JP6417341B2 - Management device and software component grouping method - Google Patents

Description

  The present invention relates to a technique for determining an appropriate arrangement of a plurality of software components constituting an application applied to a distributed system.

  In recent years, applications have been increasingly applied to hierarchical distributed systems. For example, as shown in FIG. 8, the target distributed system is composed of four layers of a data center, a rack under the switch, a physical server, and a virtual server in order from the lower layer.

A plurality of data centers are provided.
A plurality of racks are provided corresponding to one data center.
A plurality of physical servers are provided corresponding to one rack.
A plurality of virtual servers are provided corresponding to one physical server.
A plurality of processes are provided corresponding to one virtual server.
Then, software components are arranged (loaded) in a one-to-one correspondence with the process.

  An application is composed of a plurality of software components. Here, the software component refers to a group of software that provides a certain function.

  In the distributed system shown in FIG. 8, communication occurs between arbitrary software components. And communication between software components takes time, so that the hierarchy via is a lower layer. In other words, in such a hierarchical distributed system, the communication delay increases as the communication path reaches the lower layer. Therefore, if the communication path between software components with high communication frequency reaches the lower layer, The effect of delay will be noticeable.

  Therefore, since the arrangement of software components greatly affects the processing time (communication time), in order to reduce communication delay, it is necessary to perform communication between software components as much as possible.

  In view of such circumstances, Non-Patent Document 1 discloses a method for optimizing the arrangement of software components in a distributed environment so as to reduce communication delay using graph theory.

Hiroaki Kobayashi and 3 others, “Optimization of Software Component Placement in a Distributed Environment”, IEICE, IEICE Technical Report, NS2015-17 (2015-5)

  In applying an application to a distributed system, it is necessary to set predetermined constraints depending on application requirements from various viewpoints such as improvement of resource utilization efficiency and improvement of reliability / availability. However, it is not easy to manually determine the placement of software components while satisfying such constraints. Non-Patent Document 1 has neither a description nor a suggestion for solving such a problem.

  Therefore, in view of the above circumstances, the present invention has an object to determine an appropriate arrangement of a plurality of software components that constitute an application while satisfying a restriction required for an application applied to a distributed system having a plurality of layers. And

In order to solve the above problems, the invention according to claim 1, either from the first hierarchy of the hierarchy of most higher, up to N-th layer made of the most lower level (N is an integer of 2 or more) a management apparatus for determining the placement of a plurality of software components that make up an application to be applied to a distributed system having a device that functions as Kano hierarchy before Symbol software component is arranged on the first layer, wherein communication between the software components through the same equipment, or the first N is realized by way of the inter-hierarchy comprising equipment, communication between the software components, the hierarchy indicated by the device through which is subordinate more time consuming, the software component that corresponds to the device that functions as a K hierarchy (K = 1,2, ···, N ) May constitute a first K hierarchical group by the grouping processing, the management device includes a storage unit and a processing unit, the storage unit, the software components first each pair of communicating communication frequency, and, wherein is determined based on the first communication frequency, and the communication frequency data indicating the second communication frequency for each pair of the first K hierarchical group, predetermined for each pair of a particular of the software component And maximum distance constraint property data indicating a maximum distance constraint property value determined in advance for each specific pair of the K-th layer group, and the processing unit is configured to store a communication frequency of the storage unit. Referring to the data, in descending order of communication frequency pairs, the software components, and, Gurupin of the first K hierarchical group Processing, in accordance with a predetermined capacity constraint, when you execute repeatedly from said first layer to said N-th layer, each time the hierarchy the grouping process is performed is switched to one lower level, the maximum distance constraint property It decrements the value, the maximum distance constraint property values Ru to belong to the same group of pairs reached the lower limit value, characterized in that.

The invention described in Claim 5, the first layer comprising a hierarchy of most higher, functions as one of the hierarchy to the N-th layer made of the most lower level (N is an integer of 2 or more) a software component grouping method by the management apparatus to determine the arrangement of a plurality of software components that make up an application to be applied to a distributed system having a device, prior SL software component is arranged in the first hierarchy, the software communication between components is via the same device, or, the first N is realized by way of the inter-hierarchy comprising equipment, communication of the software components each other, as the hierarchy shown is device through which is subordinate time consuming, the K hierarchy (K = 1,2, ···, N ) before corresponding to the device that functions as a Software components may be configured the first K hierarchical group by the grouping processing, storage unit of the management apparatus, a first communication frequency for each pair of said software components that communicate, and the first Communication frequency data indicating a second communication frequency for each pair of the K-th layer group, determined in advance based on a communication frequency , predetermined for each specific software component pair , and the specific Kth Maximum distance constraint property data indicating a maximum distance constraint property value determined in advance for each pair of hierarchical groups, and the management device refers to the communication frequency data of the storage unit, and the communication frequency is high in order from the pair, the software components, and, the grouping process of the first K hierarchical group In accordance with a predetermined capacity constraint, when you execute repeatedly from said first layer to said N-th layer, each time the hierarchy the grouping process is performed is switched to one lower level, decrementing the maximum distance constraint property value And the pair whose maximum distance constraint property value has reached the lower limit value belongs to the same group.

According to the first and fifth aspects of the invention, by setting the maximum distance constraint property value to a specific software component pair and a specific Kth layer group pair, the communication frequency is low and is inherently low. Even a pair of software components and a pair of a K-th layer group that are arranged in different groups can be intentionally assigned to the same group with respect to application requirements.
Therefore, it is possible to determine an appropriate arrangement of a plurality of software components that constitute an application while satisfying a restriction required for the application applied to the distributed system having a plurality of hierarchies.

  Further, in the invention described in claim 2, in the invention described in claim 1, when the processing unit causes a pair whose maximum distance constraint property value has reached the lower limit value to belong to the same group, the capacity of the group is determined. If the number exceeds, the software component or the K-th layer group constituting the pair is scaled out.

  According to the second aspect of the present invention, by scaling out the software component or the Kth hierarchical group constituting the pair whose maximum distance constraint property value has reached the lower limit value, the software component or the Kth hierarchical group becomes The group to which it belongs can be formed reliably.

The invention according to claim 3, the first layer comprising a hierarchy of most higher, functions as one of the hierarchy to the N-th layer made of the most lower level (N is an integer of 2 or more) a management apparatus for determining the placement of a plurality of software components that make up an application to be applied to a distributed system having a device, prior SL software component is arranged on the first layer, wherein the software components communicate with each other is , via the same device, or, the first N is realized by way of the inter-hierarchy comprising equipment, communication of the software components each other, the hierarchy indicated by the device through which the consuming more time is subordinate, the The software component corresponding to the device functioning as the K hierarchy (K = 1, 2,..., N) is subjected to grouping processing. It is possible to constitute the first K hierarchical group by the management device includes a storage unit and a processing unit, wherein the storage unit, the first communication frequency for each pair of said software components that communicate, and Communication frequency data indicating a second communication frequency for each pair of the K-th layer group, determined based on the first communication frequency, predetermined for each specific pair of software components , and Minimum distance constraint property data indicating a predetermined minimum distance constraint property value for each specific pair of the K-th layer group, and the processing unit refers to communication frequency data in the storage unit , in descending order of communication frequency pairs, the software components, and, the grouping process of the first K hierarchical group, predetermined volume According about conditions, when you run repeatedly from said first layer to said N-th layer, each time the hierarchy the grouping process is performed is switched to one lower hierarchy, decrements the minimum distance constraint property value, the minimum distance constraint property value is not such to belong pair does not reach the lower limit value in the same group, characterized in that.

The invention of claim 6, the first layer comprising a hierarchy of most higher, functions as one of the hierarchy to the N-th layer made of the most lower level (N is an integer of 2 or more) a software component grouping method by the management apparatus to determine the arrangement of a plurality of software components that make up an application to be applied to a distributed system having a device, prior SL software component is arranged in the first hierarchy, the software communication between components is via the same device, or, the first N is realized by way of the inter-hierarchy comprising equipment, communication of the software components each other, as the hierarchy shown is device through which is subordinate time consuming, the K hierarchy (K = 1,2, ···, N ) before corresponding to the device that functions as a Software components may be configured the first K hierarchical group by the grouping processing, storage unit of the management apparatus, a first communication frequency for each pair of said software components that communicate, and the first Communication frequency data indicating a second communication frequency for each pair of the K-th layer group, determined in advance based on a communication frequency , predetermined for each specific software component pair , and the specific Kth Minimum distance constraint property data indicating a minimum distance constraint property value determined in advance for each pair of hierarchical groups, and the management device refers to the communication frequency data of the storage unit, and the communication frequency is high in order from the pair, the software components, and, the grouping process of the first K hierarchical group In accordance with a predetermined capacity constraint, when you execute repeatedly from said first layer to said N-th layer, each time the hierarchy the grouping process is performed is switched to one lower level, decrementing the minimum distance constraint property value and, wherein the minimum distance constraint property value does not belong to a pair does not reach the lower limit value in the same group, and wherein and this.

According to the third and sixth aspects of the invention, by setting the minimum distance constraint property value to a specific software component pair and a specific Kth layer group pair, communication frequency is high and inherently Even a pair of software components and a pair of a Kth hierarchical group that are arranged in the same group can be intentionally assigned to different groups with respect to application requirements.
Therefore, it is possible to determine an appropriate arrangement of a plurality of software components that constitute an application while satisfying a restriction required for the application applied to the distributed system having a plurality of hierarchies.

According to a fourth aspect of the present invention, in the third aspect of the present invention, the storage unit is predetermined for each specific pair of software components , and for each specific pair of the Kth hierarchical group. Further storing maximum distance constraint property data indicating a predetermined maximum distance constraint property value, and performing any of the grouping process based on the minimum distance constraint property data and the grouping process based on the maximum distance constraint property data. Priorities for determining whether to prioritize are included in the minimum distance constraint property data and the maximum distance constraint property data, and the processing unit stores the hierarchy in which the grouping process is performed in a lower hierarchy. Each time it switches, the maximum distance constraint property value is decremented, and the previous Pairs whose maximum distance constraint property value has reached the lower limit belong to the same group, and refer to the priority of the minimum distance constraint property data and the priority of the maximum distance constraint property data, to the minimum distance constraint property data the grouping processing based, and characterized by performing at least one of the grouping processing based on the maximum distance constraint property data.

  According to the fourth aspect of the present invention, by including priority in the minimum distance constraint property data and the maximum distance constraint property data, even if the minimum distance constraint and the maximum distance constraint conflict, at least one of the constraints And the placement destination of the software component can be determined with certainty.

  ADVANTAGE OF THE INVENTION According to this invention, the appropriate arrangement | positioning of the several software component which comprises an application can be determined, satisfy | filling the restrictions calculated | required with respect to the application applied to the distributed system which has a several hierarchy.

1 is an overall configuration diagram of a distributed system and a management apparatus according to an embodiment. It is explanatory drawing of step 1 of the grouping process of a software component. It is explanatory drawing of step 2 of the grouping process of a software component. It is explanatory drawing of step 3 of the grouping process of a software component. It is explanatory drawing of the flow of grouping which considered the maximum distance restriction | limiting. It is explanatory drawing of the case where the maximum distance restriction cannot be satisfied. It is explanatory drawing of the flow of grouping which considered the maximum distance restriction | limiting. It is explanatory drawing of the distributed system of a prior art.

  Hereinafter, embodiments of a distributed system 1 and a management apparatus 2 according to the present invention will be described with reference to the drawings.

  As shown in FIG. 1, the distributed system 1 is a system to which an application composed of a plurality of software components is applied. The data center 11, the rack 12, the physical server 13, and the virtual server 14 are sequentially installed from the lower layer. It consists of four layers.

A plurality of data centers 11 are provided.
A plurality of racks 12 are provided corresponding to one data center 11.
A plurality of physical servers 13 are provided corresponding to one rack 12.
A plurality of virtual servers 14 are provided corresponding to one physical server 13.
A plurality of processes 15 are provided corresponding to one virtual server 14.
In the process 15, software components 3 (hereinafter, may be referred to as “software components” without reference numerals) are arranged in a one-to-one correspondence.

  In the distributed system 1, the communication between the software components arranged corresponding to the process 15 takes time as the passing hierarchy is lower. For example, it takes more time to go through the physical server 13 than through the virtual server 14. Also, it takes more time to go through the rack 12 than through the physical server 13. Furthermore, it takes more time to pass through the data center 11 than through the rack 12.

  The management device 2 is a computer device that determines the arrangement of a plurality of software components that constitute an application applied to the distributed system 1, and includes a processing unit 21, a storage unit 22, an input unit 23, an output unit 24, and a communication unit 25. It is prepared for.

  The processing unit 21 determines the arrangement of a plurality of software components that constitute an application applied to the distributed system 1 based on each data stored in the storage unit 22. Specifically, the initial arrangement is determined and executed, and the rearrangement is determined and executed. The processing unit 21 is realized by, for example, a CPU (Central Processing Unit).

  The storage unit 22 stores communication frequency data, grouping data, mapping data, constraint property data, and the like, and is realized by, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), an HDD (Hard Disk Drive), or the like. .

The communication frequency data is data of communication frequency for each pair of software components, for example, data of communication frequency between software components predicted before the software components are arranged in the distributed system 1.
In this case, the processing unit 21 hierarchically groups software components based on the communication frequency data, and executes initial arrangement of the software components with respect to the distributed system 1 (details will be described later).

Further, the communication frequency data may be data obtained by tabulating actual communication frequencies between software components in the distributed system 1.
In this case, the processing unit 21 hierarchically groups software components based on the communication frequency data, and executes rearrangement of the software components with respect to the distributed system 1 (details will be described later).

  The grouping data is data obtained as a result of grouping the software components so that the processing unit 21 performs communication between the software components in an upper layer as much as possible based on the communication frequency data in the storage unit 22.

  The mapping data is data obtained as a result of the processing unit 21 placing (mapping) software components on the distributed system 1 based on the grouping data in the storage unit 22.

  The constraint property data is data including a property value given to a pair of software components. The constraint property data can also include property values that are assigned to pairs of software component groups. These property values express constraints required for the application applied to the distributed system 1 (details will be described later).

The input unit 23 is an input interface for the user of the management apparatus 2 to input information, and is realized by a keyboard, a mouse, a touch panel, or the like, for example.
The output unit 24 is an output interface for outputting information, and is realized by, for example, a liquid crystal screen display device.
The communication unit 25 is a communication interface for performing communication with an external device such as the distributed system 1.

  Next, an outline of processing in the present embodiment will be described. For the sake of brevity, descriptions that the processing subject is the processing unit 21 of the management apparatus 2 and that the processing unit 21 uses each data in the storage unit 22 are omitted as appropriate.

  The first hierarchy group is a group in which software components with high communication frequency (in some cases, simply described as “components”) are grouped. One first hierarchy group corresponds to one virtual server 14.

  The second hierarchy group is a group in which software components having a medium communication frequency are grouped. That is, a pair of software components having a medium communication frequency is not grouped as a first hierarchy group, but is grouped as a second hierarchy group. One second hierarchy group corresponds to one physical server 13.

  The third hierarchy group is a group in which software components with low communication frequency are grouped. That is, a pair of software components with low communication frequency is not grouped as a first hierarchy group or a second hierarchy group, but is grouped as a third hierarchy group. One third hierarchy group corresponds to one rack 12.

  Although not shown in FIG. 1, as shown in FIG. 4, the fourth hierarchical group is a group in which software components having a lower communication frequency are grouped. That is, a pair of software components having a lower communication frequency is not grouped as a first hierarchy group, a second hierarchy group, or a third hierarchy group, but is grouped as a fourth hierarchy group. One fourth hierarchy group corresponds to one data center 11.

In the present embodiment, software components are hierarchically grouped based on the communication frequency between software components so that a pair of software components having a high communication frequency can communicate with each other as much as possible (reference A in FIG. 1).
Then, based on the grouping result, software components are arranged in the distributed system 1 (symbol B. Mapping).

  The communication frequency data in the storage unit 22 includes not only the communication frequency for each pair of software components, but also the Kth hierarchical group (K is an integer of 1 or more, and in this embodiment, K = 1, 2, 3, 4). The communication frequency data for each pair (between the K-th layer groups) is also included. A method for determining the communication frequency of the pair of the Kth hierarchical group will be described later.

  The grouping process according to the present embodiment can be realized by the following steps 1 to 4. Here, the case where the software components are rearranged instead of the initial arrangement is described as an example.

<Step 1>
Measure the number of communications (communication frequency) between all software components that make up the application for a certain period.
<Step 2>
Based on the communication frequency between the software components (communication frequency data in the storage unit 22), grouping of the software components in the first layer is executed.

<Step 3>
Repeat grouping hierarchically. That is, for the Kth hierarchical group, grouping is sequentially performed by incrementing K.
<Step 4>
The hierarchically grouped software components are arranged in association with the hierarchical distributed system 1.

[Details of Step 1]
The step 1 will be described in detail. As shown in FIG. 2, the number of times of communication between the software components s _n and s _m included in the set S of software components constituting the application is c (s _n , s _m ). To obtain a set C of c = {c (s _n , s _m ) | s _n <s _m ∧s _n ∈ S ∧s _m ∈ S}. For example, the number of communications between s1 and s2 being 18 can be expressed as c (s1, s2) = 18.

[Details of Step 2]
Further, step 2 will be described in detail. First, for all pairs of software components to be communicated, pairs are assigned to the same first hierarchical group in order from the pair with the highest communication frequency, or a predetermined first capacity constraint condition is set. If the pair cannot belong to the same first hierarchy group, the first hierarchy grouping is completed by adding the communication frequency of the pair as the communication frequency between the first hierarchy groups to which each pair belongs. The upper limit value as the first capacity constraint condition is, for example, the number of software components that can be arranged on one virtual server 14.

Then, for all pairs in the first layer group to communicate, in order from the pair with the highest communication frequency, the pairs belong to the same second layer group, or the pairs are set to the same second layer according to a predetermined second capacity constraint condition. When it is not possible to belong to a group, the second layer grouping is completed by adding the communication frequency of the pair as the communication frequency between the second layer groups to which each belongs.
Thereafter, the same grouping process is repeated until the Nth layer (N is an integer of 2 or more).

Specific steps 2-1 to 2-5 for realizing step 2 will be described with reference to FIG.
In step 2-1, the elements c (s _n , s _m ) of the set C are sorted in descending order.
Next, in step 2-2, the top element c (s _n , s _m ) is acquired. That is, the grouping process is executed from the pair with the highest communication count.
Next, in step 2-3, grouping condition determination is performed based on a determination algorithm. This determination algorithm will be described later.

Next, in step 2-4, adding software components _s n, _{s m} to the appropriate group g. At this time, as shown by reference numeral 2-4-1, for example, when c (s ₃ , s ₉ ) is processed, for example, s ₉ may be inserted into the group g ₃ to which s ₃ already belongs. Conceivable. In that case, the communication count c (s _n , s _m ) of the pair is stored.

Further, as indicated by reference numeral 2-4-2, if the pair _{(s n} and _{s m)} is divided into different groups _g o, _{g p,} the group _g o, the number of communications between the _{g p c g (g o,} g _p )) is added (counted) to the communication count c (s _n , s _m ) of the pair. For example, when the pair (s ₈ and s ₁₁ ) is divided into different groups g ₁ and g ₂ , the communication count c (s ₈ and s ₁₁ ) of the pair (s ₈ and s ₁₁ ) is changed between the groups g ₁ and g ₂ . Add to the number of communication times c _g (g ₁ , g ₂ ) to obtain c _g (g ₁ , g ₂ ) = 79.

Next, in step 2-5 (not shown), steps 2-2 to 2-4 are repeatedly executed sequentially from the one with the largest number of communications.
That is, Steps 2-2 to 2-4 are executed for all elements of the set C.
In this way, step 2 can be realized.

Determination algorithm used in step _2-3, s n, for case analysis the _{s m} in the following cases 1 4.

Case _1: s n, if none of _{s m} does not belong to an existing group Case _2: s n, if case belonging to one or the other only existing group of _{s m 3:} s n, each _{s m} is an existing When belonging to different groups Case 4: When both s _n and s _m belong to the same existing group

In case 1, a new group is created, and the component pairs s _n and s _m to be determined are placed in the group. In that case, the communication count c (s _n , s _m ) of the pair is stored.

Case 2, when s _n is assumed to belong to an existing group g _o, can first capacitor satisfying the constraint condition when it does not reach the upper limit number of elements (a) Group g _o, g to insert a _{s m} to _o. In this case, the communication count c (s _n , s _m ) of the determination target component pair is stored. (B) On the other hand, when the number of elements of group g _o has already reached the upper limit value can not satisfy the first capacitor constraint, to create a new group g _p, the s _m to the group g _p insert. Furthermore, the number of communications between the groups _{c g (g o, g p} ) , the added number of communications _c component pair determination target _(s n, s _m).
The same applies to the case of an existing group g _o belongs to within s _n rather s _m.

In case 3, it is assumed that s _n and s _m belong to existing separate groups g _o and g _p . (A) Even if the groups g _o and g _p are merged (merged), the merged group when the element total number of does not exceed the upper limit, it is possible first capacitor satisfying the constraint condition, merges g _o and g _p (integration) is. In this case, the communication count c (s _n , s _m ) of the determination target component pair is stored. (B) On the other hand, when the groups g _o and g _p are merged, when the total number of elements of the merged group exceeds the upper limit value, the first capacity constraint condition cannot be satisfied and the merge is not performed. In this case, the group _g o, the number of communications between the _{g p c g (g o,} g p) , the number of communications _{c (s} n, s _m) of component pairs to be determined is added to (originally independent of each other The groups g _o and g _p are associated with each other using the communication count c (s _n , s _m ).

In case 4, the grouping is left as is, and information on the number of communication times c (s _n , s _m ) of the component pair to be determined is added.

[Details of Step 3]
Further, step 3 will be described in detail. The same grouping as in step 2 is repeatedly executed hierarchically. In other words, since grouping as the first hierarchical group has already been completed in step 2, first, grouping is performed in units of the first hierarchical group to create a second hierarchical group, and grouping is performed in units of the second hierarchical group. The third hierarchy group is created by performing the process until the final Nth hierarchy group is created.

  Regarding step 3, steps 3-1 to 3-5, which are specific procedures for performing the next grouping based on the grouping result of the first hierarchy group, will be described with reference to FIG. Instead of the predetermined first capacity constraint condition used in step 2, a predetermined second capacity constraint condition that is the upper limit value of the number of first hierarchy groups that can belong to the second hierarchy group in step 3 is used. Is used. The upper limit value as the predetermined second capacity constraint condition is, for example, the number of virtual servers 14 that can be mounted on one physical server 13. The second capacity constraint condition can be generally used as an upper limit value of the number of (K-1) layer groups that can belong to the Kth layer group (the 0th layer group is regarded as a software component). ).

In step 3-1, the elements c _g (g _n , g _m ) of the set C _g are sorted in descending order.
Next, in step 3-2, the leading element c _g (g _n , g _m ) is acquired. That is, the grouping process is executed from the pair with the highest communication count.
Next, in step 3-3, grouping condition determination is performed based on a determination algorithm. Since this determination algorithm is the same as the determination algorithm in step 2, description thereof is omitted.

Next, in step 3-4, g _n and g _m are added to an appropriate group (second hierarchical group). At this time, as in step 2-4, when the pair (g _n , g _m ) is divided into different groups, the communication count c _g (g _o , g _p ) of the pair is added to the communication count between the groups. .

Next, in step 3-5 (not shown), steps 3-2 to 3-4 are repeatedly executed. In other words, performing step 3-2～3-4 for all elements in the set _{C g.}

  Such step 3 is performed until the final hierarchy group is created. When the final hierarchy group is the fourth hierarchy group, the state where the hierarchical grouping is completed is as shown in FIG. 4B. The grouping result data is stored as grouping data in the storage unit 22.

[Details of Step 4]
Software components in the same first hierarchy group are arranged in the same virtual server 14. Further, the first hierarchy group in the same second hierarchy group is arranged in the same physical server 13. Further, the second hierarchy group in the same third hierarchy group is arranged in the same rack 12. Further, the third hierarchy group in the same fourth hierarchy group is arranged in the same data center 11. The software component arrangement data is stored as mapping data in the storage unit 22.

(Maximum distance constraint)
If the management apparatus 2 of this embodiment performs the process of said step 1-4, communication of the software components which comprise a pair with high communication frequency, and the Kth hierarchy group which comprises a pair with high communication frequency Software components can be arranged so that communication between software components belonging to each route passes through an upper layer, communication in lower layers can be reduced, and communication delay in the distributed system 1 can be reduced. This means that pairs with low communication frequency tend to be placed (distant) in separate groups.

  However, depending on the requirements of the application applied to the distributed system 1, it is necessary to set a constraint (maximum distance constraint) that a pair with low communication frequency is placed (close) in the same group. For example, even if the real-time requirements for processing related to a plurality of software components are strict and the communication frequency between the software components is low, the software components need to be arranged on the same physical server 13.

  In this case, the constraint condition is set so as to allow a certain amount of communication delay (for example, a delay of 1 msec or less). Specifically, a maximum distance constraint property value p (p is an integer equal to or greater than 1) is assigned to a pair of software components and a pair of a Kth hierarchical group. Here, giving the maximum distance constraint property value p to the pair of the Kth hierarchy group means giving the maximum distance constraint property value p to the pair of software components belonging to each of the Kth hierarchy group. Specifically, the maximum distance constraint property value p is a value set between software components. Therefore, the software component and the Kth hierarchy group are collectively referred to as “nodes”, and the description will be continued assuming that the maximum distance constraint property value p is given to a pair of nodes.

  The maximum distance constraint property value p is decremented each time the hierarchy on which the grouping process described in steps 2 and 3 is performed is switched to a hierarchy one level lower. When the maximum distance constraint property value p reaches the lower limit value (p = 1), the corresponding node pair is assigned to the same group with the highest priority regardless of the communication frequency. The maximum distance constraint property value p is stored as a property value of constraint property data (here, maximum distance constraint property data) in the storage unit 22 for each pair of nodes.

The specific steps A-1 to A-5 will be described with reference to FIG.
In step A-1, for example, the operator of the management apparatus 2 inputs the content of the maximum distance constraint (eg, delay of 1 msec or less) and designates the constraint.
Next, in step A-2, the designated constraint is converted so as to be suitable for the grouping process. For example, the maximum distance constraint of a delay of 1 msec or less is converted into the content of belonging to the same group at the third layer or higher (lower level). At this time, the maximum distance constraint property value p is 3. FIG. 5 shows that p = 3 (shown as a boxed numerical value) is given as an initial value to a specific pair of nodes n ₁ and n ₈ , and p is the constraint property data. Saved.

  Next, in step A-3, p is decremented every time the hierarchy in which the grouping process described in steps 2 and 3 is performed is switched to the next lower hierarchy (operation of property values). In FIG. 5, the maximum distance constraint property value p having an initial value of 3 is 2 when the first layer grouping (corresponding to step 2) is completed, and is 1 when the second layer grouping (corresponding to step 3) is completed. This is illustrated.

Next, in step A-4, when the elements c _g (g _n , g _m ) of the set C _g are sorted in descending order in step 3-1 (FIG. 4) (if the initial value of p is 1, the step 2) (when sorting the elements c (s _n , s _m ) of the set C in descending order) in FIG. 2A (FIG. 3), the pair of nodes with p = 1 is arranged at the top (priority). FIG. 5 shows that a pair of nodes n ₁ and n ₈ with p = 1 is arranged at the head.

Next, in step A-5, as in steps 2 and 3, grouping condition determination is performed from the top element based on the determination algorithm. At this time, the pair of nodes for which p = 1 is determined to belong to the same group, regardless of whether the cases 1 to 4 have already been described. FIG. 5 shows that a pair of nodes n ₁ and n ₈ arranged at the head is arranged in the group g ₁ .

  The processes of steps 1 to 4 are executed in a manner in which the maximum distance constraint property value as described above is introduced. Then, it is possible to optimize the arrangement of software components while satisfying the maximum distance constraint.

If the processes in steps A-1 to A-5 are thoroughly performed, the maximum distance constraint may not be satisfied simultaneously for a plurality of software components. For example, as shown in FIG. 6A, consider a case where a pair with a node n ₈ is formed and there are a large number of nodes n _{1 to} n ₆ having a maximum distance constraint property value p of 1. In this case, it is desired to form one group with the nodes n _{1 to} n ₆ and n ₈ , but the capacity (capacity) is exceeded because the first capacity constraint condition or the second capacity constraint condition must be satisfied. It cannot be formed.

In this case, the node as shown in FIG. 6 (b) (here, although the node n _8, which may be a other node) to scale out. Then, a plurality of groups g ₁ and g ₂ satisfying the first capacity constraint condition or the second capacity constraint condition are formed, and the nodes n _{1 to} n ₆ are set so as to satisfy the maximum distance constraint in each of the groups g ₁ and g _2. What is necessary is just to arrange.

  If the capacity (capacity) is exceeded because the first capacity constraint condition or the second capacity constraint condition must be satisfied and a group cannot be formed, an alarm notification is sent to the administrator in a predetermined manner. Also good.

(Minimum distance constraint)
Further, according to the optimization of the arrangement in the software component of the present embodiment, pairs having a high communication frequency tend to be arranged (close) in the same group. However, depending on the requirements of the application applied to the distributed system 1, it is necessary to set a constraint (minimum distance constraint) in which pairs with high communication frequency are arranged (distantly) in different groups. For example, in preparation for a catastrophic disaster, it is necessary to arrange a plurality of software components having high communication frequency in different racks.

  In this case, the constraint condition is set so as to allow a certain level of availability (for example, redundancy to withstand rack power interruption). Specifically, similarly to the maximum distance constraint property value p, a minimum distance constraint property value q (q is an integer of 0 or more) is assigned to a pair of nodes.

  The minimum distance constraint property value q is decremented each time the hierarchy in which the grouping process described in steps 2 and 3 is performed is switched to a hierarchy one level lower. Also, a pair whose minimum distance constraint property value q does not reach the lower limit (q = 0) is not allowed to belong to the same group regardless of the communication frequency. The minimum distance constraint property value q is stored as the property value of the constraint property data (here, the minimum distance constraint property data) in the storage unit 22 for each pair of nodes.

The specific steps B-1 to B-5 will be described with reference to FIG.
In Step B-1, for example, the operator of the management apparatus 2 inputs the content of the minimum distance constraint (eg, redundancy that can withstand rack power interruption) and specifies the constraint.
Next, in step B-2, the designated constraint is converted so as to be suitable for the grouping process. For example, the maximum distance constraint of redundancy that can withstand rack power interruption is converted to the content of not belonging to the same group in the second hierarchy or lower (upper). At this time, the minimum distance constraint property value q is 2. FIG. 7 illustrates that q = 2 (shown as a triangular boxed numerical value) is given as an initial value to a specific pair of nodes n ₁ and n ₈ , where q is the constraint property data. Saved.

  Next, in step B-3, q is decremented each time the hierarchy in which the grouping process described in steps 2 and 3 is performed is switched to the next lower hierarchy (property value operation). In FIG. 7, the minimum distance constraint property value q having an initial value of 2 is 1 when the first layer grouping (corresponding to step 2) is completed, and is 0 when the second layer grouping (corresponding to step 3) is completed. This is illustrated. The minimum distance constraint is canceled for a pair of nodes where q = 0.

Next, in step B-4, step 3-1 elements _{c g (g n, g m} ) of the set _{C g} by (Fig. 4) descending sort (or, step 2-1 of set C by (3) The descending order sort similar to the descending order sort of the element c (s _n , s _m ) is performed.

Next, in step B-5, as in steps 2 and 3, grouping condition determination is performed from the top element based on the determination algorithm. At this time, it is determined that a pair of nodes for which q = 0 is not satisfied (q is 1 or more) belongs to different groups, whether or not the cases 1 to 4 have already been described. FIG. 5 shows that for a pair of determination target nodes n ₁ and n ₈ (q = 1), node n ₁ is arranged in group g ₁ and node n ₈ is arranged in group g ₂ . . For a pair of nodes where q = 0, it is determined which of cases 1 to 4 is applicable.

The processes of steps 1 to 4 are executed in a manner in which the minimum distance constraint property value as described above is introduced. Then, the arrangement of software components can be optimized while satisfying the minimum distance constraint.
If the minimum distance constraint cannot be satisfied due to various circumstances such as the requirements of a plurality of software components cannot be satisfied at the same time, an alarm notification may be sent to the administrator in a predetermined manner.

There is a possibility that it may not be possible to determine in which group the node is to be arranged because the maximum distance constraint and the minimum distance constraint described above compete with each other. Therefore, priority can be set for the maximum distance constraint property data and the minimum distance constraint property data, and the constraint property data including the priority can be stored in the storage unit 22. When the maximum distance constraint and the minimum distance constraint compete, the processing unit 21 prioritizes a constraint having a larger priority value and determines the arrangement of software components. The priority value can be designated by the administrator of the management apparatus 2, for example. For example, for an application that places importance on service continuity, the priority of the minimum distance constraint property data is set higher than the priority of the maximum distance constraint property data.
If the maximum distance constraint and the minimum distance constraint do not conflict, or if the maximum distance constraint and the minimum distance constraint do not conflict, place the software components so that both the maximum distance constraint and the minimum distance constraint are satisfied. Can be determined.

As described above, according to the present embodiment, by setting the maximum distance constraint property value to a specific software component pair and a specific K-th layer group pair, the communication frequency is low and inherently different. Even a pair of software components and a pair of a Kth hierarchical group that are arranged in a group can be intentionally assigned to the same group with respect to application requirements.
Therefore, it is possible to determine an appropriate arrangement of a plurality of software components that constitute an application while satisfying a restriction required for the application applied to the distributed system having a plurality of hierarchies.

  In addition, by scaling out the software component or the K-th layer group that constitutes the pair whose maximum distance constraint property value has reached the lower limit value, the group to which these software components or the K-th layer group belongs is surely formed. Can do.

In addition, by setting a minimum distance constraint property value for a specific software component pair and a specific K-th layer group pair, a communication component pair that is frequently placed in the same group with high communication frequency , And even a pair of the K-th hierarchical group can intentionally belong to different groups with respect to application requirements.
Therefore, it is possible to determine an appropriate arrangement of a plurality of software components that constitute an application while satisfying a restriction required for the application applied to the distributed system having a plurality of hierarchies.

  In addition, by including priority in the minimum distance constraint property data and the maximum distance constraint property data, even if the minimum distance constraint and the maximum distance constraint compete, at least one of the constraints is satisfied, and the placement destination of the software component is determined. It can be determined with certainty.

  In the above-described embodiment, the software components are hierarchically grouped based on the communication frequency data obtained by counting the actual communication frequencies between the software components in the distributed system 1, and the software components are rearranged in the distributed system 1. It was supposed to be. However, other than that, based on the data of the communication frequency between the software components predicted before the software components are arranged in the distributed system 1, the software components are grouped hierarchically and the initial arrangement of the software components in the distributed system 1 is executed. You may make it do. That is, according to the method of the present embodiment, it is possible to determine the initial arrangement of the software components with respect to the distributed system 1 and then perform the rearrangement of the software components following the operation status.

(Modification)
Although description of this embodiment is finished above, the aspect of the present invention is not limited to these.
(1): For example, in the present embodiment, the case where the distributed system 1 has four hierarchies has been described. However, the present invention is not limited to this, and the case where there are two hierarchies, three hierarchies, or five or more hierarchies. Is also applicable.

  (2): The correspondence between software components and processes is not limited to one-to-one correspondence, but one-to-N correspondence (N is an integer of 2 or more), N-to-one correspondence (N is an integer of 2 or more) There may be.

  (3): The first capacity constraint condition and the second capacity constraint condition are not limited to numerical values based on the server capacity limit, and may be information on software components that cannot be grouped into the same group for a predetermined reason.

  (4): The software component grouping process is not limited to the management device 2 outside the distributed system 1, but may be another type such as a computer device in the data center 11 in the distributed system 1. It may be a device.

A technique obtained by appropriately combining various techniques described in the present embodiment can also be realized.
The software described in this embodiment can be realized as hardware, and the hardware can also be realized as software.
In addition, hardware, software, flowcharts, and the like can be changed as appropriate without departing from the spirit of the present invention.

DESCRIPTION OF SYMBOLS 1 Distributed system 2 Management apparatus 3 Software component 11 Data center 12 Rack 13 Physical server 14 Virtual server 15 Process 21 Processing part 22 Storage part 23 Input part 24 Output part 25 Communication part

Claims (6)

The first hierarchy of the most upper hierarchy, an application (in N 2 or more integer) N-th layer which is a most lower layer is applied to a distributed system having a device that acts as one of the hierarchy to A management device that determines the arrangement of a plurality of software components to be configured,
Before SL software component is arranged in the first hierarchy,
The communication between the software components is realized through the same device or between the devices in the Nth layer,
The communication between the software components, the hierarchy indicated by the device through which the consuming more time is subordinate,
The software component corresponding to the device functioning as the Kth layer (K = 1, 2,..., N) can form a Kth layer group by performing a grouping process.
The management device includes a storage unit and a processing unit,
The storage unit
Communication frequency data indicating a first communication frequency for each pair of software components to be communicated, and a second communication frequency for each pair of the K-th layer group, determined based on the first communication frequency; ,
Predetermined for each pair of a particular of the software components, and stores a, and maximum distance constraints property data indicating the maximum distance constraint property value predetermined for each pair of a particular said first K hierarchical group,
The processor is
With reference to the communication frequency data in the storage unit, the software component and the K-th layer group processing are performed in order from the first layer to the N-th layer according to a predetermined capacity constraint condition in order from the pair with the highest communication frequency . when you run repeatedly until the hierarchy,
Wherein each time the hierarchy grouping processing is performed is switched to one lower hierarchy, the decrements the maximum distance constraint property values, Ru to belong pairs where the maximum distance constraint property value has reached the lower limit value in the same group,
A management device characterized by that. The processor is
If a pair whose maximum distance constraint property value has reached the lower limit value belongs to the same group and exceeds the capacity of the group, the software component or the K-th layer group constituting the pair is scaled out.
The management apparatus according to claim 1. The first hierarchy of the most upper hierarchy, an application (in N 2 or more integer) N-th layer which is a most lower layer is applied to a distributed system having a device that acts as one of the hierarchy to A management device that determines the arrangement of a plurality of software components to be configured,
Before SL software component is arranged in the first hierarchy,
The communication between the software components is realized through the same device or between the devices in the Nth layer,
The communication between the software components, the hierarchy indicated by the device through which the consuming more time is subordinate,
The software component corresponding to the device functioning as the Kth layer (K = 1, 2,..., N) can form a Kth layer group by performing a grouping process.
The management device includes a storage unit and a processing unit,
The storage unit
Communication frequency data indicating a first communication frequency for each pair of software components to be communicated, and a second communication frequency for each pair of the K-th layer group, determined based on the first communication frequency; ,
Pre-determined for each specific pair of software components , and minimum distance constraint property data indicating a minimum distance constraint property value predetermined for each specific pair of the Kth hierarchical group ,
The processor is
With reference to the communication frequency data in the storage unit, the software component and the K-th layer group processing are performed in order from the first layer to the N-th layer according to a predetermined capacity constraint condition in order from the pair with the highest communication frequency . when you run repeatedly until the hierarchy,
Wherein each time the hierarchy grouping processing is performed is switched to one lower hierarchy, the decrements the minimum distance constraint property value, said minimum distance constraint property value is not such to belong pair does not reach the lower limit value in the same group,
A management device characterized by that. The storage unit
Predetermined for each pair of a particular of the software components, and further stores the maximum distance constraint property data indicating the maximum distance constraint property value predetermined for each pair of a particular said first K hierarchical group,
The grouping processing based on the minimum distance constraint property data, and the priority to determine whether to give priority to any of the grouping processing based on the maximum distance constraint property data, the minimum distance constraints property data and the maximum distance Store it in the constraint property data,
The processing unit decrements the maximum distance constraint property value each time the hierarchy in which the grouping process is performed is switched to a lower hierarchy, and sets pairs in which the maximum distance constraint property value has reached a lower limit to the same group. To belong,
With reference to the priority and the maximum distance priority constrained property data of the minimum distance constraint property data, the grouping processing based on the minimum distance constraint property data, and, of the grouping processing based on the maximum distance constraint Property Data The management apparatus according to claim 3, wherein at least one of them is executed. The first hierarchy of the most upper hierarchy, an application (in N 2 or more integer) N-th layer which is a most lower layer is applied to a distributed system having a device that acts as one of the hierarchy to A software component grouping method by a management device that determines the arrangement of a plurality of software components to be configured,
Before SL software component is arranged in the first hierarchy,
The communication between the software components is realized through the same device or between the devices in the Nth layer,
The communication between the software components, the hierarchy indicated by the device through which the consuming more time is subordinate,
The software component corresponding to the device functioning as the Kth layer (K = 1, 2,..., N) can form a Kth layer group by performing a grouping process.
The storage unit of the management device is
Communication frequency data indicating a first communication frequency for each pair of software components to be communicated, and a second communication frequency for each pair of the K-th layer group, determined based on the first communication frequency; ,
Predetermined for each pair of a particular of the software components, and stores a, and maximum distance constraints property data indicating the maximum distance constraint property value predetermined for each pair of a particular said first K hierarchical group,
The management device
With reference to the communication frequency data in the storage unit, the software component and the K-th layer group processing are performed in order from the first layer to the N-th layer according to a predetermined capacity constraint condition in order from the pair with the highest communication frequency . when you run repeatedly until the hierarchy,
Each time the hierarchy in which the grouping process is performed is switched to a lower hierarchy, the maximum distance constraint property value is decremented, and the pair whose maximum distance constraint property value has reached the lower limit value belongs to the same group.
A software component grouping method characterized by the above. The first hierarchy of the most upper hierarchy, an application (in N 2 or more integer) N-th layer which is a most lower layer is applied to a distributed system having a device that acts as one of the hierarchy to A software component grouping method by a management device that determines the arrangement of a plurality of software components to be configured,
Before SL software component is arranged in the first hierarchy,
The communication between the software components is realized through the same device or between the devices in the Nth layer,
The communication between the software components, the hierarchy indicated by the device through which the consuming more time is subordinate,
The software component corresponding to the device functioning as the Kth layer (K = 1, 2,..., N) can form a Kth layer group by performing a grouping process.
The storage unit of the management device is
Communication frequency data indicating a first communication frequency for each pair of software components to be communicated, and a second communication frequency for each pair of the K-th layer group, determined based on the first communication frequency; ,
Pre-determined for each specific pair of software components , and minimum distance constraint property data indicating a minimum distance constraint property value predetermined for each specific pair of the Kth hierarchical group ,
The management device
With reference to the communication frequency data in the storage unit, the software component and the K-th layer group processing are performed in order from the first layer to the N-th layer according to a predetermined capacity constraint condition in order from the pair with the highest communication frequency . when you run repeatedly until the hierarchy,
Each time the hierarchy in which the grouping process is performed is switched to one lower hierarchy, the minimum distance constraint property value is decremented, and a pair whose minimum distance constraint property value does not reach the lower limit value does not belong to the same group.
A software component grouping method characterized by the above.

Images