DESCRIPTION
INFORMATION PROCESSING SYSTEM
TECHNICAL FIELD
[001
The present invention relates to an information
processing system in which a plurality of computers
connected via a network share and execute processing of a
large-scale arithmetic operation consisting of a plurality
of jobs.
BACKGROUND ART
0002
One of the techniques for enhancing efficiency of
information processing by computers is distributed
processing. Distributed processing is a parallel
processing technique in which jobs are executed by a
plurality of machines. The techniques for sharing
processing among a plurality of machines include, in
particular, those referred to as a "multiplexing system"
and "grid computing".
0003
Among them, the multiplexing system is a technique
intended to protect data that has already been processed,
or continue service being provided, at a time of trouble,
such as machine failure. Therefore, in the case of the
multiplexing system, a plurality of machines execute the
- 1 -

same processing.
0004
On the other hand, grid computing is typically a
technique in which a plurality of computers and memory
media are connected together via a network, and virtually
handled as a large-scale, high-performance computer. For
example, by allowing a plurality of computers to share and
execute arithmetic processing that requires a large amount
of calculation, it becomes possible to substantially
quickly obtain arithmetic results.
0005
Incidentally, to share arithmetic processing among a
plurality of computers, a function of assigning jobs to
machines is necessary. Such a function is implemented by a
tool generally referred to as a "load balancer" or "load-
sharing (load-distribution) software".
0006
Next, general usage of the load balancer in grid
computing will be described. FIG. 7 is a schematic
configuration diagram of a conventional information
processing system constructed based on the grid computing
technique. As shown in FIG. 7, the information processing
system includes a client 110, a master node 120, and four
nodes 130a, 130b, 130c, and 130d. These machines 110, 120,
130a, 130b, 130c, and 130d are connected together via a
network. The client 110 is the requester of a large-scale
arithmetic operation. For example, it is assumed that the
- 2 -

client 110 requests the master node 120 to carry out a
large-scale arithmetic operation consisting of 1,000 jobs.
Here, the "job" is a unit of calculation. For example, the
job is to obtain "z" by the calculation "z=x+y". In this
case, it is assumed that n sets of variables are
substituted for x and y. Specifically, in a single job,
the number of times calculation is carried out depends on
the number of variables to be substituted. Accordingly,
although the number of jobs is 1,000, the total number of
calculations is dozens or hundreds of times greater than
that number.
0007
Here, the large-scale arithmetic operation, which the
client 110 requests the master node 120 to carry out, is
memorized in a predetermined memory device or the like.
Data for the large-scale arithmetic operation is originally
transmitted from another computer connected to the client
110 via a network, or inputted by a person in charge using
an input device of the client 110. At this time, for
example, the data for the large-scale arithmetic operation
may be transmitted or inputted after being divided into a
predetermined number (e.g., 1,000) of jobs, or transmitted
or inputted without being divided into the predetermined
number of jobs. In the latter case, the client 110 divides
the transmitted or inputted data for the large-scale
arithmetic operation into the predetermined number of jobs
in accordance with predetermined rules. Hereinafter,
- 3 -

unless otherwise specified, descriptions will be given with
respect to the case where the data for the large-scale
arithmetic operation is transmitted or inputted to the
client after being divided into 1,000 jobs.
0008
The master node 120 is a computer responsible for the
load balancing function, and performs the process of
assigning jobs to the nodes 130a, 130b, 130c, and 130d. As
for the job assignment, some techniques have been proposed,
in which a suitable number and size of jobs are transmitted
to each node in accordance with, for example, performance
and load status of the node (see, for example, patent
literature 1 and patent literature 2) . In addition, the
nodes 130a, 130b, 130c, and 130d are computers for carrying
out arithmetic processing of the jobs assigned by the
master node 120.
0009
The client 110 first receives the data for the large-
scale arithmetic operation consisting of 1,000 jobs, and
then transmits the 1,000 jobs to the master node 120. Next,
the master node 120 assigns the received 1,000 jobs to the
nodes 130a, 130b, 130c, and 130d. The nodes 130a, 130b,
130c, and 130d carry out arithmetic processing of the jobs
transmitted from the master node 120, and upon completion
of the processing, they report to the master node 120 that
the jobs have been completed. Upon receipt of such a
report from any node, if there is any unprocessed job that
- 4 -

has not yet been assigned, the master node 120 transmits
the job to that node. As such, the master node 120 repeats
both the process of transmitting any unprocessed job to
each node, and the process of receiving a report that the
job has been completed, thereby causing the four nodes 130a,
130b, 130c, and 130d to execute arithmetic processing of
all the jobs.
0010
In addition, the master node 120 simply implements
the function of efficiently assigning the jobs requested by
the client 110 to the nodes 130a, 130b, 130c, 130d, and it
does not perform any data processing on results of
arithmetic processing by the nodes 130a, 130b, 130c, and
130d. Basically, there is a premise that the results of
arithmetic processing by the nodes 130a, 130b, 130c, and
130d ultimately aggregate in the client 110, which is the
requester of the large-scale arithmetic operation.
Therefore, for example, when a result of arithmetic
processing is returned from any of the nodes 130a, 130b,
130c, and 130d, the master node 120 is required to transmit
the arithmetic processing result to the client 110.
0011
Patent Literature 1 : Japanese Unexamined Patent
Application Publication No. H07-219907
Patent Literature 2 : Japanese Unexamined Patent
Application Publication No. 2002-269062
DISCLOSURE OF INVENTION
- 5 -

TECHNICAL PROBLEM
0012
As described above, the master node performs the job
assignment process of suitably assigning and transmitting
one job after another to each node under its control, while
monitoring the processing and load status of the node.
However, as the number of nodes under control of the master
node increases, a more significant load is put on the
master node during the job assignment process, simply by
monitoring the processing and load status of each node.
Moreover, the master node performs an optimization process
for assigning suitable amounts of jobs to suitable nodes in
accordance with the processing and load status of each node,
and such an optimization process itself also creates a high
load.
0013
In addition, as the number of jobs that the client
requests the master node to process increases, the master
node more frequently performs the process of transmitting
to the client an arithmetic processing result returned from
each of the nodes. Therefore, the load on the master node
in terms of transmission of the arithmetic processing
results might increase, resulting in a reduced processing
speed of the job assignment process, which is essentially
the most important process to be performed by the master
node.
0014
- 6 -

Furthermore, even when the number of nodes is not
significantly high, as the number of requested jobs
increases, the load on the master node in terms of
selection of suitable jobs during the job assignment
process increases, and an extra load is created in terms of
maintenance of the connection with the client to meet a
request from the client to monitor the processing results.
Also, the load on the client in terms of monitoring of the
job processing results increases as the number of jobs to
be requested of the master node increases.
0015
To overcome these problems, for example, there has
been some demand to implement an information processing
system having a mechanism of controlling the load on the
master node itself.
0016
The present invention has been made in view of the
above circumstances, and an objective thereof is to provide
an information processing system capable of alleviating
excessive load on the master node, thereby allowing the
master node to efficiently perform its essential function,
i.e., the process of assigning jobs to each node.
SOLUTION TO PROBLEM
0017
The subject matter as claimed in claim 1 to achieve
the above objective is directed to an information
processing system comprising a client for requesting
- 7 -

processing of a large-scale arithmetic operation consisting
of a plurality of jobs, a plurality of nodes for executing
processing of the jobs, and a master node for assigning
processing of the jobs requested by the client to the nodes,
the client, the nodes, and the master node being connected
together via a network, wherein the client classifies the
jobs constituting the large-scale arithmetic operation,
which are memorized in a memory means, into several blocks,
and requests the master node to process the jobs block by
block, such that the master node always assigns a
predetermined number of jobs or less.
0018
The subject matter as claimed in claim 2 is directed
to the information processing system according to claim 1,
wherein the client classifies the jobs constituting the
large-scale arithmetic operation, such that a predetermined
number of jobs belong to a first block, and other jobs each
belong to an nth block, where n=2, 3, ..., N, and the
client first requests the master node to process the
predetermined number of jobs in the first block, and then
requests the master node to process a predetermined one of
the blocks that have not yet been processed upon each
receipt of a notification from the master node, indicating
that processing of one job has been completed.
0019
The subject matter as claimed in claim 3 is directed
to the information processing system according to claim 1
- 8 -

or 2, wherein the client estimates a processing load per
job included in the large-scale arithmetic operation, and
requests the master node to process the jobs in order from
the highest load to the lowest load in accordance with
estimation results.
0020
The subject matter as claimed in claim 4 to achieve
the above objective is directed to a computer-readable
recording medium having recorded thereon a program for
causing a computer to implement a function of an
information processing system according to any one of
claims 1, 2, and 3.
0021
The subject matter as claimed in claim 5 to achieve
the above objective is directed to a program for causing a
computer to implement a function of an information
processing system according to any one of claims 1, 2, and
3.
0022
The subject matter as claimed in claim 6 to achieve
the above objective is directed to an information
processing device for requesting a load balancer to process
a large-scale arithmetic operation consisting of a
plurality of jobs, the device including: a grouping process
means for grouping the jobs constituting the large-scale
arithmetic operation, which are memorized in a memory means,
based on grouping rule information memorized in the memory
- 9 -

means, thereby forming several job groups; and a management
means for requesting the load balancer to process the job
groups formed by the grouping process means, such that the
load balancer always processes a predetermined number of
job groups or less.
0023
The subject matter as claimed in claim 7 is directed
to the information processing device according to claim 6,
further comprising a load estimation means for estimating a
processing load per job group formed by the grouping
process means based on the load estimation rule information
memorized in the memory means, wherein based on a
processing load estimation result per job group, which is
obtained by the load estimation means, the management means
requests the load balancer to process the job groups in
order from the highest load to the lowest load.
0024
The subject matter as claimed in claim 8 to achieve
the above objective is directed to a program for causing a
computer to implement a function of an information
processing device according to claim 6 or 7.
ADVANTAGEOUS EFFECTS OF INVENTION
0025
In the subject matter as claimed in claim 1, the
client classifies the jobs constituting the large-scale
arithmetic operation into several blocks, and requests the
master node to process the jobs block by block, such that
- 10 -

the master node always assigns a predetermined number of
jobs or less. Thus, the client can request the master node
to process a plurality of jobs little by little in units of
blocks into which the jobs are classified such that each
block consists of small-scale jobs, rather than
simultaneously requesting the master node to carry out
processing of all the jobs as in the conventional art.
Here, the number of jobs to be included in each block is
suitably determined within the range between 1 and the
predetermined number. In addition, the predetermined
number is preferably determined in such a manner as to
allow the master node to efficiently assign the jobs to the
nodes, even if the number of nodes is significant. Such a
client conceivably has the function of controlling the load
on the master node. Thus, the master node is simply
required to always perform the process of assigning the
predetermined number of jobs or less to the nodes, and
therefore it is possible to prevent the load on the master
node from increasing. Moreover, since the load on the
master node can be reduced as such, the master node can
efficiently carry out not only the process for assigning
jobs to the nodes, but also the process for meeting the
client's demand to monitor arithmetic results. Furthermore,
the number of jobs the client requests the master node to
process is reduced, and therefore the load on the client in
terms of the process of monitoring the arithmetic results
is reduced.
- 11 -

0026
In the subject matter as claimed in claim 2, the
client classifies the jobs constituting the large-scale
arithmetic operation, such that a predetermined number of
jobs belong to the first block, and other jobs each belong
to the nth block, where n=2, 3, ..., N, and the client
first requests the master node to process the predetermined
number of jobs in the first block, and then requests the
master node to process a predetermined one of the blocks
that have not yet been processed upon each receipt of a
notification from the master node, indicating that
processing of one job has been completed. The client
requests the master node to process the jobs in such a
manner, and therefore the master node can efficiently
assign the jobs to the nodes, making it possible to prevent
the nodes from experiencing a long waiting period for
processing, and thereby to improve processing efficiency of
the nodes.
0027
In the subject matter as claimed in claim 3, the
client estimates the processing load per job included in
the large-scale arithmetic operation, and requests the
master node to process the jobs in order from the highest
load to the lowest load in accordance with estimation
results. Thus, the master node can assign processing of
the job with the highest load to a node with the highest
processing capability, thereby causing the node to first
- 12 -

execute that job, and it is possible to allow the master
node to perform the job assignment process, such that the
entire processing time can be roughly equalized among all
the nodes, thereby making it possible to shorten the entire
processing time for the large-scale arithmetic operation.
0028
In the subject matter as claimed in claim 6, the
grouping process means groups the jobs constituting the
large-scale arithmetic operation, which are memorized in
the memory means, based on the grouping rule information
memorized in the memory means, thereby forming several job
groups. For example, the grouping process means groups
1,000 jobs into 600 job groups. The management means
requests the load balancer to process the job groups formed
by the grouping process means, such that the load balancer
always processes a predetermined number of job groups or
less. For example, when the predetermined number is set at
300, the load balancer is always requested to process up to
300 job groups. Accordingly, the load balancer does not
have to always manage 1,000 jobs, and is simply required to
perform the process of assigning jobs included in 300 job
groups or less to the nodes. Therefore it is possible to
prevent the load on the load balancer from increasing.
Moreover, since the load on the load balancer can be
reduced as such, the load balancer can efficiently carry
out not only the process for assigning job groups to the
nodes, but also the process for meeting the demand to
- 13 -

monitor arithmetic results from the information processing
device. Furthermore, the number of jobs the load balancer
is requested to process is reduced, and therefore the load
on the information processing device in terms of the
process of monitoring the arithmetic results is reduced.
0029
In the subject matter as claimed in claim 7, the load
estimation means estimates the processing load per job
group formed by the grouping process means based on the
load estimation rule information memorized in the memory-
means. Based on the processing load estimation result per
job group, which is obtained by the load estimation means,
the management means requests the load balancer to process
the job groups in order from the highest load to the lowest
load. Thus, the load balancer can assign processing of the
job group with the highest load to a node with the highest
processing capability, thereby causing the node to first
execute that job group, and it is possible to allow the
load balancer to perform the job group assignment process,
such that the entire processing time can be roughly
equalized among all the nodes, thereby making it possible
to shorten the entire processing time for the large-scale
arithmetic operation.
BRIEF DESCRIPTION OF DRAWINGS
0030
FIG. 1 is a schematic configuration diagram of an
information processing system according to an embodiment of
- 14 -

the present invention;
FIG. 2 is a schematic configuration block diagram of
a client in the information processing system according to
the embodiment;
FIG. 3 is a diagram for describing exemplary jobs
constituting a large-scale arithmetic operation;
FIG. 4 is a diagram illustrating an exemplary list
generated by a list generation means of the client;
FIG. 5 is a flowchart for describing the procedure by
which the client carries out processing in accordance with
a first requesting method;
FIG. 6 is a flowchart for describing the procedure by
which the client carries out processing in accordance with
a second requesting method; and
FIG. 7 is a schematic configuration diagram of a
conventional information processing system constructed
based on the grid computing technique.
REFERENCE SIGNS OF LIST
10 client
11 input device
12 display device
13 communication portion
14 memory portion
15 control portion
15a grouping process means
15b load estimation means
15c list generation means
- 15 -

15d management means
20 master node
30a, 30b, 30c, 30d node
BEST MODE FOR CARRYING OUT THE INVENTION
0032
Hereinafter, the best mode for carrying out the
present invention related this application will be
described with reference to the drawings. FIG. 1 is a
schematic configuration diagram of an information
processing system according to an embodiment of the present
invention.
0033
For example, the information processing system
according to the present embodiment is used to share and
execute a large-scale arithmetic operation among a
plurality of computers. As shown in FIG. 1, the
information processing system includes a client 10, a
master node 20, and four nodes 30a, 30b, 30c, and 30d.
Personal computers, workstations, or the like, are used as
the client 10, the master node 20, and the nodes 30a, 30b,
30c, and 30d. In addition, the client 10, the master node
20, and the nodes 30a, 30b, 30c, and 30d are connected
together via a network. Here, for example, the internet,
or a company network, such as LAN or WAN, is used as the
network. Note that the client 10 corresponds to an
information processing device according to the subject
matter as claimed in claim 6.
- 16 -

0034
The client 10 receives data for the large-scale
arithmetic operation, and requests the master node 20 to
process the large-scale arithmetic operation. The large-
scale arithmetic operation consists of a plurality of jobs.
The "job" as used herein refers to a unit of calculation.
The data for the large-scale arithmetic operation is
typically transmitted from an external computer or the like
to the client 10 via the network, but it may be inputted by,
for example, a person in charge of the client 10
himself/herself. The data for the large-scale arithmetic
operation that is transmitted from the external computer or
the like, or inputted by the person in charge, may be
already divided or may not be divided into units of jobs
when the data is obtained by the client 10. When the
client 10 obtains the data for the large-scale arithmetic
operation that has not yet been divided into units of jobs,
the client 10 divides the data for the large-scale
arithmetic operation into a predetermined number of jobs by
means of its own job dividing function. Concrete details
of the job dividing function will be described later.
0035
In addition, the client 10 performs the process of
grouping a plurality of jobs constituting the large-scale
arithmetic operation, thereby forming job groups, each
consisting of interdependent jobs. In the present
embodiment, a request to process the large-scale arithmetic
- 17 -

operation is in fact carried out in units of such job
groups. Furthermore, the client 10 performs the process of
monitoring results of job group processing, which the
master node 20 is requested to perform. Concretely, at
regular intervals or arbitrary times as the necessity
arises, the client 10 inquires of the master node 20
whether the job group processing which the master node 20
has been requested to perform has been completed. In
addition, upon receipt of processing results for all the
job groups, the client 10 performs the process of
consolidating these results. Note that the client 10
having such a function can be implemented by, for example,
incorporating software for implementing that function into
a workstation.
0036
Incidentally, although the client 10 performs the
process of consolidating the processing results for the job
groups upon receipt of them, it does not mean that the
client 10 always carries out some information processing.
In addition, in some cases, an external machine, rather
than the client 10, might perform the process of
consolidating the processing results for the job groups.
Specifically, for example, when 1,000 results are obtained
for 1,000 jobs, if the 1,000 results themselves are
transmitted to the external machine, the client 10 does not
have to perform the consolidation process.
0037
- 18 -

The master node 20 is a computer responsible for the
load balancing function, and performs the process of
assigning the four nodes 30a, 30b, 30c, and 30d processing
of the job groups requested by the client 10. Here, for
example, the nodes 30a, 30b, 30c, and 30d are each assigned
one job group at a time. The technique as described in the
"BACKGROUND ART" section is applicable to the assignment
process itself. After determining which job group is to be
assigned to which node, the client 10 transmits the details
of the job group, i.e., calculation details for jobs
included in that job group, and variables for use in the
calculation, to the node to which the job group has been
assigned. Alternatively, the calculation details for jobs,
and the variables for use in the calculation may be stored
in a predetermined memory means (not shown), and the client
10 may transmit job group identification information to the
node to which the job group has been assigned, along with
information indicating locations of both the calculation
details for jobs included in the job group, and the
variables for use in the calculation. In such a case, the
nodes 30a, 30b, 30c, and 30d each access the memory means
to obtain the calculation details for jobs included in the
job group assigned thereto, and the variables for use in
the calculation.
0038
Also, in response to the inquiry from the client 10,
for example, the master node 20 transmits a notification to
- 19 -

the client 10, indicating that processing of the job group
in each of the nodes 30a, 30b, 30c, and 30d has been
completed. Here, the master node 20 may transmit such a
processing completion notification to the client 10 not
only upon inquiry from the client 10 but also upon receipt
of a report from any of the nodes 30a, 30b, 30c, and 30d,
indicating that job group processing has been completed.
Furthermore, upon receipt of a job group processing result
from any of the nodes 30a, 30b, 30c, and 30d, the master
node 20 transmits that processing result to the client 10.
Note that the master node 20 having such a function can be
implemented by, for example, incorporating software for
implementing that function into a workstation.
0039
The nodes 30a, 30b, 30c, and 30d each execute
processing of a job group assigned thereto in response to
an instruction from the master node 20. The instruction
from the master node 20 includes, for example, the
processing details of the job group, and information
regarding where to store (or transmit) a processing result.
Here, in the present embodiment, the master node 20 is
designated as the storage (or transmission) destination of
a processing result. Upon completion of the job group
arithmetic processing, the nodes 30a, 30b, 30c, and 30d
each report to the master node 20 that the processing of
the job group has been completed, and transmit the
processing result to the master node 20. Thereafter, the
- 20 -

nodes 30a, 30b, 30c, and 30d each wait for another
instruction to be transmitted. As such, the instruction
from the master node 20 to each of the nodes 30a, 30b, 30c,
and 30d, and the report from each of the nodes 30a, 30b,
30C, and 30d to the master node 20 are repeatedly carried
out to execute the large-scale arithmetic operation. In
this case, the master node 20 also effects its load
balancing function to perform such a control as to shorten
a waiting period between the report from each of the nodes
30a, 30b, 30c, and 30d that the processing of the job group
has been completed, and the receipt of another instruction.
0040
Next, the configuration of the client 10 will be
described in detail. FIG. 2 is a schematic configuration
block diagram of the client 10 in the information
processing system according to the present embodiment.
0041
As shown in FIG. 2, the client 10 includes an input
device 11, a display device 12, a communication portion 13,
a memory portion 14, and a control portion 15. For example,
the input device 11 inputs various instructions and data
for a large-scale arithmetic operation. A keyboard, a
mouse, or the like, can be used as the input device 11. In
addition, for example, the display device 12 displays an
input screen for inputting data for the large-scale
arithmetic operation, and it also displays processing
results for the large-scale arithmetic operation.
- 21 -

0042
The communication portion 13 transmits information to
the master node 20 and other machines (computers, databases,
etc.), and also receives information from the master node
20 and other machines. Concretely, the information to be
transmitted from other machines to the client 10 includes,
for example, data for a large-scale arithmetic operation.
In addition, the information to be transmitted from the
client 10 to the master node 20 includes, for example, a
processing instruction to request processing of job groups,
and an inquiry for monitoring processing results. The
information to be transmitted from the master node 20 to
the client 10 includes, for example, processing results per
job group.
0043
The memory portion 14 stores, for example, various
application programs and data. For example, a memory
device or hard disk drive in the client 10, or an external
memory device can be used as the memory portion 14. The
memory portion 14 stores data for a large-scale arithmetic
operation, grouping rule information, load estimation rule
information, a job group list, the maximum number of job
groups that can be assigned by the master node 20,
processing results per job group, and so on. The grouping
rule information, the load estimation rule information, the
job group list, and the maximum number of job groups that
can be assigned by the master node 20 will be described
- 22 -

later.
0044
Here, the large-scale arithmetic operation will be
concretely described. Considering now the case where data
for a large-scale arithmetic operation is transmitted from
an external machine to the client 10 via the network, when
the communication portion 13 of the client 10 receives the
data for a large-scale arithmetic operation, the data is
stored to the memory portion 14. FIG. 3 is a diagram for
describing exemplary jobs constituting the large-scale
arithmetic operation. In this example, the large-scale
arithmetic operation consists of 1,000 jobs. Among them,
job "1" is the calculation "zi-x+y", and, for example, it
is assumed that there are ten sets of variables (x,y) for
that calculation. Job "2" is the calculation "z=zixxi".
That is, the variables for job "2" are the processing
result zi for job 1, and the variable xx. When the variable
Xi takes two values, there are twenty sets of variables
(zlrXi) . In addition, job "3" is the calculation
"z=x+yi+y2" For this calculation, for example, it is
assumed that there are twenty sets of variables {x,ylry2) .
For other jobs also, their respective calculation details
and variables are set. The large-scale arithmetic
operation consists of such a collection of jobs. Note that
in the exemplary large-scale arithmetic operation shown in
FIG. 3, although a single job includes a single calculation
detail (arithmetic equation), in general, a single job
- 23 -

might include a plurality of calculation details. Also, in
the exemplary large-scale arithmetic operation shown in FIG.
3, although the jobs differ from each other in the
calculation detail (arithmetic equation), there may be a
plurality of jobs that have the same calculation detail but
different variables, among a plurality of jobs constituting
a large-scale arithmetic operation.
0045
Incidentally, in some cases, the client 10 might
receive the data for the large-scale arithmetic operation,
which has not yet been divided into units of jobs, from an
external machine, as described above. Specifically, all
calculation details and variables for use therein, which
are both included in the data for the large-scale
arithmetic operation, might be transmitted in an undivided
state from the external machine to the client 10, for
example, without being divided into 1,000 jobs in total
from job "1" to job "1,000", as shown in FIG. 3. In such a
case, the client 10 effects its job dividing function
provided thereto to divide the transmitted data for the
large-scale arithmetic operation on a calculation-by-
calculation basis, and combine each calculation obtained by
division with variables for use therein, thereby obtaining
data for the large-scale arithmetic operation that has been
divided into units of jobs. In addition, the data for a
large-scale arithmetic operation might not be directly
transmitted from the external machine to the client 10, but
- 24 -

stored in an external database. In such a case, the client
10 receives in advance from the external machine only
location information about the external database, and data
for a division criterion by which to determine how to
divide the data for a large-scale arithmetic operation.
Thereafter, the client 10 effects the job dividing function
to access the external database, and divide the data for a
large-scale arithmetic operation into a plurality of jobs
in accordance with the division criterion, thereby
obtaining data for the large-scale arithmetic operation
that has been divided into units of jobs.
0046
The control portion 15 performs overall control of
each portion of the client 10. As shown in FIG. 2, the
control portion 15 includes a grouping process means 15a, a
load estimation means 15b, a list generation means 15c, and
a management means 15d.
0047
The grouping process means 15a performs a grouping
process for grouping a plurality of jobs constituting the
large-scale arithmetic operation, thereby forming job
groups, each consisting of interdependent jobs. When one
job is related to or depends on another job, in some cases,
it might be inappropriate to process these jobs
individually. For example, when the processing result for
job "1" is used to perform the calculation for job "2" as
shown in FIG. 3, if the master node 20 is requested to
- 25 -

process job "2" before being requested to process job "1",
job "2" cannot be processed. Therefore, the grouping
process means 15a reads grouping rule information stored in
the memory portion 14, and performs the grouping process to
group a plurality of jobs constituting the large-scale
arithmetic operation in accordance with the grouping rule
information that has been read, thereby forming several job
groups. The grouping rule information has been previously
stored in the memory portion 14 for each large-scale
arithmetic operation to be performed. Also, in some cases,
the grouping rule information might be transmitted from the
external machine, along with the data for the large-scale
arithmetic operation, and then stored to the memory portion
14. For example, in the case of grouping rule information
regarding a large-scale arithmetic operation for
calculating trajectories of objects, it is possible to set
a rule of forming job groups, such that each group consists
of jobs having calculation targets in common in terms of
scientific and physical characteristics of the objects.
Also, in the case of grouping rule information regarding a
large-scale arithmetic operation for making production
schedules for products at a factory or the like, it is
possible to set a rule of forming job groups, such that
each group consists of jobs having calculation targets in
common in terms of product type and size. In addition, in
the case of the grouping rule information regarding the
large-scale arithmetic operation shown in FIG. 3, it is
- 26 -

possible to set a rule of forming job groups, such that
each group consists of jobs that are correlated in terms of
use of variables. When this rule information is applied to
the example of FIG. 3, the grouping process means 15a
analyzes the correlation between jobs in terms of use of
variables, and forms one job group by grouping jobs "1" and
"2". Furthermore, the grouping rule information may define
a rule regarding the maximum number of jobs that can be
included in one job group, such that each group is formed
by grouping a plurality of jobs, the number of which is
determined so as not to exceed the maximum number. In
addition, it is possible to set a rule by combining a
plurality of rules as described above. For example, in the
case of the grouping rule information regarding the large-
scale arithmetic operation for making production schedules
for products at a factory or the like, it is possible to
set a rule of forming job groups, such that each group
consists of jobs having calculation targets in common in
terms of product type and size, and the number of jobs
included therein does not exceed a predetermined maximum
number. Concretely, in the present embodiment, the
grouping process means 15a performs the grouping process
for 1,000 jobs as shown in FIG. 3, and obtains 600 job
groups as shown in FIG. 1.
0048
For each job group obtained by the grouping process
means 15a, the load estimation means 15b estimates a
- 27 -

processing load in accordance with load estimation rule
information stored in the memory portion 14. For example,
the load estimation rule information defines a rule of
estimating a processing load per job group, such that the
processing load is estimated based on the sum of sets of
variables for jobs included in the job group. The sum of
sets of variables is the total number of times calculation
is carried out for the jobs included in the job group
during arithmetic processing of the job group. Accordingly,
if each job has roughly the same level of calculation
detail, it is conceivable that the processing load for the
job group increases as the sum of sets of variables
increases. Note that if the jobs differ from each other in
the calculation detail significantly, the loads for the
jobs are different accordingly. In such a case, to
correctly estimate the load, the details of the load
estimation rule information may be changed considering the
details of the jobs, as well as the number of sets of
variables. To carry out such a load estimation considering
the details of the jobs, for example, it is necessary to
externally provide parameters indicating the level of
calculation detail for the job, or store previously-used
parameters to the memory portion 14, which are read again
at a later use.
0049
Note that as a result of the estimation by the load
estimation means 15b, if any job group with an extremely
- 28 -

high load has been found, the load estimation means 15b may
return the job group to the grouping process means 15a,
thereby causing the grouping process means 15a to perform
the process of further dividing the job group into smaller
groups. In such a case, the grouping rule information is
required to contain load threshold information by which to
determine whether to further divide the job group into
smaller groups, as well as a rule for that further division.
It is not possible to classify, in particular,
interdependent jobs into different groups, and therefore
the grouping rule information is required to contain a rule
for such indivisible jobs. In general, the rule contained
in the grouping rule information can be defined in various
formats, such as those of the "if-then" rule and the range
specification rule.
0050
The list generation means 15c generates a job group
list based on the processing load per job group estimated
by the load estimation means 15b, such that the job groups
are sorted in descending order of load, and the generated
list is stored to the memory portion 14. FIG. 4
illustrates an example of the list generated by the list
generation means 15c. This list includes "job group ID"
fields, and "target job" fields. Each "job group ID" field
indicates an ID number (identification information)
assigned to the job group, and each "target job" field
indicates the job number of each job included in the job
- 29 -

group. The list in FIG. 4 is directed to 600 job groups
obtained from the 1,000 jobs shown in FIG. 3. The job
group with the highest load is indicated at the top of the
list, and the loads put on by the job groups are lower as
the list goes down. In this example, the job group with ID
No. "001", which consists only of job "1000", exhibits the
highest load, and the job group with ID No. "002", which
consists of jobs "5", "6", and "7", exhibits the second
highest load. As such, in the example of FIG. 4, the list
generation means 15c assigns ID numbers to the 600 job
groups in descending order of load. Therefore, the job
group ID numbers play not only the role of identifying the
job groups, but also the role of indicating the descending
order of load. The list also includes a "request
completed" field and a "receipt completed" field for each
job group. Each "request completed" field is intended to
be assigned a "request completed" flag, which indicates
that the master node 20 has been requested to process the
job group. Each "receipt completed" field is intended to
be assigned a "receipt completed" flag, which indicates
receipt of a notification that processing of the job group
has been completed. In the present embodiment, concretely,
the "request completed" field and the "receipt completed"
field are each assigned the flag "0" when the processing
has not been "completed", and the flag "1" when the
processing has been "completed". Note that the "request
completed" flag and the "receipt completed" flag are
- 30 -

managed by the management means 15d.
0051
The management means 15d requests the master node 20
to process each job group included in the large-scale
arithmetic operation. Concretely, the management means 15d
classifies a plurality of job groups constituting the
large-scale arithmetic operation into several blocks, and
requests the master node 20 to process the job groups block
by block, such that the master node 20 always assigns a
predetermined number of job groups or less. Here, the
number of job groups to be included in each block is
suitably determined within the range between 1 and the
predetermined number. Specifically, if the predetermined
number (i.e., the maximum number of job groups to be
assigned by the master node 20) is M, a single block
contains m job groups, where in is a natural number from 1
to M, inclusively. In addition, the maximum number M of
job groups to be assigned by the master node 20 is
preferably determined in such a manner as to allow the
master node 20 to efficiently assign the job groups to the
nodes, even if the number of nodes is significant.
Specifically, the maximum number M of job groups to be
assigned by the master node 20 is determined in accordance
with the processing capability of the master node 20. In
the present embodiment, the maximum number M of job groups
to be assigned by the master node 20 is set at 300, for
example. Here, the maximum number M of job groups to be
- 31 -

assigned by the master node 20 is stored in the memory
portion 14.
0052
In addition, the management means 15d preferably
requests the master node 20 to process the job groups in
order from the highest load to the lowest load in
accordance with the list shown in FIG. 4. If the master
node 20 is requested at the last to process the job group
with the highest load, the job group is processed by one
node over a long period of time, resulting in an extremely
long processing time for the entire large-scale arithmetic
operation. By requesting the master node 20 to process the
job groups in descending order of load, the master node 20
can assign processing of the job group with the highest
load to a node with the highest processing capability,
thereby causing the node to first execute that job group,
and it becomes possible to allow the master node 20 to
perform the job group assignment process, such that the
entire processing time can be roughly equalized among all
the nodes, thereby making it possible to shorten the entire
processing time for the large-scale arithmetic operation.
0053
Incidentally, there are various possible methods for
the management means 15d to request the master node 20 to
process the job groups. In the present embodiment, the
following two requesting methods will be described as
typical examples.
- 32 -

0054
In the first requesting method, of the 600 job groups
constituting the large-scale arithmetic operation as shown
in FIG. 4, 300 job groups listed in the top three hundred
in FIG. 4 are classified into the first block, and other
job groups are each classified as the nth block (where n=2,
3, ..., N) in order from the highest position on the list
of FIG. 4. Here, N=301. Therefore, the 300 job groups
from ID No. "001" to ID No. "300" belong to the first block,
the job group with ID No. "301" belongs to the second block,
the job group with ID No. "302" belongs to the third
block, ..., and the job group with ID No. "600" belongs to
the 301st block. The management means 15d first requests
the master node 20 to process the 300 job groups in the
first block, and thereafter upon each receipt of a
notification from the master node 20, indicating that
processing of one job group has been completed, the
management means 15d requests the master node 20 to process
a predetermined one of the blocks for which processing has
not yet been requested. Here, the request to process the
one predetermined block is made for the block listed in the
highest position in FIG. 4, excluding any block for which
the processing request has already been made at that time.
For example, in the case where the master node 20 has
already been requested so as to process a total of 100
blocks, from the first to the 100th block, when the
management means 15d receives a notification that
- 33 -

processing of one job group has been completed, the master
node 20 is requested to process the 101st block.
0055
In addition, in the second requesting method, of the
600 job groups constituting the large-scale arithmetic
operation, 300 job groups listed in the top three hundred
in FIG. 4 are classified into the first block, and the next
300 job groups on the list, i.e., the 301st to 600th job
groups, are classified into the second block. The
management means 15d first requests the master node 20 to
process the 300 job groups in the first block, and then the
300 job groups in the second block after processing of all
the requested 300 job groups in the first block has been
completed.
0056
With the first requesting method, the master node 20
can efficiently assign the job groups to the nodes 30a, 30b,
30c, and 30d, making it possible to prevent the nodes 30a,
30b, 30c, and 30d from experiencing a long waiting period
for processing, and thereby to improve processing
efficiency of the nodes 30a, 30b, 30c, and 30d. On the
other hand, with the second requesting method also, it is
possible to improve the processing efficiency of the nodes
30a, 30b, 30c, and 30d. However, in the case of the second
requesting method, after the processing of the first block
has been completed, the nodes 30a, 30b, 30c, and 30d might
experience a slight waiting period for processing before
- 34 -

the processing of the second block is started. Therefore,
in consideration of this, it is conceivable that the first
requesting method is superior to the second requesting
method in the processing efficiency of each node.
0057
As such, in the present embodiment, the management
means 15d classifies the 600 job groups constituting the
large-scale arithmetic operation into several blocks, and
requests the master node 20 to process the job groups block
by block, such that the master node 20 can always perform
the assignment process for 300 job groups or less.
Therefore, hypothetically, the master node 20 performs the
process of assigning the nodes up to 300 job groups, the
number of which is relatively small as compared to that in
the conventional art, and therefore it is possible to
prevent the load on the master node 20 from increasing.
Specifically, in the present embodiment, the client 10 (the
management means 15d) functions to control the load on the
master node 20, so that the master node 20 and the client
10 are as a whole free from the state of high load.
0058.
In addition, the management means 15d uses the list
generated by the list generation means 15c to manage the
processing request status per job group. Specifically,
after requesting the master node 20 to process a job group,
the management means 15d sets the "request completed" flag
"1" in the "request completed" field for that job group on
- 35 -

the list of FIG. 4. In addition, at regular intervals or
arbitrary times as the necessity arises, the management
means 15d inquires of the master node 20 whether any job
group, which the master node 20 has been requested to
process, has been completely processed, and upon receipt of
a notification from the master node 20, indicating that the
job group has been completely processed, the management
means 15d sets the "receipt completed" flag "1" in the
"receipt completed" field for that job group on the list of
FIG. 4. Thus, for each job group, the management means 15d
can judge whether the "request completed" flag "1" is set
on the list, thereby determining whether the master node 20
has been requested to process the job group, while the
management means 15d can judge whether the "receipt
completed" flag "1" is set on the list, thereby determining
whether processing of the job group has been completed.
0059
Described next is the procedure of processing the
large-scale arithmetic operation in the information
processing system according to the present embodiment.
0060
The procedure of processing the large-scale
arithmetic operation in the information processing system
according to the present embodiment will be first described
with respect to the case where the client 10 carries out
processing in accordance with the first requesting method.
FIG. 5 is a flowchart for describing the procedure by which
- 36 -

the client 10 carries out the processing in accordance with
the first requesting method.
0061
First, for example, data for a large-scale arithmetic
operation is transmitted to the client 10 via the network.
Here, it is assumed that the large-scale arithmetic
operation consists of 1,000 jobs as shown in FIG. 3. Upon
receipt of the data for the large-scale arithmetic
operation, the client 10 performs a grouping process on the
1,000 jobs constituting the large-scale arithmetic
operation, such that interdependent jobs are classified
into one job group (Sll). Here, it is assumed that 600 job
groups are obtained through the grouping process performed
by the client 10 as shown in FIG. 1.
0062
Next, the client 10 estimates the processing load per
job group (S12). For example, the load estimation is
carried out based on the sum of sets of variables for all
the jobs included in the job group. In this case, the
client 10 determines the job group to exhibit a higher
processing load as the sum of sets of variables increases.
0063
Next, the client 10 generates a job group list, in
which the 600 job groups are sorted in descending order of
load, based on the processing load per job group obtained
by the processing in step S12 (S13). Here, it is assumed
that the list as shown in FIG. 4 is generated.
- 37 -

0064
Next, the client 10 classifies the 600 job groups
into 301 blocks. Concretely, the top 300 job groups on the
list of FIG. 4, i.e., the job groups from ID No. "001" to
ID No. "300", are classified into the first block, and
other job groups are sequentially classified into the
second block, the third block, ..., and the 301st block, in
descending order of load on the list of FIG. 4. Here, the
first block contains 300 job groups, and the nth block (n=2,
3, --., 301) contains only one job group. In this manner,
a total of 301 blocks are obtained. Thereafter, the client
10 first requests the master node 20 to process the 300 job
groups in the first block (S14). After requesting the
master node 20 to process the job groups, the client 10
sets the "request completed" flag "1" in the "request
completed" field for each of the job groups on the list.
0065
Upon request by the client 10 to process the 300 job
groups, the master node 20 assigns processing of these job
groups to the nodes 30a, 30b, 30c, and 30d. The nodes 30a,
30b, 30c, and 30d each execute processing of the job group
assigned thereto. Upon completion of the job group
processing, the nodes 30a, 30b, 30c, and 30d each transmit
a report thereof and a processing result to the master node
20.
0066
On the other hand, after requesting the master node
- 38 -

20 to process the job groups, the client 10 inquires of the
master node 20 whether any job group, which the master node
20 has been requested to process, has been completely
processed, at regular intervals or arbitrary times as the
necessity arises. Upon receipt of the inquiry, if there is
any job group that has already been processed, the master
node 20 transmits to the client 10 a notification
indicating that the job group has already been processed,
along with the processing result. Thereafter, upon receipt
of the notification indicating that the job group has
already been processed, the client 10 sets the "receipt
completed" flag "1" in the "receipt completed" field for
that job group on the list (S15).
0067
After the processing in step S15, the client 10
determines whether there is any job group for which the
"request completed" flag "1" has not yet been set on the
list (S16). If there is any job group for which the
"request completed" flag "1" has not yet been set, the
client 10 requests the master node 20 to process the job
group listed at the highest position in FIG. 4, excluding
the job groups for which processing has already been
requested so far (S17). In addition, the client 10 sets
the "request completed" flag "1" in the "request completed"
field for that job group on the list. Thereafter, the
procedure advances to step S15. The processing in steps
S15, S16, and S17 will be repeated as long as the client 10
- 39 -

determines in step S16 that there is any job group for
which the "request completed" flag "1" has not yet been set.
Here, so long as there is any job group for which the
"request completed" flag "1" has not yet been set, the
number of job groups the master node 20 has been requested
to process is always 300. Specifically, so long as there
is any job group for which the "request completed" flag "1"
has not yet been set, the master node 20 always assigns
processing of 300 job groups to the nodes 30a, 30b, 30c,
and 30d.
0068
Upon receipt of a notification that a job group has
been processed after the processing in steps S15, S16, and
S17 is repeated 300 times (S15), the client 10 determines
that there is no job group for which the "request
completed" flag "1" has not yet been set on the list (S16).
That is, at this time, the master node 20 has been
requested to process all of the 600 job groups. Next, the
client 10 determines whether there is any job group for
which the "receipt completed" flag "1" has not yet been set
on the list (S18). If there is any job group for which the
"receipt completed" flag "1" has not yet been set, the
procedure advances to step S19. In step S19, upon receipt
of a notification that a job group has been processed, the
client 10 sets the "receipt completed" flag "1" in the
"receipt completion" field for that job group on the list.
Thereafter, the procedure returns to step S18. The
- 40 -

processing in steps S18 and S19 will be repeated as long as
the client 10 determines in step S18 that there is any job
group for which the "receipt completed" flag "1" has not
yet been set. At this time, upon each processing in step
S19, the number of job groups the master node 20 has been
requested to process is decremented by 1. Accordingly,
after the "request completed" flag "1" has been set for all
the job groups on the list, the master node 20 always
assigns processing of 300 job groups or less to the nodes
30a, 30b, 30c, and 30d.
0069
After the processing in steps S18 and step S19 is
repeated 300 times, the client 10 determines that there is
no job group for which the "receipt completed" flag "1" has
not yet been set on the list (S18). Accordingly, the
client 10 confirms that the large-scale arithmetic
operation has been completely processed, and consolidates
the processing results (S20). Thus, the flow of processing
by the client 10 as shown in FIG. 5 ends.
0070
Next, the procedure of processing the large-scale
arithmetic operation in the information processing system
according to the present embodiment will be described with
respect to the case where the client 10 carries out the
processing in accordance with the second requesting method.
FIG. 6 is a flowchart for describing the procedure by which
the client 10 carries out the processing in accordance with
- 41 -

the second requesting method.
0071
Processing details of steps S51, S52, and S53 in the
process flow of FIG. 6 are the same as those of steps Sll,
S12, and S13, respectively, in the process flow of FIG. 5,
and therefore descriptions thereof will be omitted herein.
0072
After the processing in step S53, the client 10
classifies the 600 job groups into two blocks (S54).
Concretely, the top 300 job groups on the list of FIG. 4
are classified into the first block, and the next 300 job
groups on the list, i.e., the 301st to 600th job groups,
are classified into the second block.
0073
Next, the client 10 requests the master node 20 to
process one block (S55). Here, the master node is first
requested to process the first block, and then the second
block. Accordingly, in this case, the master node 20 is
first requested to process the 300 job groups in the first
block. In addition, when the master node 20 is requested
to process the job groups, the client 10 sets the "request
completed" flag "1" in the "request completed" field for
each of the job groups on the list.
0074
Thereafter, at regular intervals or arbitrary times
as the necessity arises, the client 10 inquires of the
master node 20 whether any job group, which the master node
- 42 -

20 has been requested to process, has been completely
processed. Upon receipt of a notification from the master
node 20, which responds to the inquiry, indicating that a
job group has already been processed, the client 10 sets
the "receipt completed" flag "1" in the "receipt completed"
field for that job group on the list (S56).
0075
Next, the client 10 determines whether there is any
job group for which the "request completed" flag "1" has
already been set on the list but the "receipt completed"
flag "1" has not yet been set (S57). If there is any job
group for which the "receipt completed" flag "1" has not
yet been set, the procedure returns to step S56. The
processing in steps S56 and S57 is repeated unless
processing of all the currently requested 300 job groups is
determined to have been completed.
0076
After the processing in steps S56 and S57 is repeated
299 times, when the client 10 receives a notification that
the job group has been completely processed (S56), the
client 10 determines that the "receipt completed" flag "1"
has been set for all the job groups for which the "request
completed" flag "1" has been set on the list (S57). That
is, at this time, all the 300 job groups in the first block
have been completely processed. Next, the client 10
determines whether there is any block the master node 20
has not yet been requested to process, based on the
- 43 -

"request completed" flag "1" on the list (S58). In this
case, the master node 20 has not yet been requested to
process the job groups included in the second block, and
the "request completed" flag "1" has not yet been set for
these job groups. Therefore, the determination in step S58
is affirmative. When the determination in step S58 is
affirmative, the procedure returns to step S55. In step
S55, the master node 20 is requested to process the 300 job
groups in the second block.
0077
After the 300 job groups in the second block have
been completely processed (S57), so that there is no block
the master node 20 has not yet been requested to process
(S58), the client 10 confirms that processing of the large-
scale arithmetic operation has been completed, and
consolidates the processing results (S59). Thus, the flow
of processing by the client 10 as shown in FIG. 6 ends.
0078
In the information processing system according to the
present embodiment, the client forms several job groups by
grouping a plurality of jobs constituting the large-scale
arithmetic operation, and thereafter the client further
classifies the job groups into blocks, and requests the
master node to process the job groups block by block, such
that the master node always assigns a predetermined number
of job groups or less. Thus, the client can gradually
request the master node to process a plurality of job
- 44 -

groups block by block, rather than simultaneously
requesting the master node to process all jobs as in the
conventional art. Such a client conceivably has the
function of controlling the load on the master node.
Accordingly, the master node is simply required to always
perform the process of assigning the nodes only the jobs
included in a predetermined number of job groups as
described above, and therefore it is possible to prevent
the load on the master node from increasing. Moreover, the
load on the master node can be reduced in such a manner,
and therefore the master node can efficiently perform not
only the process for assigning the job groups to the nodes,
but also the process for meeting the client's demand to
monitor arithmetic results. Furthermore, the number of
jobs the client requests the master node to process is
reduced, and therefore the load on the client in terms of
the process of monitoring the arithmetic results is reduced.
0079
Note that the present invention is not limited by the
above embodiment, and various modifications can be made
within the scope of the invention.
0080
The above embodiment has been described with respect
to the case where the master node instructs the nodes to
store the processing results to the master node, so that
each node transmits the processing results for the job
groups to the master node, which in turn transmits the
- 45 -

processing results to the client. However, the master node
may instruct the nodes to store the processing results to,
for example, a predetermined database. In such a case,
every time each node completes arithmetic processing of a
job group, the node reports to the master node that
processing of the job group has been completed, and stores
the processing result to the instructed database. Upon
receipt of a notification from the master node, indicating
processing of the job group has been completed, the client
accesses the database to obtain the processing result for
that job group.
0081
In addition, the above embodiment has been described
with respect to the case where the client estimates a
processing load per job group, and requests the master node
to process the job groups in order from the highest load to
the lowest load, in accordance with estimation results.
However, the client does not have to estimate the
processing load per job group, and request the master node
to process the job groups in descending order of load.
That is, the client may request the master node to process
the job groups in an arbitrary order.
0082
Furthermore, the above embodiment has been described
with respect to the case where the client carries out the
grouping process for forming groups, each consisting of
interdependent jobs. However, for example, when using jobs
- 46 -

that are not interdependent, the client does not have to
perform the grouping process on the jobs. In addition, a
basic unit of calculation that is independent from
arithmetic results for other basic units may be defined as
the "job". In such cases, the job and the job group can be
naturally considered to be identical. Note that when using
jobs that are not interdependent, information regarding the
maximum number of jobs that can be included in a single job
group may be used as the grouping rule information, so that
the grouping process means can classify a plurality of jobs
into several job groups, based on the information regarding
the maximum number.
0083
In addition, the above embodiment has been described
with respect to the case where a single client requests a
single master node to process jobs (or job groups).
However, for example, it is conceivable that a plurality of
clients simultaneously request a single master node to
process jobs (or job groups). In such a case, the clients
may preferably be provided with the function of achieving
cooperation between the clients, along with the function
characteristic of the present invention, i.e., the function
of making a request to process jobs (or job groups) block
by block. As a result, the clients cooperate with each
other to make adjustments, such that the total number of
jobs (or job groups) the master node is requested to
process is always within a predetermined range. As a
- 47 -

method that achieves such a cooperative function, it is
possible to use a method in which the clients communicate
with each other to mutually notify themselves of
information regarding the number of jobs (or job groups)
the master node has been currently requested to process.
In addition, another method can be used, in which the
master node, or another machine that is neither a client
nor a master node, is provided with a job number management
means for managing the number of jobs (or job groups) the
master node has been currently requested to process.
Concretely, the job number management means has information
regarding the maximum number of jobs (or job groups) the
master node can accept a request to process. When
requesting the master node to process jobs (or job groups),
each client transmits, to the job number management means,
information regarding the number of jobs (or job groups)
for which a request is to be made. When the information
regarding the number of jobs (or job groups) is sent from
each client, and the job number management means accepts
the request to process the jobs (or job groups), the job
number management means determines whether the total number
of jobs (or job groups) the master node has been requested
to process exceeds the maximum number. If the total number
is determined not to exceed the maximum number, the job
number management means transmits to the client an
instruction to accept the request to process the jobs (or
job groups). On the other hand, if the total number is
- 48 -

determined to exceed the maximum number, the job number
management means transmits to the client an instruction to
hold or cancel the request to process the jobs (or job
groups). Thus, even if a plurality of clients request a
single master node to process jobs, it is possible to
reduce the processing load on the master node.
0084
Conversely, the present invention is applicable to
the case where a single client requests a plurality of
master nodes to process jobs (or job groups). Such a job
(or job group) request form is suitable when the number of
jobs (or job groups) requested by the client is significant,
or when the jobs (or job groups) exhibit an extremely high
load. The procedure for processing in this case is roughly
the same as in the above embodiment. However, there is a
possibility where a plurality of master nodes might not
have a standardized load balancing function, e.g., they
might have their respective load balancing functions
implemented by different vendor products. In such a case,
conceivably, it is necessary to make adjustments of, for
example, the process by which to pass jobs (or job groups)
from the client to the master nodes, and the process by
which to consolidate job processing results obtained by
nodes. For example, calculation details of jobs, variables
therefor, and the job processing results obtained by the
nodes may be passed between the client, the master nodes
and the nodes via an external common database, rather than
- 49 -

being directly transmitted to/received from the client, the
master nodes, or the nodes, so that the client can
eventually collect the processing results from the external
database and carry out the consolidating process. Note
that as can be appreciated from the foregoing, the present
invention is also applicable to the case where a plurality
of clients request a plurality of master nodes to process
jobs (or job groups).
0085
Furthermore, the above embodiment has been described
with respect to the case where the client classifies a
plurality of jobs (or job groups) constituting the large-
scale arithmetic operation into several blocks, and
requests the master node to process the jobs (or job
groups) block by block, such that the master node can
always perform the process of assigning a predetermined
number of jobs (or job groups) or less. However, for
example, the client may simultaneously request the master
node to process all jobs (or job groups) as in the
conventional art, so that the master node classifies the
jobs (or job groups) into several blocks, and carries out
the process of assigning the jobs (or job groups) block by
block. That is, the function characteristic of the present
invention may be provided to the master node, rather than
to the client. As a result, for example, even if the
client requests the master node to process 1,000 jobs, it
is possible to always limit the number of jobs that are to
- 50 -

be assigned to the nodes within a predetermined range (e.g.,
300 jobs or less), thereby making it possible to
satisfactorily reduce the processing load on the master
node. Specifically, in the case of the conventional
information processing systems, for example, when the
master node is requested to process 1,000 jobs, the master
node is always required to manage all the 1,000 jobs as
targets of the load balancing function. However, the
information processing system according to the present
invention makes it possible to keep the number of jobs the
master node is always required to manage, for example,
within 300. In this case, however, the client
simultaneously requests the master node to process all the
jobs as in the conventional art, and therefore the client
might experience an unnecessary processing waiting period,
resulting in a slight reduction in processing efficiency in
the client as compared to the processing efficiency
attained in the above embodiment. However, the processing
load on the entire information processing system can be
reduced.
0086
In addition, the above embodiment has been described
with respect to the case where the client and the master
node are each implemented by a workstation or the like.
However, for example, the client and the master node may be
implemented by one workstation or the like. Specifically,
software for implementing the functions of both the client
- 51 -

and the master node may be incorporated into one
workstation, or the workstation may additionally have the
function of a node. Note that the above embodiment has
been described with respect to the case where a workstation
consistently serves as the client or the master node, and a
workstation has the functions of both the client and the
master node, but the present invention is not limited
thereto. Specifically, by installing into a plurality of
workstations such software as to allow the workstations to
function as the client, the master node, and the node, it
becomes possible to flexibly operate each workstation as
the client on one occasion, or as the master node or the
node on another occasion.
0087
It is understood that the objective of the present
invention is also achieved by providing a device according
to the present embodiment with program codes (including
executable codes) of software for implementing functions of
the device according to the above embodiment via a
recording medium having the program codes entirely or
partially recorded thereon, in which the device includes a
computer (or a CPU or MPU) that reads the program codes
stored in the recording medium to execute all or part of
the operation. In this case, the program codes read from
the recording medium by themselves implement the functions
in the present embodiment, and the recording medium having
the program codes recorded thereon constitutes one aspect
- 52 -

of the present invention.
0088
Examples of the recording medium for supplying the
program codes include a ROM, a floppy (registered
trademark) disk, a hard disk, an optical disk, a magneto-
optical disk, a CD-ROM, a CD-R, a DVD-ROM, a magnetic tape,
and a nonvolatile memory card. Furthermore, the program
codes may be supplied through downloading via a
communication line, or may be supplied and executed by
means of JAVA (registered trademark) technology or the like.
0089
In addition, it is understood that the present
invention encompasses not only the case where the functions
in the present embodiment are implemented by executing the
program codes read by a computer, but also the case where
an OS or the like that is operating on a computer performs
part or all of the actual processing in accordance with
instructions by the program codes, such that the functions
in the present embodiment are implemented by such
processing.
0090
Furthermore, it is understood that the present
invention also encompasses the case where the program codes
read from a recording medium are written to a memory
included in a function extension board inserted into a
computer or a memory included in a function extension unit
connected to a computer, and thereafter a CPU or the like
- 53 -

provided in the function extension board or the function
extension unit performs part or all of the actual
processing in accordance with instructions by the program
codes, so that the functions in the present embodiment are
implemented by such processing.
0091
In addition, the present invention may be a program
product including a program causing a computer to implement
the functions of the device according to the above
embodiment. Here, the program product refers not only to a
computer program, but also to a recording medium or a
computer, which has the program recorded thereon.
INDUSTRIAL APPLICABILITY
0092
In the information processing system of the present
invention, as described above, the client classifies a
plurality of jobs constituting a large-scale arithmetic
operation into several blocks, and requests the master node
to process the jobs block by block, such that the master
node can always perform the process of assigning a
predetermined number of jobs or less. Thus, the client can
request the master node to process a plurality of jobs
little by little in units of blocks into which the jobs are
classified such that each block consists of small-scale
jobs, rather than simultaneously requesting the master node
to carry out processing of all the jobs as in the
conventional art. Such a client conceivably has the
- 54 -

function of controlling the load on the master node. Thus,
the master node is simply required to always perform the
process of assigning the predetermined number of jobs or
less "to the nodes, and therefore it is possible to prevent
the load on the master node from increasing. Moreover,
since the load on the master node can be reduced as such,
the master node can efficiently perform not only the
process for assigning jobs to the nodes, but also the
process for meeting the client's demand to monitor
arithmetic results. Furthermore, the number of jobs the
client requests the master node to process is reduced, and
therefore the load on the client in terms of the process of
monitoring the arithmetic results is reduced. Thus, the
present invention is applicable to the information
processing system in which a plurality of computers
connected via a network share and execute processing of a
large-scale arithmetic operation consisting of a plurality
of jobs.
- 55 -

CLAIMS
1. An information processing system comprising a client
for requesting processing of a large-scale arithmetic
operation consisting of a plurality of jobs, a plurality of
nodes for executing processing of the jobs, and a master
node for assigning processing of the jobs requested by the
client to the nodes, the client, the nodes, and the master
node being connected together via a network,
wherein the client classifies the jobs constituting
the large-scale arithmetic operation, which are memorized
in a memory means, into several blocks, and requests the
master node to process the jobs block by block, such that
the master node always assigns a predetermined number of
jobs or less.
2. The information processing system according to claim
1, wherein the client classifies the jobs constituting the
large-scale arithmetic operation, such that a predetermined
number of jobs belong to a first block, and other jobs each
belong to an nth block, where n=2, 3, ..., N, and the
client first requests the master node to process the
predetermined number of jobs in the first block, and then
requests the master node to process a predetermined one of
the blocks that have not yet been processed upon each
receipt of a notification from the master node, indicating
that processing of one job has been completed.
- 56 -

3. The information processing system according to claim
1 or 2, wherein the client estimates a processing load per
job included in the large-scale arithmetic operation, and
requests the master node to process the jobs in order from
the highest load to the lowest load in accordance with
estimation results.
4. A computer-readable recording medium having recorded
thereon a program for causing a computer to implement a
function of an information processing system according to
any one of claims 1, 2, and 3.
5. A program for causing a computer to implement a
function of an information processing system according to
any one of claims 1, 2, and 3.
6. An information processing device for requesting a
load balancer to process a large-scale arithmetic operation
consisting of a plurality of jobs, the device comprising:
a grouping process means for grouping the jobs
constituting the large-scale arithmetic operation, which
are memorized in a memory means, based on grouping rule
information memorized in the memory means, thereby forming
several job groups; and
a management means for requesting the load balancer
to process the job groups formed by the grouping process
- 57 -

means, such that the load balancer always processes a
predetermined number of job groups or less.
7. The information processing device according to claim
6, further comprising a load estimation means for
estimating a processing load per job group formed by the
grouping process means based on the load estimation rule
information memorized in the memory means,
wherein based on a processing load estimation result
per job group, which is obtained by the load estimation
means, the management means requests the load balancer to
process the job groups in order from the highest load to
the lowest load.
8. A program for causing a computer to implement a
function of an information processing device according to
claim 6 or 7.
- 58 -

An information processing system is provided to
alleviate excessive load on a master node, thereby allowing
the master node to efficiently perform the process of
assigning jobs to nodes. A client 10 classifies a
plurality of jobs constituting a large-scale arithmetic
operation into several blocks, and requests a master node
20 to process the jobs block by block, such that the master
node 20 always performs the process of assigning a
predetermined number of jobs or less. Here, the
predetermined number is preferably determined in such a
manner as to allow the master node 20 to efficiently
perform the process of assigning the jobs to nodes, even if
the number of nodes is significant. As such, the client 10
has the function of controlling the load on the master node
20, and therefore it is possible to prevent the load on the
master node 20 from increasing.