Timed Mobile Agent Planning for Distributed Information Retrieval

odecrackAI and Robotics

Oct 29, 2013 (3 years and 5 months ago)

71 views

Timed Mobile Agent Planning for Distributed
Information

Retrieval
Jin
-
Wook Baek, Gyu
-
Tae Kim, Heon
-
Young Yeom

School of Computer Science and Engineering, Seoul National University

Seoul, Korea

{jwbaek,gtkim,yeom}@arirang.snu.ac.kr


ABSTRACT

The most signi
ficant performance factors in Mobile Agent
Planning (MAP) are; (a) the number of mobile agents, (b) the total
routing time consumed by the participated
agents
, and (c) the
time constraints, (such as information ready time and deadline
triggered unto each
nodes to be visited).

We propose the Timed Mobile Agent Planning (TMAP), a time
constrained mobile agent planning method for finding the minimal
number of agents and the best scheduled agents' itineraries for
information

retrieval f
rom a distributed compu
ting environment
processed
under
the
time
-
constraints while keeping the
completion and the total
routing

time minimal.

Experimental results show that th
is

method
is

required and highly
applicable directly to the time
-
constrained distributed information
ret
rieval
environment which
has

a
time window which consists of
a “ready
-
line” (information ready time) and
a
“deadline” (the end
of
the
information

s

lifetime
).

1.

I
NTRODUCTION

One major potential application area for mobile agents is
distributed information
retrieval
,

where huge amount
s

of data can
be accessed across
various
network
s
. Information is spread across
several nodes and
it
is now common for these nodes to be
geographically separated
[
3][4
]
.

If mobile agents know the network statistics
,

such as the

latency
and bandwidth of the links between nodes, and the computational
load on each node, we could better plan

these systems
.
Basically,
there are t
wo factors
that must be considered in
planning
the
mobile agents


itineraries. They are (a)
number of mobi
le agents
created
and

(b) total routing time traveled by the participating
agents.

In addition to th
ose

two factors,
certain
time constraint
s

can be
found in an application system. In an information retrieval system,
the time constraints are ready
-
line and

deadline
.

Ready
-
line is the earliest time that the expected information is
available or to be refreshed by updating the contents with the
newest information. Deadline is the latest time that information is
still available to an incoming arrival agent, whi
ch is the ready
-
line
plus the information

s lifetime.
News service
s

or stock price quote
service
s

could be situation
s

where time
-
constraint

must be
considered.

If the agent sent
,

arrives early
,

before the specified
update time and gathers
extra
information
, the
receiver
may get old
or corrupted information.

Several works have
formulate
d

mobile agent

planning
, namely
with
the Traveling Agent Problem (TAP), analogous to the Traveling
Salesman Problem,
in
deciding the sequence of nodes to visit to
minimize the

total expected time the desired information is found

[
2][5][6][7
].

Two
MAP (
Mobile Agent Planning) algorithms, BYKY1 and
BYKY2
, were

developed
[
1]. These algorithms deal with the first
two factors of agent planning; minimal number of agents and total
rout
ing time consumed by all the agents sent for an information
gathering session. However, if the situation is skewed a bit by any
kind of time constraints, such as information retrieval deadline, the
existing MAP algorithms must be amended to support the tim
e
constraints.

In this paper, we propose

that the
Timed Mobile Agent Planning
(TMAP) method
is a descendant of the above MAP algorithms
used
to find the
aforementioned first
two planning factors
and
including the third factor,
time

constraints, in d
istribu
ted
information retrieval systems. Though not a primary goal, total
routing time is still important in MAP
and/or TMAP
problem
s
,
these methods

need to factor
the traveling time of mobile agent
s

allocated in
lo
cal network
s and
it is

also

desirable to handle

the
routing time to a certain degree.
To make this possible, the
2OPT,
a simple well
-
known TSP algorithm, is employed in the
last stage
of the
proposed algorithm.

2.

PROPOSED METHODS

Table 1. Notations used in this paper

Symbols

Description



n

r

H



tour

h
1
,
h
2
,

,h
n

Number of nodes excluding the home node

Number of mobile agents needed

Home node

Completion time

A sequence of nodes which be visited by mobile agent

Node identifiers

Ready
(
h
i
)

Dead
(
h
i
)

Comp
(
h
i
)

Ls
(
h
i
,
h
j
)

Tour
(
A
i
)

Union
(
T
i
,...,
T
j
)

“ Ready
-
line

: t h
e t i me t hat i nformat i on i s ready at node
h
i

“ Deadl i ne”

of i nformat i on pi ckup at node
h
i

“ Comput at i on”

t i me at node
h
i

Short est

l at ency bet ween nodes
h
i

and
h
j

tour

of mobi l e agent
A
i

Concat enat i on

of
tours
, where
T
i
,...,
T
j

represent tours

Informally, the
timed mobile agent problem is described as
follows:

“There are
n

nodes where information is serviced. Each node has a
computation time and a time window. A computation time is
required for a mobile agent to perform the task at that node. A
mobile agent's t
ask is valid only during the time

window of a node.
A time window of a node is defined by the information ready
-
line
and the deadline.
Latencies for mobile agent to move between each
node
h
i

and
h
j

are
assumed to be
known. The planning problem is
to find t
he number of mobile agents and the each agent's itinerary
to minimize the completion time and the total execution time
under the constraint
s

of time window
s
.”

For

an agent

s process is valid
within

a time

window
.
T
he low
er
bound of the expected completion

time can
be defined

as follows:


TMAP
, the proposed algorithm,

is introduced below.

Algorithm

1.

The

TMAP

Algorithm

1.

Pre
-
processing:

Shortest
-
Latency Network

To get the latencies of all pairs of nodes before e
ntering the main body,
process the all
-
pairs shortest
-
path algorithm, and construct a shortest
-
latency network graph.

2.

Phase 1
:

Sorting the nodes

Sort the nodes in terms of
expression (1)

from home node
by decreasing
order

1)
Let the sequence (
i
a1
,i
a2
,i
a3
,

,i
an
) as the resulting sequence

2) The expected completion time


is calculated by expression (1), i.e.,
routing time of the node i
a1

3.

Phase 2
:

Planning agents

4.

i = 0

5.

do {

6.


i++

7.


Tour
(
A
i
) =
Assign
Node
(
0,

, H, H
)

8.

}
until

(
Tour
(
A
i
) == null
)

9.

Phase 3.

O
ptimization of Agent

s local network path


10.

for each agent

A
i

11.


2OPT(
T
our
(
A
i
))

12.

Sub

AssignNode

(
L
ow,
U
p, lowernode, uppernode)

13.

{

14.


// Seq1, Seq2, Seq are sequences of nodes

15.


Seq1 = Seq2 = Seq = null

16.


For each
i
ak

where
i
ak

is

unprocessed


and
ak is most lower number {

17.


Start = Max{
Ready
(
i
ak
),
L
ow+
Ls
(lowernode,i
ak
)}

18.


End

= Start+
comp
(
i
ak
)

19.


if
End
+
L
s
(
i
ak
,
uppernode
)


upper

{

20.


mark
i
ak


processed


21.


Seq1 = AssignNode(
L
ow, Start, lowernode,
i
ak
)

22.



Seq2 = AssignNode(
End
,
U
p,
i
ak
, uppernode)

23.


Seq =
Union
(Seq1,
i
ak
, Seq2)

24.


return Seq

25.


} else return NULL

26.


}

27.

}

3.

EXPERIMENT

We perform
ed

simulations
in t
he following agent environment.
First,
nodes were scattered over

the
Internet
,

thus

the latencies
are

arbitrary. Second, the latency varied between 15ms and 200ms.

Third, degree of
the
time window:
t
he number of
nodes that are
affected by the
time constrain
ts

varied 1 to

n
, where
n

is the
maximum number of nodes exclud
ing the home node.

Table
2

show
s

that the difference of the average number of agents
between optimal and TMAP is almost less than one, and the cost
of TMAP is just 40
ms

more than that of Optimal.

Table 2. Averages of agents and cost for each configuration

Nodes

Factor
s

Opt
i mal

TMAP

13

Agent

3.0

3.9

Cost

1440.3

1878.6

14

Agent

2.4

2.8

Cost

734.6

974.4

15

Agent

3.2

4.2

Cost

1334.2

1851.2

16

Agent

3.5

4.4

Cost

1
2
17.7

1693.5

Figure
1

show
s

t he differences bet ween t he
w
orst case and TMAP
in t he num
ber of agent s and cost s required respect ively when t he
number of nodes is 300 and t he t ime window is t hree. TMAP
produces 88% reduct ion of t he numbers of agent s and 83%
reduct ion of
average
cost

compared t o t he worst case.


(a)

(b)

Figure 1. Comparison of (a) the number of agents, and (b) the
costs (# of time window=3)

4.

CONCLUSIONS

The simulated result
s

show that the proposed timed mobile agent
planning methods drastically cut down the number of mobile
agents needed
,

compared
to

the worst case in which the mobile
agent is deployed to each designated node
.

Also, these methods
significantly r
e
duce
d

the total
routing

time while preserving
the
minimum completion time. Consequently, th
is

planning method
signi
ficant improve
s

the performance of the information retrieval
system using mobile agent technology.

5.

REFERENCES

[1]

Baek, J., Kim, G., Yeo, J., and Yeom, H. Cost
-
Effective
Mobile Agent Planning for Distributed Information Retrieval,
The 21
st

Int

l Conf. On Distr
ibuted Computing Systems
(ICDCS
-
21)
, Phoenix, April 16
-
19 2001.

[2]

Brewington, B.
,

Gray,
R.,
Moizumi,
K.,

Kotz,
D.
,

Cybenko,
G.
,

and
Rus,
D. Mobile agents in distributed information
retrieval, In
Intelligent Information Agents
, p
p.
355
-
395, 1999.

[3]

Kretser,
O.
,

Moffat,
A.
,
Shimmin,
T.
,
and
Zobel,
J.

Methodologies for distributed information retrieval. In
Proc
.

of the Eighteenth Int’l Conference on Distributed Computing
Systems
,
p
p.
26
-
29,
May 1998.

[4]

Miller,
L.,
Yang,
J.,
Honavar,
V., and
Wong,
J. Intelligent
mobil
e agents for information retrieval and knowledge
discovery from distributed data and knowledge sources. In
Proc. of the IEEE Information Technology Conference
, 1998.

[5]

Moizumi, K.
Mobile Agent Planning Problems
. PhD thesis,
Dartmouth College, 1998.

[6]

Moizumi,
K.
,

and Cybenko, G. The Traveling Agent
Problem
.
Mathematics of Control, Signals and Systems
,
January 1998.

[7]

Rus,
D.,
Gray,
R., and
Kotz,
D. Autonomous and adaptive
agents that gather information. In
AAAI '96 International
Workshop on Intelligent Adaptive A
gents
, August 1996.