Information Technology Reference

In-Depth Information

250

0.02

rr1

random

ENR

ECJ(R=inf)

ECJ(R=0)

ε
acc

rr1

random

ENR

ECJ(R=inf)

ECJ(R=0)

240

0.015

230

220

0.01

210

200

0.005

190

0

180

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35

0

0.05 0.1 0.15 0.2 0.25 0.3 0.35

f

f

(a) Error-rate

(b) Computation time
T

Figure 29. Error rate and Computation time for fraction
f
(
acc
=0.01
,
s =0.1
,
c =0.0
,

q =0.2
,
p
d
=0
, without blacklisting).

0.02

245

r
r
1

random

ENR

ECJ(R=inf)

EC
J
(R=
0
)

rr1

random

ENR

ECJ(R=inf)

ECJ(R=0)

ε
acc

0.018

240

0.016

235

0.014

230

0.012

225

0.01

220

0.008

215

0.006

210

0.004

205

0.002

200

0

0

0.2

0.4

0.6

0.8

1

0

0.2

0.4

0.6

0.8

1

c

c

(b) Computation time
T

(a) Error-rate

Figure 30. Error rate and Computation time for colluding rate
c
(
acc
=0.01
,
s =0.1
,

f =0.35
,
c =0.0
,
q =0.2
,
p
d
=0
, without blacklisting).

jobs having incorrect results are selected repeatedly and the number of unnecessary job al-

locations increases. For that reason,
EC
J
(R = inf)
method requires a longer computation

time, as shown in Fig.27 (b).

By selecting jobs in the ascending order of
ENR
, a job having numerous results is not

selected in job scheduling even if the job has incorrect results. In this case, the unnecessary

job allocations do not become numerous, as with the
EC
J
(R =0)
. Therefore, the varia-

tions of the proposed method
EC
J
(R =0)
, which selects jobs in the ascending order of

ENR
at first (
R =0
) and then selects a job based on
EC
J
, shows better performance.

Fig.28 (b) shows computation time
T
as a function of sabotage rate
s
for
q =0.2
.

When the value of
q
becomes larger, all saboteurs are frequently detected by more spot-

checking and no incorrect result remains in the system. In this case, the computation time