Information Technology Reference
In-Depth Information
Dispatch Policy
For each of the assignments, the estimated
failure probability (EFP) of each task-to-worker
pair is calculated with Equation (4). Then, the
overall failure probability of the assignment is:
After getting
w
tasks to dispatch, the dispatcher
finds the optimal task-to-worker assignment that
minimizes the overall failure probability of the
w
tasks.
Suppose that each task
i
has its computation
cost (
cmp
i
) and communication cost (
comm
i
) in-
formation, and each worker
j
has its performance
(
Perf
j
) and bandwidth (
Band
j
) information. This
information is available for the dispatcher. Given
a task
i
and worker
j
, the estimated process time
of task
i
on worker
j
is:
w
∏
{
OFP
=
EFP
t
' }
.
(6)
→
p
k
k
k
=
1
There are
w!
possible assignments. The dis-
patcher calculates each assignment's
OFP
, and
compares them. The assignment with the least
OFP
is used for the task dispatch. The dispatcher
sends tasks to the workers in the dispatch window,
using the decided assignment. The workflow in-
formation is updated using the enhanced workflow
management mechanism proposed in Section 4.1.
cmp
Perf
comm
Band
EPT
T
=
i
+
i
(3)
.
i j
,
j
j
Thus, the estimated failure probability of this
assignment is:
EVALUATION RESULTS
The effectiveness of the proposed LFPD is evalu-
ated using a simulator that has been developed on
a discrete event simulation environment - OM-
NeT++ (www.omnetpp.org). The purposes of this
simulation are as follows:
EPT
EFP
=
F T
(
+
CurrentTime
−
StartTime
),
{
i
→
j
}
i j
,
j
(4)
where
F(x)
is the CDF of
TTF
is distribution;
Start-
Time
j
is the start time of worker
j
in the worker
availability status list.
Suppose that the window size is
w
, the selected
tasks are {
t
1
, t
2
, t
3
, ..., t
w
}, and the worker peers in
the dispatch windows are {
p
1
, p
2
, p
3
, ..., p
w
}. For
each permutation of {
p
1
, p
2
, p
3
, ..., p
w
}, there is
an assignment. For example, the following assign-
ment is for the permutation {
p'
1
, p'
2
, p'
3
, ..., p'
w
}:
1. To prove the effectiveness of the LFPD
policy.
2. To verify the effect of the task dispatch
window.
3. To study the effect of different parameters.
4. To analyze the effect of identifying multiple
worker types.
Baseline Policies
t
→
→
→
p
'
'
'
1
1
t
p
To discuss the effectiveness of the LFPD policy,
two baselines are used.
2
2
t
p
.
(5)
3
3
'
t
→
p
w
w
Search WWH ::
Custom Search