DESCRIPTION Title of Invention: SYSTEM AND SYSTEM CONTROL METHOD Technical Field: This invention relates to a system that comprises a plurality of function blocks and that can operate with the function blocks integrated in a manner adaptive to autonomous decentralization. Background Art: Currently, an increase in system scale and a phenomenon where what has not been systematized is suddenly turned into a large-scale system are in progress in various fields. A typical example of the former is found in the IT field such as data centers and networks that are related to a technology called cloud, and the scale thereof is increasing year after year. An example of the latter is found in the infrastructure field such as electric power, energy networks, and cities, which are all of a sudden becoming targets to be controlled as a smart grid, a smart city, or other large-scale system cases. How large-scale systems are to be controlled is therefore expected to be a very important issue in the near future. Examples of control exerted in a system of the IT field include load balancing among computer resources, network load balancing, and decentralized storage arrangement. How load balancing among computer resources is controlled is described by taking as an example a load balancing method that is targeted for a data center (DC) including a plurality of servers. The basic idea of conventional control eventuates in "balancing the internal state of the system", which is rather unsophisticated. With this mindset, while the control policy is clear in the case of a uniform system (where the plurality of servers are identical servers), what index is to be the basis of balancing is not clear in a mixed-machine system where different types of machines are used. For instance, while it is known that the CPU utilization ratios of the respective servers are to be balanced in the case of servers having the same performance, it is not obvious how to balance the CPU utilization ratios of servers that have different performance characteristics in a manner that reflects the servers' respective performance. A result obtained by the queueing theory, which is an index often used in conventional control, is an index based on the theory of probability, and tells nothing about what means to use in order to lead the system to a stochastically steady state. What this means is that, although requirements to be fulfilled such as response and throughput may be satisfied eventually, there are no guarantees or limitations with regard to processes in the middle and the situation of resources that are ultimately used. In other words, the resultant control is merely for fulfilling the required performance irrespective of whether the system is in a very inefficient state in terms of energy. Under such control, the inefficiency grows as the system increases in scale, and is expected to be a serious problem. Control in the infrastructure field, typically, smart grid, is also attracting attention these days. This is because, with the arrival of energy-conscious society, situations are emerging where solar energy and other renewable energy sources are connected to conventional electric power grids, in addition to the desire for fine control to cut the waste of energy, and power supply from an electric power grid therefore becomes unstable instead unless the electric power grid is controlled as a total system including renewable energy. Efficient control of such new control targets, too, is therefore an important issue. System control methods are roughly divided into two: one is centralized control and the other is autonomous decentralized control. Centralized management is easier for systems that are not so large, and many systems have been controlled in a centralized manner up to now. Centralized control is also suitable for cases where disturbance external to the system can be predicted and cases where what components constitute the system can be envisioned. In short, centralized control is very effective for predictable environments or scales that allow an overview. Centralized control is control in which an accurate overall picture of the system is grasped constantly, to thereby determine the situation and what action is to be taken. Simply put, centralized control is a method of setting down codes by if-then-else. However, centralized systems are finding it difficult to handle large-scale systems due to the increase in system scale of these days, and autonomous decentralized systems are actively being studied. Autonomous decentralized control is control in which components such as resources operate using only relations among the components, and does not always need accurate overall information. As the system scale increases, knowing accurate overall information becomes more difficult and autonomous decentralized control is accordingly desired more. In electric power grids, typically smart grids, where predicting timing at which renewable energy is input to the system is difficult, centralized control may not be able to deal with a change in resource in real time. In IT systems, too, considering accidents such as a failure, autonomous control under which the system operates in real time so as to repair damage from an accident autonomously is naturally preferred to centralized control that is designed with such events in mind. Autonomous decentralized control is being studied actively because of these demands. While a lot of people are studying self-proclaimed autonomous decentralized systems, it is also a fact that a very high percentage of the proposed ones are systems in which components merely exist in an autonomous decentralized manner and are overall not in concert with one another. These are not autonomous decentralized systems in its true sense. It is easy to create a system in which components exist in an autonomous decentralized manner, but there is no sense in designing a system in which components are overall not in concert with one another, or components do not operate for the goal of overall optimum. It is a system in which components operate for the goal of overall optimum in an autonomous decentralized manner that should be called an autonomous decentralized system. What constitutes such a true autonomous decentralized system? Actually, such a system can be found close to home: the system of living organisms or brains is the true autonomous decentralized system. Autonomous decentralized systems taken after living organisms or brains are therefore actively being researched these days. Avery interesting study among them is a technology that uses "yuragi (which means "fluctuation" in Japanese) equation" ("Technological Innovation with Yuragi", Osaka University Yuragi Project, 2008). This is a technology of controlling components based on the yuragi equation, which is expressed by die following Expressions (1) and (2). In the expressions, Activity is a type of bias and a function that takes a large value at a point x where the system in question feels comfortable. Symbols n and U respectively represent random noise and potential for driving the system. The yuragi equation is similar to a Langevin equation expressed by the following Expression (3). These equations superficially resemble each other but there is a critical difference from the standpoint of engineering. The only difference between the yuragi equation and the Langevin equation is Activity. In a sense, the yuragi equation seems merely an expression in which potential is divided into U and Activity. This, however, creates the critical difference. Hitherto, when the Langevin equation is used to control something, the potential U needs to be obtained accurately. A problem for obtaining potential with regard to a non-linear, intricate problem is very difficult, and there are few cases where the problem can be solved analytically. Machine learning and reinforcement learning which are popular nowadays are in actuality physically equal to obtaining the potential of a system, and use a neural network or the like to obtain the potential by a black-box approach. At present, there is no other way than the black-box approach to obtain potential for a non-linear, intricate problem with conventional technologies. Anyway, obtaining potential is equal to setting a model, and aiming for more complex operation therefore complicates a potential function extremely, which gives rise to a problem in that an event cannot fully be expressed by a single potential function. A solution to this is the yuragi equation. In the yuragi equation, comfortableness felt by the system is separated as Activity from the overall potential, thereby enabling a person to specify the operation of the system in a top-down manner. Specifically, Activity is set so that operation expected from the system is implemented. For instance, when one wishes to move a system to a point, a function that takes a maximum value at the desired point is given as Activity. In the case where the system is far from the desired point, the first term of the right-hand side of Expression (1) takes a small value, and components of the system are therefore driven by random noise of the second term of the right-hand side. This is a state similar to Brown motion if the components of the system are to be likened to particles. The system positioned far from the desired point in some cases accidentally comes close to the desired point through random movement such as Brown motion. The first term of the right-hand side of Expression (1) in this case takes a large value because of the action of the Activity function. Then what governs system operation shifts from the second term of the right-hand side to the first term of the right-hand side. Once the first term of the right-hand side starts controlling the system, the system now naturally settles at a potential point where Activity is maximum. In short, the desired point is reached. What is important here is that, when the yuragi equation is used, only Activity is determined in a top-down manner without issuing a special command to the components of the system or exerting special control on the components of the system. An output to be given to each component of the system is an output that is in proportion to the value of the right-hand side of Expression (1). Cases where there are a plurality of components can similarly be dealt with by giving outputs that are in proportion to the right-hand side of Expression (1). In this case, random noise in one driving unit is independent of random noise in another driving unit. Activity, on the other hand, is the same for all components. This way, despite the respective components simply operating in accordance with their individual versions of Expression (1), the action of Activity which integrates the overall system causes the components to eventually operate in concert with one another on the whole and move to a point where Activity is maximum. A system comprising a plurality of components, too, can thus operate in an autonomous-decentralized, concerted manner by using the yuragi equation. As disclosed in Patent Literature 1, for example, autonomous decentralized control of network paths using this technology is being studied and results indicating effectiveness such as autonomous avoidance of failure have been obtained (Patent Literature 1). Citation List: Patent Literature Patent Literature 1: Japanese Unexamined Patent Application Publication (JP-A) No. 2005-285016 Disclosure of the Invention: Problems to be Solved by the Invention: As described above, the "yuragi equation" which is a prospective technology is being studied, and has been found to have room for improvement in two points. One is that control by the yuragi equation is stochastic control. Specifically, because random noise is used in Expression (1) as the source of power for each component, the component does not always operate deterministically. This means that the component's operation varies depending on the probability at the time, which can lead to redundant control. Components of a true autonomous decentralized system should operate deterministically for the goal of overall optimum. The other point concerns setting Activity. As described above, the Activity function can be set by a person in a top-down manner. In actuality, however, there are many cases where a person cannot imagine a function of the overall maximum itself. Only functions about the components can be set or checked by a person, and it is actually very often the case that a function indicating a state where the components are in concert with one another cannot be imagined. This invention has been made in view of the problems described above, and an object of this invention is therefore to provide a system in which components can overall operate in concert with one another by using only functions about the components, without using the theory of probability. Means to Solve the Problems: In order to accomplish the above-mentioned object, according to a first aspect of this invention, there is provided a system, including a plurality of function blocks correlated to one another, in which the plurality of function blocks each include: a storing unit including an evaluation function that is associated with the each of the plurality of function blocks; a restraint condition fulfillment control unit for controlling an operation level of the each of the plurality of function blocks based on a restraint condition that is to be fulfilled by the each of the plurality of function blocks and all of the plurality of function blocks together; a profit maximization control unit for controlling the operation level of the each of the plurality of function blocks by information about the evaluation function of another of the plurality of function blocks that has a relation with the each of the plurality of function blocks; and an activation/shutdown problem solution control unit for controlling an active state/shutdown state of the each of the plurality of function blocks depending on the operation level of the each of the plurality of function blocks, by using results of the control of the restraint condition fulfillment control unit and the control of the profit maximization control unit. According to a second aspect of this invention, there is provided a system control method for use in the system as described in the first aspect. Effect of the Invention: According to this invention, it is possible to provide the system in which the components can overall operate in concert with one another by using only the functions about the components, without using the theory of probability. Brief Description of the Drawings: Fig. 1 is a schematic system diagram studied in the invention of this application. Fig. 2 is a schematic architecture diagram according to the invention of this application. Fig. 3 is a schematic graph of evaluation functions that are used in the invention of this application. Fig. 4A is a graph of a control image according to the invention of this application that is for a case where an activation/shutdown problem solving unit is not put into operation. Fig. 4B is a graph of a control image according to the invention of this application that is for the case where the activation/shutdown problem solving unit is not put into operation. Fig. 4C is a graph of a control image according to the invention of this application that is for the case where the activation/shutdown problem solving unit is not put into operation. Fig. 5 A is a graph of a control image according to the invention of this application that is for a case where the activation/shutdown problem solving unit is put into operation. Fig. 5B is a graph of a control image according to the invention of this application that is for the case where the activation/shutdown problem solving unit is put into operation. Fig. 5C is a graph of a control image according to the invention of this application that is for the case where the activation/shutdown problem solving unit is put into operation. Fig. 6 is a graph showing an example of an evaluation function that is used in the invention of this application. Fig. 7Ais a graph showing an experiment result in an embodiment of the invention of this application. Fig. 7B is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 8A is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 8B is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 9A is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 9B is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 10 is a graph showing an example of an evaluation function that is used in the invention of this application. Fig. HAis a graph showing an experiment result in an embodiment of the invention of this application. Fig. 1 IB is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 12 is a schematic system diagram studied in the invention of this application. Fig. 13Ais a graph showing an experiment result in an embodiment of the invention of this application. Fig. 13B is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 14 is a schematic system diagram studied in the invention of this application. Fig. 15A is a graph showing an experiment result in an embodiment of the invention of this application. Fig. 15B is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 16 is a schematic system diagram studied in the invention of this application. Fig. 17 is a graph showing an example of an evaluation function that is used in the invention of this application. Fig. 18A is a graph showing an experiment result in an embodiment of the invention of this application. Fig. 18B is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 19 is a schematic system diagram studied in the invention of this application. Fig. 20A is a graph showing an experiment result in an embodiment of the invention of this application. Fig. 20B is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 21A is a schematic system diagram studied in the invention of this application. Fig. 2IB is a schematic system diagram studied in the invention of this application. Fig. 22 is a graph showing an experiment result in an embodiment of the invention of this application. Fig. 23 A is a schematic system diagram studied in the invention of this application. Fig. 23B is a schematic system diagram studied in the invention of this application. Fig. 24A is a graph showing an experiment result in an embodiment of the invention of this application. Fig. 24B is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 25 A is a graph showing an experiment result in the embodiment of the invention of this application. Fig. 25B is a graph showing an experiment result in the embodiment of the invention of this application. Mode for Embodying the Invention: A preferred embodiment mode of this invention is described in detail below with reference to the drawings. The operation and configuration of this invention are described first taking as an example load balancing among resources of a data center. Discussed here as an example in which a problem cannot be solved satisfactorily with conventional technologies is load balancing in a mixed-machine data center where servers of different performance characteristics are used. In a data center where a plurality of (N) servers of different performance characteristics are coupled by a network, processing requests are distributed among the servers as seen fit while requested response values are satisfied. At the same time, the overall power consumption of the data center is reduced as much as possible. Such a problem is examined in the following. A schematic diagram of this problem is illustrated in Fig. 1. As a result of studying the problem, the inventors of this invention have reached the following conclusion. The problem to be solved has a structure in which two problems are intermingled. One is an activation/shutdown problem and the other is an optimization problem. The activation/shutdown problem is a problem of determining or managing which resource is to be used and which resource is not to be used when there is a group of resources (a group of function blocks). The optimization problem is a problem of how a task assigned to the overall system is to be distributed in order to accomplish optimization based on an index when resources are allocated. Hitherto, these problems have often been attempted to be solved separately by stochastic approaches. With the simple theory of probability, however, it is difficult to quickly deal with a situation in real time in a robust manner in an environment that keeps changing freely. The inventors of this invention have devised architecture with which a situation can be quickly dealt with in real time in a robust manner depending on the circumstances by regarding the two problems as a unitary problem. The inventors of this invention have also devised a method of solving the respective problems by a deterministic approach or a relational approach, instead of a stochastic approach. Fig. 2 illustrates a schematic diagram of the architecture of a function block devised by the inventors of this invention and a system that bundles the function block with others. As illustrated in Fig. 2, the function block of this application mainly includes an activation/shutdown problem solution control unit and an optimization problem solution control unit, and constitutes the system together with a correlated function block (another function block). A detailed description is given next on solving means of the respective parts. A solution to the optimization problem is described first. Two requirements need to be considered in order to solve the optimization problem. One is demand-supply balancing and the other is the maximization of the overall profit. It cannot be said that optimization is accomplished unless the two are both satisfied. Demand-supply balancing control is also control for fulfilling a restraint condition, and may therefore be called restraint condition fulfilling control. In short, the optimization problem solution control unit in the architecture of Fig. 2 comprises a profit maximization control unit and a restraint condition fulfillment control unit. Demand-supply balancing is in many cases a restraint condition in that the total amount of a mission or task required of the overall system needs to be satisfied. Unless this is fulfilled, there is no sense in maximizing efficiency or profit. The demand in this example is the sum of values of an operation level A4 (i represents a function block number), and is expressed by the following Expression (4). To accomplish demand-supply balancing, the restraint condition fulfillment control unit determines the operation level of each function block (resource) with the use of an equation that satisfies the following Relation (5). In (5), At represents the operation level of another function block that has a relation with a function block of the operation level A. This equation is more specifically expressed by the following Expression (6). In this expression, the symbol Ki is a coefficient equivalent to a gain of a change in operation level. The symbol ^om,j represents a normalization coefficient of the function block i (an amount equivalent to the scale of the function block), and is not always necessary. However, multiplying by a normalization coefficient overall is often preferred in the case of heterogeneous components, which is why A,nom,i is introduced in Expression (6). The restraint condition fulfillment control unit controls the operation levels of the function blocks in a manner determined by Expression (6), thereby causing the overall system to operate so as to fulfill Dem, which is the restraint condition. Next to be solved is overall profit maximization, which may be reworded as overall efficiency maximization. To solve overall profit maximization, such evaluation functions as those of Fig. 3 which are related to the respective components (function blocks) are introduced (in Fig. 2, the evaluation functions are stored in a storing unit of the function block). In this example, the axis of abscissa represents a parameter for controlling each function block, which corresponds to a processing request amount X or the like. The axis of ordinate represents an index related to some form of efficiency, and concrete examples of the index are described in detail in embodiments discussed later. As is obvious from Fig. 3, the evaluation functions are convex functions. The use of convex functions is another point of the invention of this application. This is because some form of efficiency, system stability, and the like can be expressed by such convex functions as those illustrated in Fig. 3 in many systems. While it is a common practice to call functions that are convexed upward as those in Fig. 3 as concave functions and to call functions that are convexed downward as convex functions, the expression that distinguishes functions by the functions' properties is employed herein, and concave functions, too, are expressed as convex functions. A problem for accomplishing overall optimization (a state where the sum of values of evaluation functions of the respective components is maximum) through cooperation between components whose evaluation functions are convex functions is known as "convex programming problem". It has mathematically been proven that, in a convex programming problem, optimization is accomplished under a situation where differential values of evaluation functions at the operation levels of the respective components are equal to one another. This principle is applied here, which is the reason why convex functions are used here as evaluation functions. Considering this principle, an equation for accomplishing overall profit maximization satisfies the following Relation (7). In Expression (7), f; is an evaluation function of the function block i (the i-th function block), and fk is an evaluation function of another function block k that has a relation with the function block. The equation is more specifically expressed by the following Expression (8). In this expression, the symbol K2 is a coefficient equivalent to a gain of a change in operation level. The profit maximization control unit controls the respective function blocks so that Expression (8) is satisfied, thereby causing the function blocks to operate in a manner that equalizes differential values of evaluation functions at the operation levels. To solve the optimization problem ultimately, the demand-supply balancing and overall profit maximization described above need to be solved simultaneously. The optimization problem solution control unit therefore controls the operation levels of the respective function blocks by the following Expression (9), which is created by integrating Expression (6) and Expression (8). A solution to the activation/shutdown problem is described next. Expression (9) actually contains a function of activation/shutdown a component. The second term of the right-hand side has an effect in that one component and another component suppress each other via gradients of evaluation functions. The phenomenon that actually occurs is that a component whose evaluation function is greater in gradient (greater in efficiency improvement) suppresses, or even stops in some cases, a component whose evaluation functions is smaller in gradient (smaller in efficiency improvement). It can therefore be said that the second term of the right-hand side of Expression (9) contains a function of solving a shutdown problem out of the activation/shutdown problem. On the other hand, the first term of the right-hand side is capable of activating a component that has shut down or that has been shut down. When the demand falls short, the first term of the right-hand side takes a positive value and moves each component toward a direction in which the component operates more. This means that as a result, even a component that has shut down is moved toward a direction in which the component is activated. The first term of the right-hand side of Expression (9) therefore contains a function of solving an activation problem out of the activation/shutdown problem. However, simply applying Expression (9) does not constitute solving the optimization problem after the activation/shutdown problem is solved, because Expression (9) as it is just contains the function of activating and the function of shutting down. Figs. 4A to 4C show an example of how function blocks operate in the case where the activation/shutdown problem solution control unit simply solves Expression (9). In this example, the activation/shutdown problem solution control unit uses Expression (9) to control how much load (power) each component needs to bear in order to accomplish a load of 0.4 in total when three function blocks (here, corresponding to servers) coupled in series comprise evaluation functions shown in Fig. 4A. A limiter is applied so that a load equal to or smaller than 0 is treated as 0. Fig. 4B shows changes with time of the load of the respective components, and Fig. 4C shows changes with time of the differential values of the evaluation functions of the respective components. It can be seen in Fig. 4(b) that a "node 3" (a function block 3) which is poor in efficiency has automatically shut down. Though it superficially seems that the activation/shutdown problem is solved by Expression (9), it is understood from Fig. 4C that differential values of a "node 1" and a "node 2" do not match. While there is no need for differential values of the "node 1" and the "node 2" to match the differential value of the "node 3" which has shut down, a mismatch between differential values of the active "node 1" and "node 2" means that the system has not reached an optimized state. In other words, the activation/shutdown has been accomplished but not the optimization. This is because simply applying Expression (9) causes a function block that has shut down to suppress a function block that is active. In addition, a rapid increase in demand, for example, may not lead to the activation of the "node 3" which has shut down in Figs. 4 even when the activation function of the first term of the right-hand side of Expression (9) is in effect, depending on suppression by the second term of the right-hand side of Expression (9). For instance, the "node 3" remains shut down when an increment of the first term of Expression (9) is less than the suppression by the second term. Thus, solving the activation/shutdown problem and the optimization problem both cannot be achieved by simply solving Expression (9) by the activation/shutdown problem solution control unit. This invention addresses this by giving a function of managing the activation/shutdown state of function blocks and adjusting control of Expression (9) to the optimization problem solution control unit controlled by Expression (9). Specifically, the activation/shutdown problem solution control unit is given the following functions. The first function is for excluding a function block that has shut down (a function block that has reached the minimum value of its share of the load) or a function block that has reached an upper limit to the load in the calculation of the second term of the right-hand side of Expression (9). The second function is for excluding the second term of the right-hand side of Expression (9) and using only the first term of the right-hand side when calculating the load of a function block that has shut down or a function block that has reached an upper limit to the load (namely, a function of circumventing the use of the profit maximization control unit). The point is that the first term of the right-hand side of Expression (9) and the second term of the right-hand side of Expression (9) are handled independently of each other, depending on the state of function blocks. This is where the significance of existence of the activation/shutdown problem solution control unit resides. The system can thus solve the activation/shutdown problem (including a function block that has reached an upper limit and is therefore substantially the same as a function block that has shut down) while separating a function block that has shut down and therefore substantially cannot be changed in operation level, or a function block that has reached an upper limit to the load, from the control of the profit maximization control unit, and can use, if necessary, the restraint condition fulfillment control unit to bring the function block that has substantially shut down into an operating state (active state). More concrete descriptions are given on the two functions described above. Excluding a function block that has shut down or a function block that has reached an upper limit to the load in the calculation of the second term of the right-hand side of Expression (9) means that, when calculating the load of a function block that is active, for example, 0 is set to the numerator of the second term of the right-hand side of Expression (9) for a function block that has shut down or a function block that has reached an upper limit to the load. Because the numerator of the second term of the right-hand side is of course not 0 for function blocks that are active, there is a chance that the second term of the right-hand side has some value in the case where a function block that is active is connected to a function block for which the load is to be calculated. When calculating the load of a function that has shut down or a function block that has reached an upper limit, on the other hand, the second term of the right-hand side is ignored, in other words, the second term of the right-hand side is treated as 0. This way, only the first term of the right-hand side of Expression (9) affects a function block that has shut down or a function block that has reached an upper limit, and these function blocks can be activated again when there is a problem in demand and supply. Control for solving the activation/shutdown problem and the optimization problem unitarily is thus completed by the activation/shutdown problem solution control unit and optimization problem solution control unit proposed in the invention of this application. Figs. 5A to 5C show an example of controlling load balancing among three "nodes" (function blocks) when the demand changes from 0.4 to 0.8 (at a time 2). In Figs. 5A to 5C, which use the same notation as in Figs. 4A to 4C, control for shutting down the "node 3" when the demand is 0.4 and activating the "node 3" anew when the demand changes to 0.8 is implemented. The gradients of evaluation functions ultimately converge into the same values at any one of the two values of the demand, which proves that the optimization problem, too, has been solved. Although demand-supply balancing is often a restraint condition in that the total amount of a mission or task required of the overall system needs to be satisfied as stated in the description given above on demand-supply balancing, there are other cases. Control in that case involves transforming the demand-supply balancing terms described above so that the restraint condition is fulfilled. In this case, too, it is vital to manage the situations of function blocks with the activation/shutdown problem solution control unit. Otherwise, activation and shutdown are not implemented properly. The important point here is that, while evaluation functions of the respective function blocks are set, an evaluation function for the function blocks as a whole is not set. Not needing to set an overall evaluation function is a huge advantage because, as opposed to evaluation functions of individual function blocks which can be set in various ways, how to set an overall evaluation function becomes increasingly unclear as the system grows more complicated. In addition, the system here may allow each component to operate based only on information about its own state and the state of a component related to itself. Although information used in demand-supply balancing may be regarded as overall information, in an electric power system or other cases where a poor demand-supply balance can be detected from fluctuations in alternating frequency, demand-supply balancing, too, can be implemented in a completely autonomous decentralized manner. In short, a system can be a true autonomous decentralized system which does not need overall information if the system is capable of obtaining information relevant to demand-supply balancing from itself. This system basically operates independently in an autonomous decentralized manner and, when there is a failure in a component, other components autonomously execute a recovery operation so as to cover the loss of signals from the failed component. The system can also autonomously and gradually be adjusted toward a proper operation after a sudden addition or removal of a component. In other words, the system is very robust with respect to external disturbance, and has a scalability with which components can be added or removed freely. Conventional systems are powerless to failures and need to prepare various sequences for errors. Conventional systems also have no guarantees that the overall system stability is maintained against an unplanned addition or removal of components (resources), and require reviewing programs and processing each time. The system of this application can solve all of those problems by control adaptive to autonomous decentralization. Embodiments More detailed descriptions are given below on embodiments of this invention with reference to the drawings. (First Embodiment) The invention of this application is carried out for the load balancing among resources of a data center that is illustrated in Fig. 1. The operation level (n represents a function block number) of a resource in this case corresponds to a task amount allocated to the resource. When load balancing among resources of a data center is considered, restraint conditions that matter to an administrator are response, throughput, energy, and the like. An embodiment of the invention of this application in which response and throughput are restraint conditions is described first. Response and throughput, which have a strong relation to each other, are respectively defined as follows. A response R is the very time required for processing, and a throughput S is a workload processed per unit time. Therefore, simply, the following Expression (10) is established in relation to a requested workload X. S=9iR - (10) (S<1) However, an effective response Re of an actual resource is not like this, and requires considering a stochastic queueing time by the queueing theory. This is because the effective response depends not only on the processing performance of the resource of its own but also on how much work arrives and how much work is unprocessed and has built up, which are clarified by the queueing theory through stochastic calculation. According to the queueing theory, a time Tm from the arrival of work to the completion of the processing of the work is expressed by the following Expression (11). Tm=l/(n-Jl) - (11) In this expression, the symbol \i represents a processing rate per unit time, and is the reciprocal of the response R. With the effective response Re, which can be regarded as the reciprocal of Tm, and the relation of (10), the following Expression (12) is established. Re=\i-X =1/R-S/R =(1-S)/R - (12) The effective response Re, too, thus has a close relation with the throughput. What is understood from Expression (12) is that the effective response can be improved by reducing the throughput, whereas the throughput can be improved by deteriorating the effective response. In short, the effective response and the throughput are in a trade-off relation with each other in which one of the two is in an invalid state (becomes 0) at the limit value of the other, and the administrator needs to balance the two at an appropriate balance. How the administrator is to implement the load balancing in a data center with die use of the invention of this application? The effective response and the throughput are in a trade-off relation as described above, which means that a rather light workload is to be distributed among components (resources) when response is valued more, whereas a heavy workload is to be distributed when throughput is valued more. Accordingly, functions set to the respective components are set by, for example, the following method: Fig. 6 is a schematic diagram of evaluation functions. The axis of abscissa represents the workload because the subject studied here is load balancing. As described above, convex functions are preferred as evaluation functions, and a quadratic function that is convexed upward is used here. The symbol X^ax represents a limit of the resource in question, and the symbol Xpeak represents a peak point of the evaluation function. By normalization in which the peak of the evaluation function is set as 1, the normalized evaluation function is determined as a quadratic function that passes through (0, 0) and (Xpeak, 1)- An evaluation function is obtained for each component (resource) by multiplying the normalized evaluation function by a coefficient a;, which is in proportion to a difference in performance between components. The point Xpeak is where the throughput S is 1 and the effective response Re is 0. As described above, a rather light workload is distributed among the components (resources) when response is valued more, and a rather heavy workload is distributed when throughput is valued more. Accordingly, in the case where the system is to process work by putting importance on response, the point Xpeak is set rather small whereas the point A^ak is set rather large when the system is to process work by putting importance on throughput. The system uses the method of this invention of this application to operate while solving the activation/shutdown problem and the optimization problem, and aims for a point as close to the point Xpeak of each component as possible. Therefore, with the point Xpeak set rather small, work is distributed among components bearing relatively light loads, and the overall system operates in a manner that values response more. With the point Xpeak set rather large, work is distributed among components bearing relatively heavy loads, and the overall system operates in a manner that values throughput more. Resources being wasted are shut down as seen fit, and the resultant control is free from a waste of energy as well. The axis of ordinate of the evaluation function in this case does not really represent some form of efficiency. The axis of ordinate rather indicates a point where the administrator wishes for the system to operate, and setting the evaluation function is not ruled by any physical quantity or principle. A person who wishes to run the system can set the axis of ordinate and the axis of abscissa at his/her own discretion. In a sense, once the system is given an instruction from a person in the form of an evaluation function, the system operates autonomously in a concerted manner from then on. While the evaluation functions here are obtained by multiplying a quadratic function that passes through (0, 0) and (A^ak, 1) by the coefficient