common
and
most
unusual
experiments