CN115840638A - Function filling model based on resource fragment space-time feature perception and method thereof - Google Patents
Function filling model based on resource fragment space-time feature perception and method thereof Download PDFInfo
- Publication number
- CN115840638A CN115840638A CN202211258523.7A CN202211258523A CN115840638A CN 115840638 A CN115840638 A CN 115840638A CN 202211258523 A CN202211258523 A CN 202211258523A CN 115840638 A CN115840638 A CN 115840638A
- Authority
- CN
- China
- Prior art keywords
- resource
- time
- function
- unallocated
- space
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a function filling model based on source fragment space-time characteristic perception and a method thereof.A non-allocated resource predictor periodically sends the non-allocated resource quantity in a period of time to a phase space on a node according to a discrete stable period and a local intensive period; the unused resource predictor is used for predicting periodic points and unused resources through a deep neural learning model, then acquiring available non-periodic points and unused resources at the current moment and uniformly transmitting the sizes to a facies space; the scheduling unit sends an optimal scheduling scheme to the phase space according to the function scheduling request according to the resource state of 'unallocated', 'unused' and the function request resource amount; the method and the system can effectively sense the change of the resource fragments generated by the virtual machine service in a period of time in the future, and dispatch the functions to the fragments which can meet the requirement of the resources in the execution time, thereby reducing the resource competition.
Description
The technical field is as follows:
the invention belongs to the technical field of mixed deployment of different applications for improving resource utilization rate in a public cloud environment in cloud computing, and particularly relates to a function filling model based on resource fragment space-time feature perception and a method thereof.
Background art:
cloud data centers have evolved into the key infrastructure of modern digital economies. However, under the background of large-scale data center construction with high cost, the utilization rate of data center resources is generally low, which causes a great deal of resource waste. For example, data published by google cloud, microsoft cloud, and ali cloud show that the average utilization rate of the CPU is 25% -60%. In order to reduce cost and improve efficiency, cloud service providers mostly adopt a hybrid deployment technology to improve the resource utilization rate of a data center. However, resource contention interference between mixed applications causes performance degradation of cloud services, especially the tail delay of delay-sensitive interactive services increases significantly.
In order to solve the problem of cloud service performance reduction in the application mixing process, the existing work respectively provides an active control type and a feedback regulation type hybrid scheduling deployment technology. The 'active control type' mixed part technology realizes accurate prediction of interference by describing the execution characteristics of application in an early stage and establishing a prediction model. The feedback adjustment type hybrid scheduling deployment technology monitors the running performance of the cloud service in real time, and when the application performance breaks through a critical value, a management controller intervenes in a resource adjustment and allocation process to realize the rapid recovery of the cloud service performance. However, the above solutions deeply depend on offline analysis or real-time monitoring of the application, and in a public cloud environment, the tenant application exists in the form of a "black box" in the virtual machine, and it is difficult for a cloud service provider to accurately depict the application execution characteristics.
In a public cloud data center, on one hand, the server fragmentation "unallocated resources" after virtual machine scheduling presents a relatively stable and locally dense characteristic, and resources that have been allocated to a virtual machine but not yet fully used, namely "resources that are not fully used" present a partially periodic characteristic. For example, the average life cycle of microsoft "unallocated resources" is 61.5 days, and the average time interval of CPU increase and decrease is 17.8 hours, which shows the relative stability of unallocated resource fluctuation; the services run by 75% of the virtual machines in the data center are periodic and constant, i.e., resources allocated to the virtual machines but not used in time exhibit partial periodicity. On the other hand, the emerging Serverless (Serverless) model has been rapidly developed in recent years. The basic computing task in the serverless computing, namely the function, has obvious small-volume and short-period characteristics. 90% of the Azure function requests less than 400MB of memory, 50% of the application program allocates up to 170mb of memory when running, 50% of the call has an average execution time of less than 1 second, and 96% of the function has an average execution time of less than 60 seconds. Therefore, the server-free function calculation is executed by utilizing the fragment resource scheduling of the server, so that the resource utilization rate of the data center resource is greatly improved, and the system throughput is improved.
The invention content is as follows:
aiming at the problems in the prior art, the invention provides a function filling model based on resource fragment space-time characteristic perception, which is applied to a running system under a public cloud environment, can effectively perceive the size change of resource fragments generated by virtual machine service in a period of time in the future, and schedules a function to the fragments which can meet the required resources in the execution time of the function, thereby reducing resource competition. Meanwhile, the invention is used as an independent runtime system, which is not strongly coupled with the platform, so that the invention can be operated on most server-free computing platforms. The requirements of various applications on resources in the execution time of the applications can be met when the hybrid deployment is ensured, so that resource competition after the hybrid deployment is reduced, and the resource utilization rate and effective throughput are greatly improved on the premise of ensuring the service quality.
The invention solves the practical problem by adopting the following technical scheme:
a function filling model based on resource fragment space-time feature perception is applied to a public cloud system containing virtual machines, and comprises an unallocated resource space-time predictor, a shared and unused resource space-time predictor, a periodic classifier of VM service, a resource phase space and a scheduling unit; wherein:
the unallocated resource space-time predictor is used for analyzing the virtual machine creation time interval of the node and predicting the fluctuation time point of the next unallocated resource to obtain the unallocated resource amount in the future time
The point-unused resource space-time predictor is used for predicting the point-unused resource amount generated by the periodic service in the future time and collecting the non-periodic point-unused resource amount generated by the non-periodic service every second;
the periodic classifier is used for dividing the resources of the virtual machine on the node into periodicity and influencing the prediction range of the unused resources space-time predictor according to the division result;
the phase space is used for recording the unallocated resource amount, the periodic score but unused resource amount in a future period of time and the non-periodic score but unused resource amount in the current time;
and the scheduling unit outputs an optimal matching resource scheduling scheme according to the unallocated resource amount, the unused resources and the function request resource amount.
Further, the scheduling unit comprises a function carver and a function scheduler; wherein:
the function carver obtains the required resource amount at any time in the execution time of the function carver by carving the function, and sends the carved function to the function scheduler for scheduling;
and the function scheduler selects the node with optimal stability and sufficient resource supply in the function execution time to deploy the function service by analyzing the phase space resource stability of each global node.
The invention can also be implemented using the following technical solution,
a function filling model based on resource fragment space-time feature perception comprises the following steps:
the unallocated resource predictor periodically sends unallocated resource amount in a future period of time to a phase space in a discrete stable period and a local intensive period according to a virtual machine time interval;
the used resource predictor predicts the periodic score and the unused resource through a deep neural learning model, then acquires the size of available non-periodic resources at the current moment and sends the size to a facies space;
and the scheduling unit sends an optimal scheduling scheme to the phase space according to the function scheduling request according to the resource state of 'unallocated', the resource state of 'unused' and the resource quantity of the function request.
Further, the unallocated resource predictor periodically transmits an amount of unallocated resource for a future period of time to a facies space process:
initializing the locally intensive creation track input of all virtual machines on a node by the unallocated resource space-time predictor;
the unallocated resource space-time predictor judges whether the current time is in a discrete stable period or a local intensive period by analyzing the virtual machine creation time interval on the current node, extracts an intensive creation track with the closest average interval time when the unallocated resource space-time predictor is in the local intensive period, calculates the shortest creation time, the longest creation time and the average resource amount of the track, sets the time range from the shortest creation time to the longest creation time as the fluctuation time of the unallocated resource next time, and sets the average resource amount as the reduction amount of the unallocated resource next time;
the unallocated resource spatiotemporal predictor converts the corresponding resource amount from the unallocated resource to the non-periodic sharing of the unused resource near the predicted virtual machine creation time, and after the predicted time is exceeded, no new real virtual machine is created, and then the corresponding periodic sharing of the unused resource is converted back to the unallocated resource.
Further, the score without resource predictor predicts a score without resource send phase space procedure:
and the resource space-time predictor without using the resources only predicts the periodic resources divided by the periodic classifier, predicts the size of the periodic resources in each time scale in the future through a deep neural learning model, and calculates the variance between the predicted value and the consumption value of the virtual machine during execution, wherein when the variance is greater than a threshold value, the corresponding periodic resources are converted into non-periodic resources.
Further, the scheduling unit sends an optimal scheduling scheme process to the phase space according to the function scheduling request according to the resource status of 'unallocated', 'unused' and the function request resource amount:
the function carver obtains the resource usage amount and the execution time of the function after executing each newly arrived function, and schedules a function request of a two-dimensional required resource label;
the function scheduler schedules the function request of the required resource label in a node phase space with optimal stability, unallocated resources plus periodic points and unused resources greater than the required resources in the function execution time according to the function with the execution time greater than 1 s; and scheduling the function with the execution time less than 1s in the non-periodic resources with the resource amount meeting the non-periodic time.
Has the beneficial effects that:
in order to reduce the service quality degradation caused by the mixed part of the virtual machine application and the function application, previous researches propose that the resources distributed to the function service are fed back and adjusted by monitoring the IPC fluctuation of the virtual machine, and then the resources at the current moment are distributed to the function service according to the quantity. The method only concerns the allocable resource amount at the current moment during scheduling, and ignores the competition of the function on the resource in the whole execution time, which may affect the service quality of the virtual machine. In order to make up for the defects of the scheme, the invention provides a function filling technology based on resource space-time feature perception, and the resource utilization rate and the effective throughput rate are greatly improved under the condition of ensuring the delay of the original service tail to reach the standard.
Compared with the prior art, the method considers that the requirement of the function application on the resource in the execution time is met during the scheduling. Through the classification and prediction of the distributable fragment resources, the size of the fluctuating fragment resources in a period of time in the future is obtained, and the function is scheduled to the resource fragments with any resource requirements in execution time, so that the possibility of resource competition which can occur in the future is reduced. The evaluation result shows that compared with the most advanced technology, the method can improve the CPU utilization rate by 19% and improve the effective throughput by 47% on the premise of ensuring the 99-branch delay compliance of the LC service.
Description of the drawings:
FIG. 1 is a schematic diagram of a function filling model structure based on resource fragment space-time feature perception according to the present invention.
Detailed Description
As shown in FIG. 1, the invention provides a function filling model based on resource fragment space-time feature perception, which is composed of core concepts such as feature classification and space-time prediction of resource fragments, space-time filling of functions and the like. The system consists of a node-level unallocated resource space-time predictor, a classified unused resource space-time predictor, a periodic classifier, a resource phase space, a cluster-level function characterizer and a function scheduler. The unallocated resource space-time predictor predicts the fluctuation time point of the next unallocated resource through a statistical model by analyzing the fluctuation interval of the unallocated resource on a node, so as to calculate the unallocated resource amount in the future time; predicting the resource amount of the periodic resources in the future time through a deep learning model by a space-time predictor of the unused resources, and collecting and reporting the non-periodic resource amount every second; the periodic classifier divides the virtual machine service on the node into periodicity by adopting a fast Fourier transform method, and influences the prediction range of the space-time predictor according to the division result; recording the available resource quantity of each moment from the current time point to a future period of time by the phase space for the scheduling function service of the scheduler; the function carver carves the function to obtain the required resource amount at any time in the execution time of the function carver, and sends the carved function to the function scheduler for scheduling; and the function scheduler selects the node with optimal stability and sufficient resource supply in the function execution time to deploy the function service by analyzing the phase space resource stability of each global node.
1. Periodic classifier
The periodic classifier collects a section of virtual machine resource utilization rate through virt-top, performs fast Fourier transform on the section of resource utilization rate to obtain a spectrogram of the virtual machine resource utilization rate, and analyzes whether the spectrogram has the following characteristics at the same time: 1. the spectrogram has significant outliers, the significant outlier data mode points have a difference of magnitude 2, the spectrogram signal is a discrete signal 3, the spectrogram is close to the range of the domain boundary, and the magnitude of the point set is equal and close to 0. When the above characteristics are all satisfied, the original service is a periodic service with predictability, otherwise, the original service is an aperiodic service.
2. Unallocated resource spatio-temporal predictor
The unallocated resource space-time predictor judges whether the current interval is approximately equal to the last interval and is smaller than the threshold time by analyzing a virtual machine creation time interval graph, if so, the interval average value is set as the next virtual machine creation time, the resource size average value of the virtual machine in the two intervals is set as the amount of resource reduction of the next unallocated resource, the corresponding resource amount is converted into the non-periodic sharing without using the resource near the predicted virtual machine creation time, and if no new real virtual machine is created after the predicted time is exceeded, the corresponding periodic sharing without using the resource is converted back into the unallocated resource.
Prediction of the spatio-temporal distribution of "unallocated" fragmented resources:
the "unallocated" fragmented resources refer to resources left unallocated to any virtual machine on the server after the scheduling of the virtual machine is completed. Its fluctuation time series has the typical "relatively stable, locally dense" characteristic. According to the characteristics, the statistical analysis and the modeling can be respectively carried out aiming at the discrete stationary phase and the local intensive phase. In a discrete stable period, the scheduling request of the virtual machine is sparse, and the prediction difficulty is high, so that a prediction method is not designed. In a local intensive period, predicting the request reaching time of an intensive area by using a statistical method, and calculating the size of the unallocated fragment resource in the next future phase space window to serve as a prediction result to be reported to a scheduler.
3. Spatio-temporal predictor of fractional unused resources
The method comprises the steps that a resource space-time predictor is not used, only periodic resources divided by a periodic classifier are predicted, the size of the periodic resources in each time scale in the future is predicted through an LSTM model, meanwhile, a real resource fluctuation track is collected to be compared with a predicted value, when a variance is larger than a threshold value, the corresponding periodic resources are converted into non-periodic resources, and meanwhile, the periodic classifier is requested to divide service periodicity again; and the spatial-temporal predictor of the resources which are not used for the time division predicts the periodic resources, then acquires the size of the available non-periodic resources at the current moment, and sends the size to a phase space in a unified way.
"Create unused" resources refer to resources that have been allocated to a virtual machine but have not been used by the virtual machine internal applications. The resource usage rules are different for different applications. To improve the accuracy of the prediction, it is first distinguished by fast fourier transform whether the resources are periodic (i.e., predictive), all the predictive services are predicted for a future period of time at regular intervals, and a periodic "fraction unused" fragmentation resource future window is reported to the scheduler, while an aperiodic "fraction unused" fragmentation resource is collected per second and reported to the scheduler.
4. Phase space
The facies space records the amount of unallocated resources, the periodic score but the amount of unused resources for a period of time in the future, and the aperiodic score but the amount of unused resources for the current time.
6. The scheduling unit is based on a function filling type mixed part scheduling method of the fragment resource portrait, and when the scheduler receives a function scheduling request, the scheduler decides an optimal scheduling scheme according to an unallocated resource state, a distributed unused resource state and a function request resource amount. A reinforcement learning method is adopted to establish a resource state-scheduling decision learning model, and the resource scheduling is carried out with the aim of maximizing the long-term effective throughput rate.
5.1 function describer
The function carver monitors the resource usage amount of each newly arrived function during execution, records the execution time after execution is completed, further gives a two-dimensional required resource label of the function resource and the execution time, directly gives a corresponding label to the same function and sends the same function to the function scheduler for scheduling.
6.2 function scheduler
And the function scheduler performs variance calculation on the phase space views of all the servers in the whole situation, the variance represents the resource stability of the servers, and stability descending arrangement is performed on all the nodes. After receiving the plotted function request, the function scheduler schedules the function with the execution time larger than 1s in a node phase space which has optimal stability and does not allocate resources plus periodic points and does not use the resources larger than the needed resources in the function execution time; and scheduling the function with the execution time less than 1s in the non-periodic resources with the resource amount meeting the non-periodic time.
The present invention is not limited to the above-described embodiments. The foregoing description of the specific embodiments is intended to describe and illustrate the technical solutions of the present invention, and the above specific embodiments are merely illustrative and not restrictive. Those skilled in the art can make many changes and modifications to the invention without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (6)
1. A function filling model based on resource fragment space-time feature perception is applied to a public cloud system containing virtual machines, and is characterized in that: the filling model comprises an unallocated resource space-time predictor, a divided and unused resource space-time predictor, a periodic classifier, a resource phase space and a scheduling unit; wherein:
the unallocated resource space-time predictor is used for analyzing a virtual machine creation time interval of the node to predict a fluctuation time point of the next unallocated resource to obtain an unallocated resource amount in the future time;
the point-unused resource space-time predictor is used for predicting the point-unused resource amount of the periodic virtual machine service in the future time and collecting the non-periodic point-unused resource amount of the non-periodic virtual machine every second;
the periodic classifier is used for dividing the resources of the virtual machine on the node into periodicity and influencing the prediction range of the spatial-temporal predictor without using the resources according to the division result;
the phase space is used for recording the unallocated resource amount, the periodic points but the unused resource amount in a future period of time and the non-periodic points but the unused resource amount in the current time;
and the scheduling unit outputs an optimal matching resource scheduling scheme according to the unallocated resource amount, the unused resources and the function request resource amount.
2. The function filling model based on resource fragment space-time feature perception according to claim 1, wherein: the scheduling unit comprises a function carver and a function scheduler; wherein:
the function carver obtains the required resource amount at any time in the execution time of the function carver by carving the function, and sends the carved function to the function scheduler for scheduling;
and the function scheduler selects the node with optimal stability and sufficient resource supply in the function execution time to deploy the function service by analyzing the phase space resource stability of each global node.
3. A function filling method based on resource fragment space-time feature perception is characterized by comprising the following steps:
the unallocated resource predictor predicts the unallocated resource amount in a future period of time on the nodes in a discrete stable period and a local intensive period according to the time interval of the virtual machine and periodically sends the unallocated resource amount to a phase space;
the score unused resource predictor is used for predicting the size of available non-periodic resources at the current moment through a deep neural learning model after predicting the periodic score and unused resources and uniformly sending the size to a facies space;
and the scheduling unit sends an optimal scheduling scheme to the phase space according to the function scheduling request according to the resource state of 'unallocated', the resource state of 'unused' and the function request resource quantity.
4. The method as claimed in claim 3, wherein the unallocated resource predictor periodically sends the amount of unallocated resources in a future period to the facies space process:
initializing the locally intensive creation track input of all virtual machines on a node by the unallocated resource space-time predictor;
the unallocated resource space-time predictor judges whether the current time is in a discrete stable period or a local intensive period by analyzing the virtual machine creation time interval on the current node, extracts an intensive creation track with the closest average interval time when the unallocated resource space-time predictor is in the local intensive period, calculates the shortest creation time, the longest creation time and the average resource amount of the track, sets the time range from the shortest creation time to the longest creation time as the fluctuation time of the unallocated resource next time, and sets the average resource amount as the reduction amount of the unallocated resource next time;
the unallocated resource spatiotemporal predictor converts the corresponding resource amount from the unallocated resource to the non-periodic sharing of the unused resource near the predicted virtual machine creation time, and after the predicted time is exceeded, no new real virtual machine is created, and then the corresponding periodic sharing of the unused resource is converted back to the unallocated resource.
5. The method as claimed in claim 3, wherein the score-unused resource predictor predicts score-unused resource transmit phase-space process:
and the resource space-time predictor is used for predicting only the periodic resources divided by the periodic classifier, predicting the size of the periodic resources in each time scale in the future through a deep neural learning model, and calculating the variance between the predicted value and the consumption value of the virtual machine during execution, wherein when the variance is greater than a threshold value, the corresponding periodic resources are converted into non-periodic resources.
6. The method for filling function based on resource fragment space-time feature perception according to claim 3,
the scheduling unit sends the optimal scheduling scheme to the phase space according to the function scheduling request according to the resource state of 'unallocated', 'shared but unused' and the function request resource quantity:
the function carver obtains the resource usage amount and the execution time of the function after executing each newly arrived function, and schedules a function request of a two-dimensional required resource label;
the function scheduler schedules the function request of the required resource label in a node phase space with optimal stability, unallocated resources plus periodic points and unused resources greater than the required resources in the function execution time according to the function with the execution time greater than 1 s;
and scheduling the function with the execution time less than 1s in the non-periodic resources with the resource amount meeting the non-periodic time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211258523.7A CN115840638A (en) | 2022-10-14 | 2022-10-14 | Function filling model based on resource fragment space-time feature perception and method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211258523.7A CN115840638A (en) | 2022-10-14 | 2022-10-14 | Function filling model based on resource fragment space-time feature perception and method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115840638A true CN115840638A (en) | 2023-03-24 |
Family
ID=85575826
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211258523.7A Pending CN115840638A (en) | 2022-10-14 | 2022-10-14 | Function filling model based on resource fragment space-time feature perception and method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115840638A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116401055A (en) * | 2023-04-07 | 2023-07-07 | 天津大学 | Resource efficiency optimization-oriented server non-perception computing workflow arrangement method |
-
2022
- 2022-10-14 CN CN202211258523.7A patent/CN115840638A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116401055A (en) * | 2023-04-07 | 2023-07-07 | 天津大学 | Resource efficiency optimization-oriented server non-perception computing workflow arrangement method |
| CN116401055B (en) * | 2023-04-07 | 2023-10-03 | 天津大学 | Resource efficiency optimization-oriented server non-perception computing workflow arrangement method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jiang et al. | Joint task offloading and resource allocation for energy-constrained mobile edge computing | |
| WO2020206705A1 (en) | Cluster node load state prediction-based job scheduling method | |
| CN107743100B (en) | Online adaptive network slice virtual resource allocation method based on service prediction | |
| CN109271015B (en) | Method for reducing energy consumption of large-scale distributed machine learning system | |
| CN102004671B (en) | Resource management method of data center based on statistic model in cloud computing environment | |
| CN111026553B (en) | Resource scheduling method and server system for offline mixed part operation | |
| CN111491006A (en) | Load-aware cloud computing resource elastic distribution system and method | |
| CN110858161A (en) | Resource allocation method, device, system, equipment and medium | |
| CN109150738B (en) | Industrial internet resource management method and system, readable storage medium and terminal | |
| CN110990138B (en) | Resource scheduling method, device, server and storage medium | |
| CN113535409B (en) | Server-free computing resource distribution system oriented to energy consumption optimization | |
| CN111628887B (en) | Internet of things slice resource distribution system and method, electronic equipment and storage medium | |
| CN115408152A (en) | Adaptive resource matching obtaining method and system | |
| CN110990160B (en) | Static security analysis container cloud elastic telescoping method based on load prediction | |
| Al-Sinayyid et al. | Job scheduler for streaming applications in heterogeneous distributed processing systems | |
| CN109936471B (en) | Multi-cluster resource allocation method and device | |
| CN115840638A (en) | Function filling model based on resource fragment space-time feature perception and method thereof | |
| Lu et al. | On the performance-driven load distribution for heterogeneous computational grids | |
| Tuli et al. | SplitPlace: AI augmented splitting and placement of large-scale neural networks in mobile edge environments | |
| Abdullah et al. | A reliable, TOPSIS-based multi-criteria, and hierarchical load balancing method for computational grid | |
| CN114741200A (en) | Data center station-oriented computing resource allocation method and device and electronic equipment | |
| EP4300305A1 (en) | Methods and systems for energy-efficient scheduling of periodic tasks on a group of processing devices | |
| CN117076882A (en) | Dynamic prediction management method for cloud service resources | |
| Kanagaraj et al. | Uniform distribution elephant herding optimization (UDEHO) based virtual machine consolidation for energy-efficient cloud data centres | |
| Salimian et al. | Energy-efficient placement of virtual machines in cloud data centres based on fuzzy decision making |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |