FR3013473A1 - METHOD AND DEVICE FOR CONTROLLING THE DEPLOYMENT OF SOFTWARE ENTIRE SETS IN A HIGH PERFORMANCE COMPUTER - Google Patents

Abstract

A method is provided for controlling the deployment of software entities of different types in a high performance computer (CHP) comprising a main management node (NGP) coupled to at least two auxiliary management nodes (NGA1-NGAN) each coupled to a service node group (NS11-NSNM (N)). This method comprises a step (i) in which storage units (BP, BA1-BAN) are stored with sets of at least two different software entities each associated with at least one auxiliary management node (NGA1) or node of the service (N11) in which they are to be deployed, and a step (ii) in which the auxiliary management nodes (NGA1-NGAN) are transmitted via the main management node (NGP) and to the service nodes (NS11- NSNM (N)), via their respective auxiliary management nodes (NGA1-NGAN), the sets respectively associated with them for local deployment.

Description

The invention relates to the deployment of software entities of different types in "high performance" type calculators (or sucalculators). Some high performance computers include a main management node which is coupled to auxiliary management nodes, themselves coupled to groups of service nodes. Here, the term "service node" means electronic equipment in which software entities (defining resources) must be deployed (or set up) and then configured using configuration rules. This configuration of the resources of the service nodes, which is intended to meet the specific needs of their respective users, is performed by a configuration tool (such as the one called "Kconf"), in a mode that is generally of the client / type. server (where the client is a service node). Here is meant by "resource" a "configurable service" useful to a computer or an application running in a computer. It may for example be a service (nfs or postgresql), or a file system (or "file system") shared between nodes (service) of a group called high availability (or HA ) on a storage array (remote or local), or an Internet Protocol (IF) address. The set of resources of a (service) node is sometimes called a "technical state". This technical state is the state in which the node must be configured to perform its role, that is to say the use that we want to make him in a computer. There are usually several typical roles, such as management (or MNGT), work (or MWS), calculation (or COMPUTE), connection (or LOGIN), or storage (or 10). To control the deployment of all software entities in all the ancillary management nodes and service nodes of a high-performance computer, a deployment tool (such as "ksis") is used. This deployment being currently software entity after software entity to each auxiliary management node and each service node, the time it requires can vary very strongly from one computer to another according to the number of nodes that it understands and depending on the number of types of software entities to deploy. When the high performance computer only has a few hundred nodes, the deployment time remains reasonable. On the other hand, when the high performance computer comprises several thousand or even tens of thousands of nodes, the deployment time becomes very important, which can be a real inconvenience for its administrator and delays its use.

The invention therefore aims to improve the situation, and in particular to allow a significant reduction in the time of deployment of the software entities in a high performance computer. It proposes for this purpose a method for controlling the deployment of software entities of different types in a high performance computer comprising a main management node coupled to at least two auxiliary management nodes each coupled to a group of service nodes. and comprising: - a step (i) in which sets of at least two different software entities each associated with at least one auxiliary management node or service node in which they are to be deployed are stored in storage means, and a step (ii) in which the sets associated respectively with them for deployment are transmitted to the auxiliary management nodes, via the main management node, and to the service nodes, via their respective auxiliary management nodes. local. The method according to the invention can comprise other characteristics that can be taken separately or in combination, and in particular: a set associated with an auxiliary management node and transmitted to the latter can be modified, as a function of the service nodes which are coupled to this auxiliary management node, and at least some of the sets associated with service nodes coupled to an auxiliary management node can be defined according to the modified set associated therewith; certain sets can be defined by combining selected software entities contained in at least two other sets; the software entities may have types that are selected from (at least) a so-called image type, a so-called virtual type and a so-called diskless type; in main storage means, associated with the main management node, each set associated with at least one auxiliary management node can be stored, and auxiliary storage means, associated respectively with the auxiliary management nodes, can be stored in sets associated respectively with the service nodes that are coupled to these auxiliary management nodes; in the presence of service nodes of different types, it is possible to associate different sets with subgroups of a group respectively comprising service nodes of different types; the sets may comprise at least one software entity of a first type and / or at least one software entity of a second type and / or at least one software entity of a third type. The invention also proposes a computer program product comprising a set of instructions which, when executed by processing means, is suitable for implementing a control method of the type of that presented above for controlling the deployment of software entities of different types in a high performance computer. The invention also proposes a control device capable of being coupled to a high-performance computer comprising a main management node coupled to at least two auxiliary management nodes, each coupled to a group of service nodes, and comprising: means interface devices arranged to store in storage means sets of at least two different software entities each associated with at least one auxiliary management node or service node in which they are to be deployed, and control means arranged to triggering transmission to the ancillary management nodes, via the main management node, and to the service nodes, via their respective auxiliary management nodes, sets respectively associated with them for local deployment. The device according to the invention can comprise other characteristics that can be taken separately or in combination, and in particular: it can comprise processing means arranged to modify a set associated with an auxiliary management node and transmitted to the latter, in function of the service nodes coupled to this auxiliary management node, and to define at least some sets associated with service nodes coupled to an auxiliary management node according to the modified set associated therewith; the processing means may be arranged to define certain sets by combining selected software entities contained in at least two other sets. The invention also proposes a high performance computer comprising a main management node, coupled to at least two auxiliary management nodes each coupled to a group of service nodes, and a control device of the type of that presented above. Other features and advantages of the invention will appear on examining the detailed description below, and the accompanying drawings, in which: - Figure 1 schematically and functionally illustrates a high performance computer equipped with an example embodiment of a control device according to the invention, - Figure 2 schematically illustrates an example set of software entities associated with a service node, and - Figure 3 schematically illustrates an example of set of software entities defined from the set of Figure 2 for a service node. The object of the invention is in particular to propose a control method, and an associated control device D, for a high performance computer CHP comprising a main management node NGP coupled to at least two auxiliary management nodes NGAi, themselves coupled. respectively to groups Gi NSii service nodes. FIG. 1 schematically illustrates a nonlimiting example of a computer (high performance) CHP. In this example, the calculator CHP comprises N auxiliary management nodes NGAi (i = 1 to N, with N 3) io coupled to a main management node NGP and respectively to N groups Gi of M (i) service nodes NSq ( with j = 1 to M (i)). For example, each group Gi is high availability (or HA). For example, N is equal to 100 and M (i) is equal to 50 regardless of the group Gi considered (and therefore whatever the value of the index i). But the number of service nodes NSii could vary from one group Gi to the other Gr. On the other hand, the number N of groups Gi can take any value greater than or equal to one (1). Similarly, the number M (i) of NSii service nodes of a group Gi can take any value greater than or equal to two (2). Note that a group Gi may be subdivided into SGik subgroups 20 including NSii service nodes of different types. Software entities (defining resources) must be deployed (or implemented) in each NGAi auxiliary management node and each NSq service node, and then configured by an OC configuration tool using configuration rules. The resources of a service node NSq of a group Gi are generally shared with the other service nodes NSir (j 'j) of this group Gi, under the control of a high availability software (or HA (" High Availability ")). In the nonlimiting example illustrated in FIG. 1, the configuration tool OC is part of the main management node NGP. But it could be external to him. For example, it could be implemented in a server dedicated to configuration. It will be noted that the configuration commands that are generated by the configuration tool OC can be transmitted to the auxiliary management nodes NGAi and the service nodes NSii by at least one configuration server (then part of the CHP computer, for example) . Each configuration server then constitutes a service associated with at least one node Nq in order to allow the configuration of its resources. For example, several configuration servers may be distributed in different NSii service nodes and / or different NGAi auxiliary management nodes. The resources (and therefore the software entities) can be of any type as long as they are configurable services that are useful to the CHP calculator or to an application running in this CHP calculator. For example, in the case of an OC configuration tool such as kconf, which uses the "Puppet" application, the resources are of limited types and defined by the Puppet definition language. Thus, it may for example be a service (nfs or postgresql), or a file system (or "file system") 15 shared between NSii service nodes of a group Gi on a bay of storage, or an Internet Protocol (IF) address, or static or dynamic configuration files, or user managers, or script executors. The calculator CHP also comprises an operating system 20 (or "operating system") SO responsible for providing the interface between software applications (or "software") and hardware (or "hardware") that includes . For example, the operating system can be of type Linux. As indicated above, the invention proposes a method for enabling the control of the deployment of software entities of different types in the high performance computer CHP. This method comprises first (i) and second (ii) steps that can be implemented by a control device D according to the invention. In the nonlimiting example illustrated in FIG. 1, the control device D forms part of the computer CHP, and more specifically of its main management node NGP as well as possibly of its auxiliary management nodes NGAi. But this is not obligatory. It could indeed be equipment that is external to the CHP computer but accessible by the latter (CHP), for example because of a computer connection. Therefore, the control device D can be realized, either in the form of software modules (or computer, or "software"); it is then in the presence of a computer program product comprising a set of instructions which, when executed by processing means such as electronic circuits (or "hardware"), is adapted to implement the method control system, either in the form of a combination of software modules and electronic circuits. For example, the control device D may be a sub-part of a deployment tool (such as for example ksis) or this deployment tool. During the first step (i) of the method according to the invention, sets of at least two different software entities each associated with at least one auxiliary management node NGAi or service node NSq are stored in storage means BP, BAi. in which they must be deployed. For example, the software entities that are combined within EEL sets may have a first type called image, or a second type called virtual or a third type called diskless (or "diskless"). A software entity of the first type (image-ELI) is intended to support the hardware and the software application of a service node NSii. A software entity of the second type (virtual - ELV) is intended to support a virtual machine of a service node NSq. A third type (diskless - ELSD) software entity is intended to support the hardware and software application of an NSq service node that does not have local local storage (or physical disk). It is important to note that all software entities are software module photographs that will be used by different objects according to their respective types. In the case presented above, the EEL sets may comprise at least one software entity of a first type (image - ELI) and / or at least one software entity of a second type (virtual - ELV) and / or at least one software entity of a third type (without disk - ELSD). In the presence of NSq service nodes of different types, it is possible to associate different EEL sets with SGik subgroups of a group Gi respectively comprising NSq service nodes of different types. An exemplary EEL set of software entities associated with a service node NSii is illustrated not limitatively in FIG. 2. In this example, the set EEL comprises in combination a software entity of the first type (image) ELI A, a software entity of the second type (virtual) ELV1, another software entity of the second type (virtual) ELV2, and a software entity of the third type (without disk) ELSD 5. Of course, this is only an example purely illustrative. Numerous other combinations of software entities comprising at least two software entities of the same type and / or of different types can indeed be defined. An EEL set may be either defined by a specialist, then provided to the main management node NGP using interface means MI of the control device D, and then transmitted to at least one auxiliary management node NGAi by triggering by control means MC of the control device D, is defined by processing means MT i of the control device D from at least one other previously defined set. During the second step (ii) of the method according to the invention, on the one hand, to the NGAi auxiliary management nodes, via the main management node NGP, the associated EEL sets are transmitted to them for deployment. local, and on the other hand, to the NSii service nodes, via their respective NGAi auxiliary management nodes, the sets respectively associated with them for local deployment. It will be understood that by transmitting sets of EEL software entities one after the other, instead of transmitting the software entities one after the other, the time required for the deployment of the software entities in all the management nodes is very significantly reduced. NGAi auxiliaries and all NSii service nodes. It is important to note that this transmission can be done either in a single substep or in two substeps. In the first alternative (single substep), the control means MC of the device D triggers the transmission to the auxiliary management nodes NGAi, via the main management node NGP, of the sets respectively associated with them and which are associated with the nodes NSq service of Gi groups that they manage respectively. In this case, each auxiliary management node NGAi is responsible for retransmitting to NSq service nodes of the group Gi that it manages the received sets associated with them respectively. In the second alternative (two substeps), the control means MC of the device D triggers the transmission to the auxiliary management nodes NGAi, via the main management node NGP, sets respectively associated with them. Then, each auxiliary management node NGAi comprises control means MC responsible for triggering the transmission to the service nodes NSq of the group Gi that it manages the sets respectively associated with them and which have been defined locally and not at the level of the main management node NGP.

Preferably, and as illustrated without limitation in FIG. 1, the storage means comprise main storage means BP and N auxiliary storage means BAi. The main storage means BP are associated with the main management node NGP. It will be noted that in the example illustrated in non-limiting manner in FIG. 1, the main storage means BP form part of the main management node NGP. But this is not obligatory. They could indeed be part of a device that is external to the main management node NGP but accessible by the latter (NGP). The interface means MI of the (control) device D are arranged to store in these primary storage means BP each set associated with at least one auxiliary management node NGAi. Preferably, these primary storage means BP are arranged in the form of a database in which are stored all the EEL sets, which are transmitted to the NGAi auxiliary management nodes and to the NSq service nodes via the management node. NGP, and the associated NGAi and NSq node IDs. These identifiers may, for example, be IP addresses. The main storage means BP can also store all the information 'o relating to the specificities of each of the auxiliary management nodes NGAi, including the materials that constitute them and their current states (ie "hidden" or "in operation" "). The main storage means BP can also store all the data representative of the operations performed locally by a management tool on each software entity EEL they store ("create", "delete", "list", "deploy", "export" Or "import", for example). The N auxiliary storage means BAi are associated respectively with the N auxiliary management nodes NGAi. It will be noted that in the example illustrated nonlimitingly in FIG. 1, the N auxiliary storage means BAi are respectively part of the N auxiliary management nodes NGAi. But this is not obligatory. They could indeed be part of one or more equipment external to the NGAi auxiliary management nodes but accessible by them (NGAi).

The interface means MI of the (control) device D are arranged to store in an auxiliary storage means BAi associated with an auxiliary management node NGAi each set associated thereto (NGAi) and each set associated with one of the less NSq service nodes of the Gi group that it manages.

Preferably, each auxiliary storage means BAi is arranged in the form of a database in which all the EEL sets are stored, which are transmitted to the associated auxiliary management node NGAi and to the service nodes NSii via this management node. NGAi auxiliary, and the identifiers of NSii service nodes associated with them. These identifiers may, for example, be IF addresses. Each auxiliary storage means NGAi can also store all the information relating to the specificities of each of the service nodes NSii of the group Gi that it manages, and in particular the materials that constitute them and their current states (namely "out of 'group', 'hidden' or 'in operation').

Each auxiliary storage means NGAi can also store all the data representative of the operations performed locally by a management tool on each software entity EEL that it stores. It will be noted that it is possible to modify locally a set which has been associated with an auxiliary management node NGAi and transmitted to the latter (NGAi), as a function of the service nodes NSq which are coupled to this auxiliary management node NGAi. This modification is then performed by the processing means MT i which are associated with the auxiliary management node NGAi as a function of the knowledge they have of the service nodes NSq of the group Gi managed by their auxiliary management node NGAi, and which results information stored in the auxiliary storage means BAi associated therewith (NGAi). In this case, it is possible (the processing means MT) to define at least some sets that are associated with NSq service nodes coupled to an auxiliary management node NGAi as a function of a modified set. which is associated with the latter (NGAi). For example, a modification of an EEL set may consist of an addition and / or deletion of one or more software entities. These types of modifications are intended, for example, to simplify the updates or to replace one software entity with another to change the working environment. Note also that one can locally modify a set that has been associated with at least one NSq service node coupled to an auxiliary management node NGAi. This modification is then performed by the processing means MT i which are associated with the auxiliary management node NGAi as a function of the knowledge they have of the service nodes NSq of the group Gi managed by their auxiliary management node NGAi, and which results information stored in the auxiliary storage means BAi associated therewith (NGAi). For example, a modification of an EEL set may consist of an addition and / or deletion of one or more software entities (for the same reasons as mentioned in the previous paragraph). Note that an addition may consist of a duplication of a software entity that is already contained in the set concerned. When an EEL set comprises several identical EL software entities (due to a local duplication), as in the example shown non-limitatively in FIG. 3, these identical EL software entities are configured in different ways by the configuration tool OC to respectively concern different functions of each associated NSq service node. In this example, we started with the set EEL of FIG. 2 to define a set EEL 'modified which comprises in combination a software entity of the first type (picture) ELI A, three software entities of the second type (virtual) ELV1, identical, another software entity of the second type (virtual) ELV2, and two software entities of the third type (without disk) ELSD 5, identical. Of course, this is only a purely illustrative example of modification of an EEL set. Instead of duplicating software entities in a set, it is possible to associate with the software entity to duplicate a dedicated field defining this p number P times, and to associate this dedicated field with this software entity. This makes it possible to limit the weight of the EEL sets that must be transmitted to the NSq service nodes. It will also be noted that one (the processing means MT) can define certain sets by combining certain selected EL software entities that are contained in at least two other sets. This makes it possible to create new sets adapted to at least one NGAi auxiliary management node or at least one NSq service node from software entities EL associated with at least one similar NGAi auxiliary management node or at least one node of similar NSii service. This operation is possible because all the manipulations performed on each software entity associated with a node are stored in an NGAi auxiliary storage means associated with this node and can therefore be used to design a set to be associated with at least one node. similar. It should also be noted that in order to deploy the software entities EL, it is possible, for example, to use what the person skilled in the art calls the "boot iPXE" which makes it possible to remotely reboot nodes so that they actually take over. account the software entities that were transmitted to them. It will also be noted that in order to transport the EEL sets, the computing network that connects the NSq service nodes to each other and to their respective NGAi auxiliary management nodes is preferably used. This computing network, which is often of the "InfiniBand" type, is indeed faster than the administration network, which is often of the Ethernet type.

It will also be noted that the device D can also allow the transmission of a single software entity to at least one chosen node, rather than an EEL set, when this is necessary or better adapted. The invention may advantageously allow a specialist to display on a screen coupled to the main management node NGP EEL sets that have been deployed in at least one NGAi auxiliary management node. For this purpose, it is possible for example to display on the screen, following the provision of a dedicated command (or request), the identifier of each set deployed on an auxiliary management node o NGAi and the identifiers of each of the entities software of the set considered. It should be noted that the software entities that have been deployed, but which have not been activated, may possibly be displayed with the ones activated, but with a different display mode (for example in square brackets).

Similarly, the invention may advantageously allow a specialist to display on a screen coupled to an auxiliary management node NGA i EEL sets that have been deployed in at least one NSii service node. For this purpose, it is possible for example to display on the screen, following the provision of a dedicated command (or request), the identifier of each set deployed on NSq service nodes coupled to the auxiliary management node NGAi concerned. , and the identifiers of these NSii service nodes in correspondence of the identifiers of the software entities of the set considered. The invention is not limited to the high performance control method, control device, and computer embodiments described above, by way of example only, but encompasses all variants that the skilled in the art within the scope of the claims below.

Claims (12)

REVENDICATIONS1. A method of controlling the deployment of software entities of different types in a high performance computer (CHP) comprising a main management node (NGP) coupled to at least two auxiliary management nodes (NGAi) each coupled to a group of service nodes (NSii), characterized in that it comprises a step (i) in which sets of at least two different software entities each associated with at least one auxiliary management node are stored in storage means (BP, BAi). (NGAi) or service node (NSii) in which they are to be deployed, and a step (ii) in which said ancillary management nodes (NGAi) are transmitted via said main management node (NGP), and said nodes service (NSq), via their respective auxiliary management nodes (NGAi), said sets respectively associated with them for local deployment. 2. Method according to claim 1, characterized in that one modifies a set associated with an auxiliary management node (NGAi) and transmitted to the latter (NGAi), depending on the service nodes (NSq) coupled to this node auxiliary management system (NGAi), and at least some of said sets associated with service nodes (NSii) coupled to an auxiliary management node (NGAi) are defined as a function of said modified set associated therewith (NGAi). 3. Method according to one of claims 1 and 2, characterized in that certain sets are defined by combining selected software entities contained in at least two other sets. 4. Method according to one of claims 1 to 3, characterized in that said software entities have types selected from a group comprising at least one type called image, a so-called virtual type and a so-called no disk type. 5. Method according to one of claims 1 to 4, characterized in that is stored in main storage means (BP), associated with said main management node (NGP), each set associated with at least one management node auxiliary (NGAi), and stored in auxiliary storage means (BAi) associated respectively with said auxiliary management nodes (NGAi), said sets respectively associated with the service nodes (NSq) which are coupled to these management nodes auxiliaries (NGAi). 6. Method according to one of claims 1 to 5, characterized in that in the presence of service nodes (NSii) of different types, different sets are associated with subgroups of a group respectively comprising nodes of service (NSq) of different types. 7. Method according to one of claims 1 to 6, characterized in that said sets comprise at least one software entity of a first type and / or at least one software entity of a second type and / or at least one entity. software of a third type. 8. Computer program product comprising a set of instructions which, when executed by processing means, is adapted to implement the method according to one of the preceding claims for controlling the deployment of software entities of different types in a high performance computer (CHP) comprising a main management node (NGP) coupled to at least two auxiliary management nodes (NGAi) each coupled to a group of service nodes (NSq). A device (D) for controlling the deployment of software entities of different types in a high performance computer (CHP) comprising a main management node (NGP) coupled to at least two auxiliary management nodes (NGAi) each coupled to a group of service nodes (NSii), characterized in that it comprises (i) interface means (Ml) arranged to store in storage means (BP, BAi) sets of at least two different software entities each associated with at least one auxiliary management node (NGAi) or service node (NSq) in which they are to be deployed, and (ii) control means (MC) arranged to trigger transmission to said auxiliary management nodes (NGAi). ), via said main management node (NGP), and said service nodes (NSq), via their respective auxiliary management nodes (NGAi), said sets respectively associated with them for their local deployment. 10. Device according to claim 9, characterized in that it comprises processing means (MT) arranged to modify a set associated with an auxiliary management node (NGAi) and transmitted to the latter (NGAi), depending on the nodes of NSii) coupled to this auxiliary management node (NGAi), and to define at least some of said sets associated with service nodes (NSq) coupled to an auxiliary management node (NGAi) according to said modified set associated with this last (NGAi). 11. Device according to claim 10, characterized in that said processing means (MT) are arranged to define certain sets by combining selected software entities contained in at least two other sets. A high performance computer (CHP) comprising a main management node (NGP) coupled to at least two auxiliary management nodes (NGAi) each coupled to a group of service nodes (NSii), characterized in that it comprises in addition to a control device (D) according to one of claims 9 to 11.