Arrangement and methods for safe and efficient management of processing loads in space-based computing cluster

Abstract

The nodes can be bidirectional communication wise connectable to a switch (210) in a first topology, and in a ring (230) in a second topology, and when the switch becomes unavailable, the ring-based connection can be used instead, and the control system can take into account additional communication tasks of the nodes for performing scheduling and / or assigning occurring due to the use of the ring-based connection. The invention also relates to control methods for enhancing operating lifetime of the arrangement.

Description

1ARRANGEMENT AND METHODS FOR SAFE AND EFFICIENT MANAGEMENTOF PROCESSING LOADS IN SPACE-BASED COMPUTING CLUSTERTECHNICAL FIELD OF THE INVENTION5

[0001] The present invention relates to compute architectures and control methods thereof, which are suitable for use in space (or other highly autonomous contexts), especially for demanding compute applications as one can find in a true data center. Particularly, the present invention relates to an arrangement comprising compute processing nodes, which is to be used10 in space vehicles, and methods for enhancing operating lifetime of such arrangement by providing task scheduling and load balancing.BACKGROUND OF THE INVENTION15

[0002] Existing compute arrangements (architectures) and their control methods do not take into account specific space conditions such as dynamics (e.g., power availability, communication aspects, exposure to the sun etc.) due to flight conditions and / or other harsh conditions, such as radiation, and their interrelation.20

[0003] The known infrastructure scheduling solutions, which however do not take into account such physical aspects like heat or available power and link availability, are disclosed, for example, in US11533234B2, US10791063B1, CN111614785B and US10317888B2.

[0004] Therefore, there is a need to provide methodology and computer-implemented25 architectures for safe and efficient management and distribution of processing tasks in a computing cluster operating in harsh environment of space, taking into account the limitations inherent to space systems, such as occurrence of faults, thermal management and limited resources.30 AIM OF THE INVENTION

[0005] The invention is generally aimed at providing task scheduling and load balancing for the field of the invention.2

[0006] The invention is particularly aimed at providing topologies for groups of processing nodes suited for the same field and adapted accordingly, and providing task scheduling and load balancing while exploiting those topologies, taking into consideration the features of those topologies.5SUMMARY OF THE INVENTION

[0007] The present invention is set out in the appended set of claims.10

[0008] Particularly, according to one aspect of the present invention, a compute arrangement comprising a plurality of compute processing nodes and a control system for scheduling tasks in time and assigning said tasks to one or more of the nodes in place, wherein for performing the scheduling and / or assigning the control system is configured to compute an estimated behavior of the one or more of the nodes based on a model thereof, the model estimating at15 least an expected temperature of the one or more of the nodes.

[0009] In one embodiment, the nodes are bidirectional communication wise connectable in a first topology, and wherein the nodes are bidirectional communication wise connectable in a second topology with more direct connections between the nodes than in the first topology, and20 wherein when the first topology becomes unavailable the arrangement is configured to exploit the second topology instead, and wherein the control system is configured to take into account additional communication tasks of one or more of the nodes for performing the scheduling and / or assigning when using the second topology.25

[0010] Preferably, in the first topology the nodes are connected to a switch.

[0011] Preferably, in the second topology the nodes are connected in a ring.

[0012] In one embodiment, the nodes are bidirectional communication wise connectable to a30 switch, and the nodes are bidirectional communication wise connectable in a ring, and when the switch becomes unavailable, the arrangement is configured to exploit the ring-based connection instead, and the control system is configured to take into account additional communication tasks of one or more of the nodes for performing scheduling and / or assigning occurring due to the use of the ring-based connection.3

[0013] In another embodiment, the arrangement further comprises communication means to enable communication between the nodes, wherein the nodes are connected to the communication means.5

[0014] In another embodiment, the control system is configured to receive information related to an actual behavior of the nodes and based thereon, to re-compute the scheduling and / or assigning.

[0015] In another embodiment, the information related to the actual behavior of the nodes10 pertains to a temperature of one or more of the nodes and / or their performance degradation due to radiation induced faults.

[0016] In another embodiment, the control system is configured to receive information related to actual available nodes of the nodes and their connection topology and based thereon, to15 compute and / or re-compute the scheduling and / or assigning.

[0017] According to another aspect of the present invention, a space vehicle is provided comprising one or more of the arrangements according to a first aspect of the invention. The space vehicle can be a satellite, a space station or a spaceship.20

[0018] According to another aspect of the present invention, a space vehicle system is provided comprising a plurality of space vehicles, preferably selected from satellites, space stations and / or spaceships, wherein each of the space vehicles comprises one or more compute processing nodes, wherein the nodes together are comprised in the arrangement according to a25 first aspect of the invention. Optionally, an actual availability of the nodes depends on a place in the orbit of the space vehicle they belong to (e.g. due to the available energy, for instance, from solar panels).

[0019] According to another aspect of the present invention, a method is provided for30 enhancing operating lifetime of the arrangement according to a first aspect of the invention, wherein based on estimating and / or detecting temperature of one or more of the nodes, the control system performs scheduling tasks in time and assigning said tasks to one or more of the nodes in place, so as to reduce temperature issues in the nodes. The temperature issues4 optionally comprise high amplitude and frequency of temperature variations and / or differences amongst the nodes.

[0020] In one embodiment of the method, based on estimating and / or detecting undesirable5 temperature of at least one of the nodes, a control signal is generated by the control system to switch off the node and switch on the node after a predetermined period has lapsed.

[0021] In one more embodiment of the method, prior to switching off the node, when possible, at least one task running on the node and / or a state of the node is transferred to another node.10

[0022] In one more embodiment of the method, the control system manages the arrangement such that one or more nodes are reserved to ensure that, prior to switching off the node with undesirable temperature, the at least one task running on the node and / or the state of the node is transferred to one of the reserved nodes.15

[0023] In one more embodiment of the method, the control system takes into account that one or more of the nodes may invoke precautionary measures for re-active or pro-active fault removal in them, optionally related to radiation induced, preferably non-transient, faults.20

[0024] In another embodiment of the method, further based on estimating and / or detecting unacceptable performance of at least one of the nodes, which includes not respecting communication timeouts, reporting errors and / or unsuccessful task completion, a control signal is generated by the control system to switch off the node and switch on the node after a predetermined period has lapsed.25

[0025] According to another aspect of the present invention, a computer program is provided comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method for enhancing operating lifetime of the arrangement, as described above.30

[0026] According to another aspect of the present invention, a computer-readable medium is provided comprising instructions which, when executed by a computer, cause the computer to carry out the method for enhancing operating lifetime of the arrangement, as described above.5

[0027] The computer-readable medium can be a non-transitory program storage device tangibly embodying a program of instructions executable by the computer to perform the method for enhancing operating lifetime of the arrangement. Such a non-transitory program storage device can be, without limitation, an electronic, magnetic, optical, electromagnetic,5 infrared, or semiconductor device, or any suitable combination of the foregoing. It is to be appreciated that the following, while providing more specific examples, is merely an illustrative and not exhaustive listing as readily appreciated by one of ordinary skill in the art: a portable computer diskette, a hard disk, a ROM, an EPROM (Erasable Programmable ROM) or a Flash memory, a portable CD-ROM (Compact-Disc ROM).10

[0028] As can be seen from the above summary, the present invention beneficially provides compute architectures suitable for use in the harsh circumstances occurring in space, such as radiation. The invention also pertains to digital control of the above-mentioned architectures by task scheduling in time (i.e., which task is executed and when) and load balancing (where15 the task is executed) between a plurality of compute processing nodes. It is a contribution of the invention to provide models, representative for the performance of said architectures and use them as estimators in the above-mentioned control.

[0029] A particular issue addressed by the invention is that while using compute architectures20 in space one cannot or can only to a very limited extent use the heat disposal means commonly used at earth. Therefore, the digital control of the mentioned architectures takes into account heat dissipation resulting from executing a task on the processing node, so as to minimize occurrence of temperature gradients, which leads to mechanical stress and eventually results in electrical failures. It is therefore a further contribution of the invention to provide models25 representative for the temperature performance of said architectures and use them as estimators in the mentioned control. Thus, the invention can achieve minimization of the amplitude and frequency of temperature changes experienced by the processing nodes.

[0030] Said models are beneficially used by the digital control system in open loop to estimate30 performance for scheduling and task-node assignments and, based thereon, a schedule and tasknode assignment is determined or selected when being within predetermined constraints on said performance. Further, beneficially, in a sort of closed loop, while executing said selected schedule and task-node assignment, one or more of said performances can be measured by use6 of sensors, and in case said predetermined constraints are violated, the execution can be terminated, and a new schedule and task-node assignment can be determined.

[0031] Because said compute architectures are used in the harsh circumstances of space, one5 may not assume that the full capabilities of said architectures remain available. For task scheduling and / or load balancing of the architectures, the digital control system advantageously takes into account these varying diminishing capabilities thereof. It is thus a contribution of the invention to adapt the models, representative for the performance of said architectures and use of said adapted models as estimators in the mentioned control.10

[0032] The architectures according to the invention are characterized by having a plurality of (compute) processing nodes connected in a predetermined configuration. However, because said architectures are used in the harsh circumstances of space, one may not assume that this predetermined configuration of said architectures remain feasible. It is thus a contribution of15 the invention to provide architectures which can be configured in a plurality of ways. It is further contribution of the invention by providing an approach of selecting another, still feasible, configuration (which may be less performing though) amongst those configurations. It is also a contribution of the invention by using configuration dependent models, representative for the performance of said architectures and use of the appropriate model as20 estimator in the mentioned control. The configuration dependent models can be and preferably are of the adaptive kind mentioned above. It is a further contribution of the invention to realize that the use of such, still feasible configuration (which may be less performing though) invokes additional tasks, which use needs to be controlled by the control system, as beneficially provided by the invention.25

[0033] The invention advantageously provides that the compute processing nodes of the compute architecture can be placed within one satellite and the digital control discussed above is confined to said one satellite (or a cluster of the nodes therein). According to the invention, said compute processing nodes can be advantageously placed across a plurality of satellites and30 the digital control discussed above is applied to plurality of satellites (or a cluster of the cluster of the nodes therein) as a whole or jointly. For example, one may determine a (rough) task scheduling and task-satellite assignment and thereafter a more detailed task scheduling and task-note assignment within the satellite though. It is a further contribution of the invention to7 realize that the use of such multi-satellite arrangement invokes additional (communication) tasks, which use needs to be controlled by the control system.

[0034] When dealing with multiple (varying) configurations as outlined above, one may5 consider that one or more satellites in the plurality of satellite arrangements is no longer available (or in the opposite case, becomes available (again)), in particular because the actual availability of the processing nodes depends on the place in the orbit of the space vehicle they belong to (e.g. due to the available energy, for instance, from solar panels). Advantageously, in the invention means to measure the current energy state are provided. Those can be used to10 make predictions, taking into account the dynamics of movement around a planet.

[0035] When dealing with multiple (varying) configurations as outlined above, in said architectures (within a single satellite), which are characterized by having a plurality of (compute) processing nodes connected in a predetermined configuration and supported by15 connecting elements such as switches, the connecting elements can be no longer available (e.g., because of failures due to the harsh conditions) and the present invention advantageously provides that the processing nodes themselves have to act as connecting elements instead (for instance, by organizing themselves as a ring of nodes).20

[0036] The embodiments with changing availability of satellites and changing internal configurations can be combined and the invention advantageously provides digital control capable of handling these situations.

[0037] The invention beneficially adapts the above digital control and / or architecture such that25 those can act as a true data center.

[0038] Summarizing, the invention provides low cost, low effort, alternative network to overcome switch failures, and efficient task scheduling algorithm that allows to deploy systems in resource constrained and challenging environments.30BRIEF DESCRIPTION OF THE DRAWINGS8

[0039] The present disclosure will be better understood, and other specific features and advantages will emerge upon reading the following description of particular and non-restrictive illustrative embodiments, the description making reference to the annexed drawings wherein:5

[0040] Figure 1 describes a method flowchart in accordance with one of the aspects of the invention.

[0041] Figure 2 describes a compute arrangement (architecture) topology in accordance with another aspect of the invention, which can be beneficially used with the aspect of the invention10 described in Figure 1.ILLUSTRATIVE EMBODIMENTS

[0042] The present description illustrates the principles of the present disclosure. It will thus15 be appreciated that those skilled in the art will be able to devise various arrangements that, although not explicitly described or shown herein, embody the principles of the disclosure and are included within its scope.

[0043] All examples and conditional language recited herein are intended for educational20 purposes to aid the reader in understanding the principles of the disclosure and the concepts contributed by the inventor to furthering the art, and are to be construed as being without limitation to such specifically recited examples and conditions.

[0044] Moreover, all statements herein reciting principles, aspects, and embodiments of the25 disclosure, as well as specific examples thereof, are intended to encompass both structural and functional equivalents thereof. Additionally, it is intended that such equivalents include both currently known equivalents as well as equivalents developed in the future, i.e., any elements developed that perform the same function, regardless of structure.30

[0045] It will be appreciated that some aspects of the present invention may be substantially implemented with the help of computer readable media and so executed by a computer or processor, whether or not such computer or processor is explicitly shown in the figures.9

[0046] The functions of the various elements shown in the figures may be provided through the use of dedicated hardware as well as hardware capable of executing software in association with appropriate software.5

[0047] The terms “adapted” and “configured” are used in the present disclosure as broadly encompassing initial configuration, later adaptation or complementation of the components or aspects of the invention, or any combination thereof, whether effected through material or software means.10

[0048] The present disclosure will be described with reference to the drawings.Tasks and processing jobs scheduling

[0049] The invention pertains to computer-implemented tasks and processing jobs scheduling15 taking into account several aspects when dealing within the context of the invention.

[0050] The invention pertains to a computer-implemented (compute) arrangement (architecture) comprising a plurality of (compute) processing nodes, and a control system for scheduling tasks (in time), and assigning said tasks to one or more of the nodes (in place). For20 performing the scheduling and / or assigning, the control system is configured to compute an estimated behavior of the one or more of the nodes based on a model thereof, the model estimating at least an expected temperature of the one or more of the nodes.

[0051] The invention also pertains to processing nodes in a satellite or the like, a constellation25 of processing nodes within such satellite, and constellations of satellites.

[0052] The invention also pertains to control methods for enhancing operating lifetime of the arrangement (architecture).30

[0053] The invention deals with issues of possible temporary unavailability of the nodes for deployment of tasks, inputs provisions, and getting the results back. The problems to deal with can stem from: availability of the communication link or links,10 availability of the power / energy at the end of processing (also pertains to a single satellite cluster), other tasks deployed in the meantime (also pertains to a single satellite cluster), thermal status of the system nodes (also pertains to a single satellite cluster).5

[0054] The invention provides the use of a (distributed) algorithm containing a digital twin or model of a cluster or a cluster of clusters (distributed among a variety of satellites). The algorithm is deployed on one or more of the processing nodes participating in the scheduling and load balancing scheme.10

[0055] The algorithm takes into account information about:- network topology of the system - what processing node is connected to what processing node, via which communication links,- physical implementation of the system - which processing nodes are close to each other15 (thermal coupling), supplied from the same power sources, and- status of the particular subsystems:- operation temperature,- remaining power,- deployed processing tasks,20 - processing tasks to be deployed,- efficiency of given tasks processing,- operational status (on, off, number of tasks already deployed, criticality of tasks),- elapsed time of operation.25

[0056] Figure 1 provides an illustration of a flowchart of the algorithm used to implement one embodiment of the invention, which comprises the following steps: step (1): start the algorithm; step (2): update a network topology and a physical system architecture (digital twin); step (3): get status information of all the processing nodes in the system30 (arrangement / architecture) ; step (4): propose which processing task shall be deployed and on which processing node; step (5): estimate results of the proposed task to node mapping in terms of thermal, power budget and task criticality management;11 step (6): determine whether this mapping is safe according to system constraints and limits (yes (Y) or no (N)): if no (N), return to step (4); if yes (Y), perform step (7): deploy processing tasks; and5 perform step (8): determine whether to end scheduling (yes (Y) or no (N)): if yes (Y), perform step (10): end of the algorithm; if no (N), perform step (9): determine whether to update network topology and physical system architecture (yes (Y) or no (N)): if yes (Y), return to step (2);10 if no (N), return to step (3).

[0057] Within Figure 1 one can recognize a method step executed by said control system for performing scheduling and / or assigning tasks to the processing nodes. One can further observe in Figure 1 the step of receiving information related to the actual behavior (status) of said15 processing nodes and, based thereon, re-computing said scheduling and / or assigning. One further can find in Figure 1 the step of receiving information related to the actual available processing nodes and their connection topology and, based thereon, (re-)computing said scheduling and / or assigning.20 Communication with the payloads and the system data sinks

[0058] Problem existing in the art consists in that processing nodes are connected through a switch (first topology) combined with a remote terminal unit, with the rest of the satellites and all payloads. Switches are a single point of failures, but the redundancy is not always possible25 due to mass or volume constraints.

[0059] Solution of the mentioned problem by the invention is as follows. Where the above provides the first topology, the invention provides for an alternative independent information distribution system as shown in Figure 2 defining a second topology.30

[0060] Figure 2 particularly shows that processing nodes (200) are (bidirectional communication wise) connected / connectable to a switch (210) and said processing nodes (200) are (bidirectional communication wise) connected / connectable in a ring (230). Figure 2 further shows that the processing nodes (200) can be connected to dedicated external communication12 means (220) called payload. Each of the processing nodes (200) can interface to one or more payloads (220) for data upload and download.

[0061] In this embodiment all the processing nodes (200) are put in the bi-directional ring (230)5 network (as shown by dashed arrows in Figure 2) ensuring expansion of fault tolerance at minimal cost (cost can be potentially affected only by some high-speed harness, extra data links are typically available in COTS components).

[0062] The payload can be a data source (e.g., a camera or a scientific instrument) but also a10 data sink (downlink).

[0063] The processing nodes, as a minimum, can be able to feed forward the data through ring network, one or the other way.15

[0064] Each of the processing nodes can serve as mini-routers connecting payload, ring network and the main switch.

[0065] The ring network can be used to communicate to nearby nodes that need to proactive refresh / power cycle as well.20

[0066] The invention can be applied in any reliability challenging and resource constrained environment, i.e. autonomous cars.

[0067] In an embodiment of the load balancing described above, the particularities of the25 second topology (ring connection) are taken into account because the further you send data the more you add extra load to the nodes along the path.

[0068] The present invention thus advantageously provides an algorithm for scheduling and distributing processing tasks by measuring the physical state of the processing nodes,30 cluster and host satellite, and analyzing the cluster model (digital twin) taking into account long term needs of the system operation. Further, the present invention provides cluster architecture extension to provide fault tolerance without need to implement cluster switch redundancy.13Challenges associated with the task scheduling / distribution model on spacecraft / constellation level

[0069] Main challenges associated with building task scheduling / distribution models for5 space-based processing infrastructure stems from the fact that unlike, similar, ground-based infrastructure, space micro data centres operate in resource limited environments.

[0070] Environment limits the amount of energy available for processing, as it’s proportional to the time spent by spacecraft in the sunlight and sizing of its solar panels, and inversely10 proportional to the amount of energy spent on other spacecraft activities (communication, attitude control, temperature control). Therefore, it shall be expected that the total amount of energy available for data processing activity shall depend on orbit and mission plan, and shall be modelled in order to deploy processing tasks efficiently and with certainty that the processing can be finished before the system runs out of available energy.15

[0071] Likewise, execution of tasks causes the dissipation of waste heat and as a result increases the temperature of operation of processing nodes. The capability of a space-based data processing to cool down passively is limited by unavoidable thermal resistances and effectiveness of heat irradiation into deep space. At the same time, active cooling of a space¬20 based data canter is limited, again, by available energy. Cooling capacity of processing nodes shall be considered a limited resource, and included into the task scheduling / distribution models to ensure the allowed operating temperature range of processing node is respected and that number and depth of thermal cycles (temperature change frequency and amplitude) is limited to ensure the data processing node longevity.25Challenges associated with the task scheduling / distribution model ring backup network

[0072] When ring backup network is activated, processing nodes, beside performing processing tasks, utilise their own processing resources (processor time, storage) to act as a30 local network switch and they do it less efficiently compared to dedicated switch nodes.

[0073] It is expected that during backup ring network operation the total processing capability of the system will be reduced because of communication (switching) overhead.14

[0074] Moreover, this overhead will increase proportionally to the number of processing nodes acting as local switches, which the data stream will have to traverse, in order to reach it’s target node.5 Rationale behind central processing switch and ring backup network

[0075] Space based computing clusters are constructed out of selection of the processing nodes, switches, communication links to the payloads, to the switch(es) and adjacent processing nodes. Since said equipment operates in space and is subject to the influence of ionising10 radiation and resulting effect in microelectronics, it shall be expected to fail over time, in numerous, temporary and permanent ways. In order to address the fault occurrences and extend the operational lifetime of the cluster, the system shall be equipped with several, independent, communication pathways, implemented in a way that doesn’t immediately violate size, weight and power constraints of the spacecraft.15

[0076] The first topology is based on a central switch (star topology) which is a system node dedicated to quick and effective packets (data streams) switching and transfer between nodes, and between nodes and other parts of the system (spacecraft platform, interspacecraft communication links, other processing clusters, data centers). It is an effective and efficient20 solution, which is a single point of failure. It could be made redundant (double star topology) but it may not always be possible due to mass and volume constraints of the system hosting computing cluster.

[0077] In order to overcome this limitation, a ring backup network is proposed according to25 the present invention, which bidirectionally links subsequent nodes, enforcing them to act like local switches, ensuring capacity of a system to transfer packets from any node to any other node, without use of a central switch.

[0078] It’s not required by the invention that the system work in either centralised or ring30 network variant. Intermediate variants are possible, where only part of the ring network is activated while the central switch remains operational. Such examples could easily address the occurrence of a broken communication link between one node and a switch. In this case, the node that cannot communicate with the switch directly would have to route data back and forth through nearby nodes in ring network which are still connected to the central switch.15

[0079] The above disclosed subject-matter is to be considered illustrative, and not restrictive, and serves to provide a better understanding of the present invention set out in the appended set of claims.

Claims

16CLAIMSA. compute arrangement comprising a plurality of compute processing nodes (200) and a control system for scheduling tasks in time and assigning said tasks to one or more of the5 nodes in place, wherein for performing the scheduling and / or assigning the control system is configured, to compute an estimated behavior of the one or more of the nodes based on a model thereof, the model estimating at least an expected temperature of the one or more of the nodes; wherein the nodes are bidirectional communication wise connectable in a first topology, and wherein the nodes are bidirectional communication wise connectable in a second topology with10 more direct connections between the nodes than in. the first topology, and wherein when the first topology becomes unavailable the arrangement is configured to exploit the second topology instead, and wherein the control system, is configured to take into account additional communication tasks of one or more of the nodes for performing the scheduling, and / or assigning when using the second topology.

152. The arrangement of claim 1, wherein in the first topology the nodes are connected to a switch (210), and / or in the second topology the nodes are connected in a ring (230).

3. The arrangement of claim 1 , wherein, the nodes are bidirectional communication wise20 connectable to a switch (210), and wherein die nodes are bidirectional communication wise connectable in a ring (230), and wherein when the switch becomes unavailable, the arrangement is configured to exploit the ring-based connection instead, and wherein the control system is configured to take into account additional communication tasks of one or more of the nodes for performing scheduling and / or assigning occurring due to the use of the ring-based25 connection.

4. The arrangement of claim L farther comprising communication means (220) to enable communication between the nodes, wherein each of the nodes is connected to one or more of the communication means.

305. The arrangement of claim 1 , wherein the control system is configured to receive information, related to an actual behavior of the nodes and based thereon, to re-compute the scheduling and / or assigning, wherein preferably the information related to the actual behavior17 of the nodes pertains to a temperature of one or more of the nodes and / or thei r performance degradation due to radiation induced faults.

6. The arrangement of claim 1 , wherein the control system is configured to receive5 information related to actual available nodes of the nodes and their connection topology' and based thereon, to compute and / or re-compute the scheduling and / or assigning.

7. A space vehicle comprising one or more of the arrangements of any one of claims 1 to 6, the space vehicle preferably selected from a satellite, space station and spaceship.

108. A space vehicle system, comprising a plurality of space vehicles, preferably selected from satellites, space stations and / or spaceships, wherein each, of the space vehicles comprises one or more compute processing nodes (200), wherein the nodes together are comprised in the arrangement of any one of claims 1 to 6, wherein optionally an actual availability of the nodes15 depends on a place in the orbi t of the space vehicle they belong to.

9. A. .method for enhancing operating lifetime of the arrangement of any one of claims 1 to 6, wherein based on estimating and / or detecting temperature of one or more of the nodes, the control system performs scheduling tasks in time and assigning said tasks to one or more20 of the nodes in place, so as to reduce temperature issues in the nodes, the temperature issues optionally comprising high amplitude and frequency of temperature variations and / or differences amongst the nodes.

10. The method of claim 9, wherein based on estimating and / or detecting, undesirable25 temperature of at least one of the nodes, a. control signal is generated by the control system to switch off the node and s wi tch on the node after a predetermined period has lapsed.I L The method of claim 10, wherein prior to switching off the node, when possible, at least one task running on the node and / or a. state of the node is transferred to another node.3012. The method of claim 1 1 , wherein the control system manages the arrangement such that one or more nodes are reserved to ensure that, prior to switching off the node with undesirable temperature, the at least one task running on the node and / or the state of the node is transferred to one of the reserved nodes.1813. The method of claim. 12, wherein the control system takes into account that one or more of the nodes may invoke precautionary measures for re-active or pro-active fault removal in them, optionally related to radiation induced, preferably non-transient, faults.

514. The method of claim 9, wherein further based on estimating and / or detecting unacceptable performance of at least one of the nodes, which, includes not respecting communication timeouts, reporting errors and / or unsuccessful task completion, a control signal is generated by the control system to switch off the node and switch on the node after a10 predetermined period has lapsed.

15. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of claims 9 to 14.

Images