Nothing Special   »   [go: up one dir, main page]

CN101788920A - CPU virtualization method based on processor partitioning technology - Google Patents

CPU virtualization method based on processor partitioning technology Download PDF

Info

Publication number
CN101788920A
CN101788920A CN201010108081A CN201010108081A CN101788920A CN 101788920 A CN101788920 A CN 101788920A CN 201010108081 A CN201010108081 A CN 201010108081A CN 201010108081 A CN201010108081 A CN 201010108081A CN 101788920 A CN101788920 A CN 101788920A
Authority
CN
China
Prior art keywords
core
subregion
cpu
domain
general purpose
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
Application number
CN201010108081A
Other languages
Chinese (zh)
Inventor
张炯
龙翔
吕紫旭
高小鹏
白跃斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beihang University
Original Assignee
Beihang University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beihang University filed Critical Beihang University
Priority to CN201010108081A priority Critical patent/CN101788920A/en
Publication of CN101788920A publication Critical patent/CN101788920A/en
Pending legal-status Critical Current

Links

Images

Landscapes

  • Multi Processors (AREA)

Abstract

The invention provides a CPU virtualization method based on processor partitioning technology, which comprises the following steps: aiming at a multi-core or many-core platform, firstly, dividing a CPU core into a management core (H-Core) partition and a plurality of general core (G-Core) partitions; then, allocating management tasks and privileged domains in a VMM into the H-Core partition for operating; dividing general client domains operating on the VMM into different virtualization task subsets according to individual characteristics; and allocating the virtualization task subsets into the corresponding G-Core partitions for operating. Different scheduling policies and resource allocation policies can be realized aiming at different types of virtualization tasks in different CPU partitions. The CPU virtualization method of the invention improves the safety and the reliability of the virtualization system on the multi-core or many-core platform, reduces the complexity of CPU virtualization, simultaneously simplifies the complexity of overall resource allocation and scheduling, improves the capacity of the VMM for allocating and scheduling computing resources, and has the advantages of safety, reliability, high efficiency and high performance.

Description

A kind of CPU virtual method based on processor partitioning technology
Technical field
The present invention proposes a kind of CPU virtual method, relates in particular to towards multinuclear or many nuclear platforms, and the CPU virtual method based on processor partitioning technology belongs to computer operating system and technical field of virtualization.
Background technology
Along with the development of information industry, all trades and professions improve constantly the requirement of arithmetic capability.Meanwhile, be subjected to the restriction of microelectronics production technology, traditional method of passing through to improve circuit level, CPU frequency can not solve such as problems such as design complexity, chip power-consumption, circuit leakage current, CPU heat radiations.Therefore, increasing processor production firm turns to chip multiprocessors (CMP, Chip Multi-processing) technology and while multithreading (SMT, SimultaneousMulti-Threading) technology utilizes the Thread-Level Parallelism of processor to improve the interior processing power of single processor encapsulation.SUN company had released the processor-server Niagara 2 that is integrated with 32 threads already at present, Intel Company has also designed the picture processing chip Larrabee that is integrated with 32 processor cores, at present the processor products such as double-core, four nuclears of Intel and AMD have been popularized in personal desktop, commerce server field, calculate the field especially at a plurality of polycaryon processors of mainboard deploy in science, in single computer systems, processor core quantity is than greatly increasing before.Can predict, will be integrated with more processor core or processor thread in following single processor encapsulation, its application will extensively cover fields such as server, high-performance calculation, Flame Image Process, embedded system, and following multi-core platform will be packaged with more processor core in single-chip crowd examines the platform development.
Along with improving constantly of computing power, under the prerequisite that guarantees safety and isolate, make full use of system resources in computation, Intel Virtualization Technology arises at the historic moment, its basic thought is to utilize virtual machine monitor (VMM) that the bottom hardware resource is managed, provide a plurality of virtual execution environments that are isolated from each other to the upper strata simultaneously, these virtual hardware execution environments become a virtual machine (Virtual Machine, be called for short VM), thus make and can move a plurality of different Client OSs simultaneously on the physical host.As far back as the sixties in 20th century, IBM has just used the system virtualization technology on 360 serial large-scale machines, server resource is distributed to a plurality of users with multiplexing form.In recent years, Intel Virtualization Technology has all had application widely in Server Consolidation, service isolation, Software deployment, science calculating and security fault-tolerance field.At present a plurality of successful virtual projects occur, comprised versions such as the VMWare Server of VMWare company and Workstation, the SolarisContainer of Sun Microsystems etc., and the KVM of the community that increases income and Xen etc.Most of at present existing Intel Virtualization Technology, by between original system software (operating system) and hardware platform (bare machine), inserting the software thin layer, use directly and carry out, binary translation (Dynamic Binary Translation), hardware is assisted virtual (Hardware-assistedvirtualization) or general virtual (Para-virtualization) technology, realize the parallel running of a plurality of Client OSs on multi-core platform, thereby improved the overall utilization rate of CPU, and can reduce cost, streamlining management, improve security and reliability etc.
Dummy machine system is controlled by VMM, a plurality of separate territories (Domain) are provided on it, all include complete virtual hardware execution environment in each territory, the basic calculating resource that provides comprises processor, storer and equipment, is used for moving Client OS separately and carries the actual calculation task.Resources allocation and task scheduling between the territory are controlled by VMM, and are responsible for each other isolation and protection.
In order to utilize the improvement on the performance that polycaryon processor brings, existing virtualization system is mostly continued to use the symmetric multiprocessor (SMP of original operating system, Symmetrical Multi-Processing) support pattern of system, the performance improvement that can not utilize the Thread-Level Parallelism technology of multi-core platform to bring fully equally also can't utilize the improvement of the Thread-Level Parallelism of many nuclear platforms.Typical in the Credit dispatching algorithm among the Xen, all core cpu in the VMM control system, for each core cpu is set up a scheduling queue, the CPU time of distributing corresponding ratio globally according to the CPU allocation proportion of design in advance for each Domain, and according to the time corresponding priorities, from high to low all Domain of scheduling in each cpu queue.When the last load of certain CPU was overweight, VMM was responsible for the task on it is sought an idle a little core cpu in the overall situation, and task immigration is arrived the target core cpu.Under the less situation of CPU scale, the distribution of this overall situation and operating strategy can satisfy the demands, but increase along with integrated processor core number in the computing system, this method can't satisfy the calculated amount of the distribution rapid expansion corresponding with operating strategy of the overall situation soon, finally can't satisfy system and examine dispatching requirement on the platform the crowd.
Simultaneously, traditional CPU virtual method manages all CPU by VMM is unified, and they is distributed to different client application according to certain scheduling strategy, the different level of privilege requirements that do not embody different client application and had.For example, in Xen, have franchise Domain such as Domain0, bear such as functions such as administration view, Domain establishment, device model, rear end drivings, and all have authorization code in these functions, some also needs to operate in the root user pattern.Traditional CPU virtual method, franchise Domain and Universal Client Domain are put together, on the CPU of overall situation view, dispatch, franchise Domain appears in meeting and Universal Client Domain seizes same core cpu each other, on same core cpu, replace execution, thereby can cause the decline of entire system reliability.In addition, traditional C PU virtual method can't guarantee that franchise Domain can have enough computational resources, can the system performance bottleneck occur in these key areas in some cases.Therefore, be necessary to utilize the CPU partitioning technique that the CPU at franchise Domain place and the CPU at Universal Client Domain place are carried out physical segmentation.
In addition, traditional CPU virtual method uses unified scheduling strategy, can not satisfy the characteristic of different client application.Because the development of computing power in order to improve the utilization factor of system, can be disposed the Universal Client Domain of number of different types usually in actual multinuclear or many karyonides system.Among these Domain, some application is had relatively high expectations to time response, needs to adopt response time parameter scheduling strategy preferably; Some application is had relatively high expectations to I/O or network throughput, then needs to adopt the scheduling strategy that device access is optimized.In the CPU subregion at Universal Client Domain place, we can further carry out subregion, are the different scheduling strategy of CPU subregion distribution of different Universal Client Domain, thereby the running environment of the separation that is fit to its characteristic is provided for different application.
Therefore, how to make full use of the system processor resource of multinuclear or many nuclear platforms, the reliability and security of raising system, the performance of while optimization system, adapt to more targeted environment, support the more applications environment, become and in Intel Virtualization Technology, use CPU partitioning technique key issue to be solved.
Summary of the invention
The present invention proposes a kind of CPU virtual method based on processor partitioning technology, this method improvement the virtualized implementation of existing virtual machine monitor CPU, core cpu under multinuclear or the many nuclear platforms is divided into management core subregion and a plurality of general purpose core subregion, H-Core (Hyper-Core, management core) is assigned portion C PU core, franchise Domain of operation and management role on it, remaining core cpu is as G-Core (General-Core, general purpose core), and a plurality of G-Core subregions have been divided into according to the constraint condition that will move Universal Client Domain among the VMM dissimilar, and different scheduling strategies is set, thereby in different G-Core subregions, dispose and dispatch dissimilar Universal Client Domain for different G-Core subregions.
A kind of CPU virtual method based on processor partitioning technology that the present invention proposes specifically comprises the steps:
Step 1: virtual machine monitor VMM provides H-Core subregion and a plurality of G-Core subregion to the processor resource subregion of multinuclear or many nuclear platforms; During system start-up, virtual machine monitor VMM is with the processor core in multinuclear or the many nuclear platforms, be divided into management core H-Core and general purpose core G-Core two class CPU subregions, general purpose core G-Core class subregion wherein, dissimilar according to the constraint condition of the Universal Client Domain of system expection needs operation, be divided into the general purpose core G-Core subregion of respective amount, VMM is that each CPU subregion makes up data structure CPU subregion BITMAP (bitmap), the concurrent physical processor ID that this subregion of record has in CPU subregion BITMAP, BITMAP creates different CPU subregions by traversal CPU subregion, the different CPU subregion is assigned the core cpu that quantity does not wait, by the VMM unified management;
Step 2: from the management correlation function of VMM and Domain and code, isolate management role H-Task and franchise Domain H-Domain among the VMM, they are deployed to management core H-Core subregion in the step 1;
Step 3: the Universal Client Domain that moves on multinuclear or the many nuclear platforms is classified, and be deployed to corresponding general purpose core G-Core subregion; When Universal Client Domain starts,, be divided in the different virtualized tasks subclass according to the constraint condition of himself, then, in the general purpose core G-Core subregion of the correspondence that VMM is deployed to each virtualized tasks subclass in the step 1 to be divided;
Step 4: make up the scheduling sublayer module in the general purpose core G-Core subregion, and design the scheduling strategy that meets; In system's operational process, VMM makes up the scheduling sublayer module in each general purpose core G-Core subregion, the scheduling sublayer module of each general purpose core G-Core subregion, have separately independently scheduling queue and scheduling strategy, be responsible for the operation of Universal Client Domain in the scheduling subregion, and be responsible for safeguarding the resources allocation of the Universal Client Domain in this subregion, and configuration meets the various strategies of the Universal Client Domain constraint condition of moving it on;
Step 5: in the operational process of system, VMM carries out load balancing to general purpose core G-Core subregion and virtualized tasks subclass; Task scheduling, resources allocation work in each general purpose core G-Core subregion, finish by the scheduling sublayer module in each subregion, when computational resource in the general purpose core G-Core subregion breaks down or virtualized tasks subclass when changing, VMM adjusts the virtualized tasks subclass in general purpose core G-Core by stages, perhaps repartition general purpose core G-Core subregion, the virtualized tasks subclass is deployed in the new general purpose core G-Core subregion.
Described constraint condition comprises: response time requirement, device access characteristic, reliability properties, security feature.
The method that the subclass of virtualized tasks described in the step 3 is disposed, specifically when Universal Client Domain starts, set up mapping relations between the general purpose core G-Core subregion of VMM with the Universal Client Domain in this virtualized tasks subclass and its distribution, the general purpose core G-Core subregion BITMAP that specifies this client computer Domain to move by CPU mapping parameters in the Domain structure, thus a class virtualized tasks subclass and corresponding general purpose core G-Core subregion binding made.
The operation of Universal Client Domain in the module schedules of scheduling sublayer described in the step 4 subregion, specifically the VCPU (virtual processor) at the Universal Client Domain correspondence of moving in the general purpose core G-Core subregion carries out, VCPU is come it is dispatched by the scheduling code on each concurrent physical processor by in the scheduling queue of the scheduling sublayer module assignment of place general purpose core G-Core subregion each physical cpu correspondence in the general purpose core G-Core subregion of place then.
Compared with prior art, the present invention has following advantage and good effect:
(1) simplified the virtualized difficulty of CPU under multinuclear or the many nuclear platforms
Under multinuclear or many nuclear platforms, increasing core cpu resource improved the computational complexity that centralized management brought greatly, and under most of scene, single client computer Domain can't take whole core cpu resources.By processor is carried out subregion, virtual machine monitor does not need directly core cpu to be assigned to each independent clients Domain, and just need be according to the classification of task, earlier core cpu is carried out subregion, then with dissimilar task deployments in different CPU subregions.The right computational resource scale of each subregion demand side significantly reduces like this, and needs the computation requirement of the virtualized tasks of serving seemingly closer on each subregion, thereby can be different task scheduling of different zoning design and resource allocation policy easily.As long as it is just passable with the virtualized tasks assignment that virtual machine monitor is finished the maintenance of CPU subregion in system's operational process, therefore, reduced the virtualized difficulty of CPU under multinuclear or the many nuclear platforms.
(2) extendability and the adaptability of CPU virtual method have been improved
The present invention has reduced virtual machine monitor in the difficulty of CPU aspect virtual, thereby make virtual machine monitor can support more physical cpu quantity, when the physical processor core number significantly increased along with the development of many nuclear technology, the method that the present invention proposes had better extensibility; In addition, support different application demands by different CPU subregions, by design and realization more applications kind, method provided by the invention can be supported more applications, has good extendability.
Pass through partitioning technique; can in different subregions, use different scheduling strategy and resource allocation policy; and can not increase the overall complexity that virtual machine monitor is realized; thereby can support to exist in the same system multiple different applied scene (as in science calculating field; there are some system monitoring tasks to have higher real-time requirement; general operation task then needs for a long time not operation with shutting down), thus the more applications scene adapted to.
(3) improved the reliability and security of system
It is virtual to carry out CPU by processor partitioning technology, the client computer Domain of different privilege levels can be cut apart, management role in franchise Domain and the virtual machine monitor is assigned to the H-Core subregion to be moved, thereby the running environment (CPU time of franchise Domain and management role, memory source, hardware device) can not be subjected to the influence of other Domain, avoided in original CPU virtual method, privilege Domain and Universal Client Domain operate on the same CPU, thereby by the situation of the application of unauthenticated among Universal Client Domain influence.
In addition, by the reliability design and the safety Design of different stage are provided, reliability and security design and difficulty and the cost realized, the reliability and the security that have improved system from another aspect have been reduced in the G-Core subregion.
(4) multinuclear or many overall performances of examining virtualization system under the platform have been improved
Because the scale of CPU subregion is less with respect to entire system, therefore can in subregion, adopt comparatively complicated optimizing scheduling and resource allocation optimization easily, and the subregion planted agent is with having similar demand, so can more convenient design optimisation strategy targetedly.
In addition, to the redeploying of failure condition, and concern the computational resource of to distribute according to need during operation according to the deployment that environmental demand is adjusted subregion and task, farthest guarantee the computational resource of system core task, thereby can obtain the better overall performance.
Description of drawings
The flow chart of steps that Fig. 1 proposes for the present invention based on the CPU virtual method of processor partitioning technology;
Fig. 2 is the exemplary block diagram of embodiment of the invention virtual machine monitor;
The static illustrative diagram of Fig. 3 in the two 4 nuclear platforms of the embodiment of the invention some virtualized tasks being disposed.
Embodiment
Below in conjunction with accompanying drawing the present invention is further described in detail.
In the embodiment of the invention, based on bare machine virtual (Bare-metal virtualization) method, a kind of CPU virtual method based on processor partitioning technology has been proposed, characteristics at multinuclear or many nuclear platforms, with the classification of task of the numerous processor cores in the system, be divided into the G-Core subregion of H-Core subregion and respective amount according to expection; Franchise Domain and the management role moved in the virtualization system are deployed among the H-Core, are its distributing independent core cpu and computational resource; The Universal Client Domain that moves in the virtualization system according to characteristics such as time-constrain, IO visiting demand, security and reliability requirements, is divided into a plurality of virtualized tasks subclass, is deployed in the corresponding G-Core subregion; Meanwhile, on different G-Core subregions, characteristic according to the virtualized tasks subclass of operation on it, use own scheduling strategy, resource allocation policy, equipment simulating method and access control policy, thereby can adapt to of the different requirements of different virtual task subclass at aspects such as response time, calculated performance, security and reliabilities.
Below in conjunction with embodiment and accompanying drawing, further describe.
A kind of CPU virtual method based on processor partitioning technology comprises the steps, as shown in Figure 1:
Step 1 is carried out subregion to the processor resource on multinuclear or the many nuclear platforms, and H-Core subregion and a plurality of G-Core subregion are provided.
During system start-up, virtual machine monitor is with the processor core in multinuclear or the many nuclear platforms, be divided into H-Core and G-Core two class CPU subregions, G-Core class CPU subregion is expected according to system needs the type of the constraint condition of the client computer Domain of operation to be divided into the different G-Core subregion of respective amount.These subregions have been assigned with core cpu and other computational resources that quantity does not wait, and the core cpu that the different CPU subregion does not wait according to the computation requirement quantity allotted by the VMM unified management, is waited for the deployment of H-Task, H-Domain and G-Domain.System's expection needs the constraint condition of the client computer Domain of operation specifically to set forth in step 3.
Privilege Domain, promptly H-Domain refers to have higher level of privilege, carries out the virtual client computer Domain of privilege task, has born such as functional modules such as administration view, Domain establishment, device model, driving Domain.For example, the Domain0 in the Xen system, and Driver-Domain, stub-Domain etc.
Management role, promptly H-Task refers to operate among the VMM, has the mission critical of management characteristic, as mission criticals such as system health condition monitoring, scheduling decision, interrupt responses, operates in privileged mode, and is most important to the normal operation of system.For example, being responsible for the parts such as scheduler, access control and event channel of maintenance system scheduling decision in Xen, all is key points of security of system reliability service.
Universal Client Domain, i.e. G-Domain is generally the calculation task of the required service of virtual platform.Universal Client Domain can comprise multiple application, has different constraint requirements.
VMM the unloading phase, be that each CPU subregion makes up data structure---CPU subregion BITMAP, and in CPU subregion BITMAP, preserve the actual physical processor ID that has in this subregion, thereby can safeguard the structure of CPU subregion; In the startup and operational process of VMM, VMM is responsible for detecting the BITMAP of each subregion, and safeguard finger daemon in each subregion, make the VMM task carried out in the finger daemon and client computer Domain all by physical restriction in this subregion, thereby realize the Physical Extents of CPU.
Physical Extents by CPU subregion BITMAP realization, dispatch with respect to layering scheduling and subregion that scheduler is realized, can limit computational resource such as VMM task and CPU that client computer Domain can visit the subregion from system level, thereby can obtain the inaccessiable reliability and security of scheduling strategy in the realization of VMM upper strata.
In two 4 nuclear platform example systems that execution science is as shown in Figure 3 calculated, need the multiple different application of operation, in the present embodiment these Universal Clients Domain is divided into 4 class virtualized tasks subclass according to response time requirement, device access characteristic, the task subclass formed of privileged trading Domain and management role in addition, we are divided into all 8 core cpu F0-F7 in 4 different G-Core subregions and 1 the H-Core subregion, and processor core quantity corresponding in each CPU subregion is followed successively by 1,2,2,2,1.In the VMM start-up course, 5 CPU subregions of initialization bitmap (BITMAP) at first, when VMM at BP (Bootstrap Processor, bootstrap processor) after starting on and finishing the initial work of CPU, BP will oneself join first CPU BITMAP, and then BP can start remaining 7 AP (Application processor) successively.The process that starts AP is similar with the process that starts BP, and after the initialization of the CPU of AP was finished, we joined them among 5 CPU BITMAP respectively.So far, for this virtualization system has made up 5 CPU BITMAP, wherein deposited the ID of corresponding whole 8 processor core correspondences.As shown in Figure 3, core cpu F0 is BP, divides in a G-Core subregion, core cpu F1-F7 is AP, and F1, F2 divide in the 2nd G-Core subregion, and F3, F4 divide in the 3rd G-Core subregion, F5, F6 divide in the 4th G-Core subregion, and F7 divides in the H-Core subregion.
Step 2 is isolated H-Domain and H-Task, and they are deployed in the H-Core subregion.
In the virtualization system operational process, the system monitoring module, policy selection module of the normal operation of a series of assurance systems etc. are arranged in VMM, they operate on the level of privilege, have higher priority, and the normal operation of system is played crucial effects.And functional modules such as the administration view of moving among some client computer Domain, Domain establishment, device model, the operation to system plays an important role equally.These franchise Domain and management role are the key modules of system.
Step 1 has been distributed a H-Core subregion for these franchise Domain and management role, guarantees that they have independent CPUs and computational resource, can not disturbed by other calculation tasks.VMM separates H-Domain and H-Task from system, and set up mapping management with between they and the H-Core subregion, thereby guarantee that these mission criticals are served by H-Core independently, thereby the computational resource and the CPU time that guarantee mission critical are fully guaranteed, the H-Core subregion is also isolated other calculation tasks in these mission criticals and the system in addition, thereby has guaranteed the reliability service of system.
As shown in Figure 3, in the present embodiment, according to franchise Domain and the needed computational resource of management role, F7 is assigned as the H-Core subregion, and above-mentioned franchise Domain and management role are deployed in wherein.In the H-Core subregion, F7 as proprietary core cpu, is offered the operation of franchise Domain and management role specially.
Because the physical isolation of the CPU subregion that step 1 realized, thereby other non-critical task on the G-Core subregion can not be visited institute's distribution calculation resource in the H-Core subregion, equally also can't interfere with the operation of mission critical.
Step 3 is classified to a plurality of Universal Client Domain that move on multinuclear or the many nuclear platforms, and is deployed to corresponding G-Core subregion.
In system's operational process, when Universal Client Domain creates, according to its characteristic, be categorized as different virtualized tasks subclass, VMM is deployed to different virtualized tasks subclass in the different G-Core subregions of step 1 division.
Because multinuclear or many nuclear platform computing powers are powerful, generally can dispose the virtualized tasks of number of different types on it, to make full use of its computational resource.And each task all has its constraint condition, generally comprise: response time requirement (requiring in real time to require), device access characteristic (IO intensive applications and CPU intensive applications), reliability properties (key area is used with unessential field and used), security feature (security sensitive is used and the insensitive application of safety) etc. with non real-time, as shown in table 1.
Figure GSA00000019630800081
Table 1
The task of various boundary conditions needs strategies and its adaptation such as different scheduling, resources allocation.The embodiment of the invention has been chosen server has been used the constraint condition that directive significance is all arranged with embedded calculating, comprise that response time requirement, device access characteristic classify, task-set is divided into: real-time IO intensive applications, real-time CPU intensive applications, non real-time IO intensive applications, non real-time CPU intensive applications.It is corresponding respectively:
The IO intensive applications---real-time data acquisition is used in real time, needs and can obtain data designated in the time limit in regulation, and may relate to a large amount of device accesses;
CPU intensive applications---Industry Control Application need by a large amount of calculating, be finished the control to production line at the data of external world's sampling acquisition in real time, needs in time to make response, needs to carry out a large amount of relatively calculating simultaneously;
Non real-time IO intensive applications---data-storage applications, after getting access to lot of data, it need be kept in the outside mass-memory unit, for subsequent treatment, and this class application need be carried out a large amount of relatively device accesses usually, and is then not high to the time requirement of visit;
Non real-time CPU intensive applications---off-line data is handled and is used, for supporting industry control is used, often need further analyze the data of obtaining, thereby make that the data of storage are reasonable more and effective, these application do not have the requirement of response time, but need take a large amount of processor time.
The present invention requires Universal Client Domain to classify according to the difference of task, thereby has simplified the complicacy that polytype task is provided support and brought.The embodiment of the invention is in the structure of Domain correspondence, be provided with CPU mapping parameters domain_cpu_map and be used for specifying relation between Domain and the CPU BITMAP, special structure is set again, at this Domain and the last characteristic that is had of using thereof, for structure is given different parameters, comprise and be used to specify parameter is_rt_app that whether has real-time to require and the rt_config of association that indicates the real-time characteristic parameter; Be used for the parameter is_cpu_bound of indication equipment access characteristics, and the parameter c pu_weight etc. that uses shared load, be used for indicating grouping of Domain place and the performance requirement that has.
VMM is responsible for different Universal Client Domain is deployed in the different G-Core subregions according to the requirement of its computational resource.When virtualized tasks starts, set up mapping relations between the CPU subregion that VMM is distributed Universal Client Domain and its virtualized tasks subclass, the CPU BITMAP that specifies this application to move by CPU mapping parameters in the Domain structure, thus the virtualized tasks subclass of a certain class and the corresponding CPU subregion binding of satisfying the Domain performance requirement made.As shown in Figure 3, in the present embodiment, the data storage subclass of non real-time IO intensive task correspondence is deployed in the G-CORE subregion that step 1 distributes, is moving data storage Domain and the intensive virtualized tasks of other similar IO on this G-CORE subregion; The real-time task subclass of real-time IO intensive task correspondence is deployed in the 2nd G-CORE subregion, is moving device drives Domain and data acquisition Domain on this G-CORE subregion; The scientific computing subset of non real-time CPU intensive task correspondence is deployed in the 3rd G-CORE subregion, is moving on this G-CORE subregion and calculating operation Domain and calculate acceleration Domain; The data processing subclass of real-time CPU intensive task correspondence is deployed in the 4th G-CORE subregion, is moving data processing Domain and the intensive virtualized tasks of other similar CPU on this G-CORE subregion.Such distribution method satisfies the demand of different task to computational resource, and can adopt different optimal way in different subregions.
The virtualized tasks subclass is deployed to corresponding G-Core partition running, and its computational resource of participating in the distribution and can't use computational resource outside the subregion all in this G-Core subregion; Simultaneously, the Universal Client Domain in the different G-Core subregions does not influence each other.
In the embodiment of the invention, by the BITMAP of the VCPU CPU subregion that can move among the control Domain, thereby limit the physical cpu subregion at corresponding Domain place, thereby realize that dissimilar virtualized tasks subclass is deployed in the different CPU subregions.Operation queue that VCPU is constructed and waiting list make up on the physical cpu in the CPU subregion BITMAP of its place.
Step 4, VMM makes up the scheduling sublayer module in the G-Core subregion, and designs the scheduling strategy that meets.At different G-Core subregions, the strategy of aspect such as the constraint condition characteristic according to virtualized tasks subclass on it disposes satisfactory task scheduling, resources allocation, and is safe and reliable.
In system's operational process, whole Universal Client Domain is limited in finishing in the corresponding G-Core subregion in the virtualized tasks subclass, VMM sets up independently task queue and carries out task scheduling and resources allocation in each G-Core subregion, each G-Core subregion is selected the scheduling strategy and the resource allocation policy that are fit to according to the constraint condition of virtualized tasks subclass in the subregion.
Because each virtualized tasks subclass is to be made of the application with approximate characteristic, therefore the present invention can be at different application types, define different task scheduling, resources allocation, reliability properties, and owing to move corresponding virtual task subclass on each G-Core subregion, therefore just be that each G-Core subregion is realized the support to a kind of type application, just realize CPU virtual method based on subregion.
In virtual machine monitor, by design scheduling sublayer module in each G-Core subregion, these scheduling sublayer modules provide only scheduling strategy according to the constraint condition of the virtualization task subclass in the G-Core subregion.Scheduling in the G-Core subregion, VCPU at the Universal Client Domain correspondence of moving in this subregion carries out, VCPU is assigned in the scheduling queue of each physical cpu correspondence in the subregion, is come it is dispatched by the scheduler on each concurrent physical processor.
Scheduler on the concurrent physical processor is meant the scheduling code that moves on the concurrent physical processor.Scheduling code on each concurrent physical processor is responsible for the operation of VCPU scheduling queue on this processor.
In this step, the present invention is embodiment illustrated in fig. 3 to have made up the scheduler (Scheduler) that is suitable for task subset property on it for each G-Core subregion, satisfies its different characteristics at aspect such as real-time and data-intensive or CPU be intensive respectively.When a physical cpu free time gets off, VMM can put into operation by the VCPU of policy selection according to scheduler from the VCPU current self waiting list, because the restriction of step 1 and step 3, this VCPU must be the VCPU that is deployed among the Universal Client Domain of this physical cpu place G-Core subregion, and the scheduling strategy of scheduler, then the characteristic at the virtualized tasks subclass of being disposed in the current G-Core subregion makes up, thereby has guaranteed the optimization of performance in the subregion.
Step 5 in the operational process of system, is carried out load balancing to general purpose core G-Core subregion and virtualized tasks subclass.
The normal operating phase of resource requirement in satisfying the G-Core subregion, VMM only need control the deployment of G-Core subregion and task, be responsible for safeguarding that the VCPU of initiate Universal Client Domain correspondence and the corresponding relation of G-Core subregion BITMAP get final product, work such as task scheduling, resources allocation are finished by the subtask of VMM in the subregion in each G-Core subregion.Computational resource breaks down in the G-Core subregion, and when perhaps the virtualized tasks subclass changed, VMM need redistribute the G-Core subregion, perhaps the virtualized tasks subclass was re-deployed in the new G-Core subregion.
When needs redeploy the virtualized tasks subclass, VMM then needs all VCPU on the required virtualized tasks subclass that redeploys are stopped, again safeguard the mapping relations of the BITMAP of all VCPU and G-Core subregion in the virtualized tasks subclass afterwards, then use the virtual machine (vm) migration technology, VCPU is moved among the BITMAP of corresponding G-Core subregion.Here require the target G-Core subregion of migration to satisfy by of the requirement of migration task subclass to computational resource.
The deployment of virtualized tasks subclass and distribution can be static the deployment, also can be dynamically to adjust.The embodiment of the invention is for the virtualized tasks collection that can pre-determine quantity and constraint condition characteristic, general purpose core G-Core subregion is carried out static partition, VMM has specified the sorting technique of client computer Domain by the configuration file that reads Universal Client Domain, the unloading phase of Universal Client Domain, this client computer Domain is assigned to the G-Core subregion of virtualized tasks subclass correspondence under it, in operational process, do not need to adjust its mode classification and partition method, this running environment is suitable for static the deployment, can reach more excellent deployment effect.And at the virtualized tasks collection that can't pre-determine quantity and constraint condition characteristic, when moving by VMM, the embodiment of the invention gathers the parameter of Universal Client Domain operation, judge whether the G-Core subregion can satisfy the computing demand of virtualized tasks subclass on it, if satisfy and then directly this client computer Domain be deployed to corresponding G-Core subregion, do not need to adjust, if do not satisfy then according to above-mentioned G-Core subregion again, virtual machine (vm) migration is finished the dynamic adjustment of G-Core subregion and Task Distribution, thus satisfy can't predetermined task-set the subregion virtual method.Comprise information such as the needed computer resource of this client computer Domain requires, constraint condition characteristic in the configuration file of described Universal Client Domain.Described virtualized tasks collection is the set of virtualized tasks subclass.
The embodiment of the invention is by supporting the dynamic adjustment of G-Core subregion and virtualized tasks subclass, make CPU Intel Virtualization Technology in the face of polycaryon processor can provide real-time or non real-time based on subregion, IO is intensive or computation-intensive, safe and reliable running environment, and this running environment is practical, efficiently.
Fig. 2 has provided the method for the present invention exemplary block diagram of virtual monitor in force.
As shown in Figure 2, have a plurality of core cpus in multinuclear or the many nuclear plateform systems, the embodiment of the invention select core cpu E0 to 12 core cpus of core cpu E11 as demonstration, expect the different general virtual task of operation 4 classes on it, different resource demand according to every kind of application, we have carried out subregion with above-mentioned core cpu, be divided into a G-Core subregion successively, the 2nd G-Core subregion, the 3rd G-Core subregion and the 4th G-Core subregion, wherein be assigned processor core E0 in the G-Core subregion, E1 and E2, be assigned processor core E3 and E4 in the 2nd G-Core subregion, be assigned processor core E5 in the 3rd G-Core subregion, E6 and E7 are assigned processor core E8 and E9 in the 4th G-Core subregion; In addition processor core E10 and E11 are assigned as the H-Core subregion.
After finishing the CPU subregion, the embodiment of the invention is deployed in the virtualized tasks subclass that dissimilar Universal Client Domain forms in the different CPU subregions, as shown in Figure 2, in the embodiment of the invention, virtualized tasks subclass A, B, C, D are deployed in respectively on a G-Core subregion, the 2nd G-Core subregion, the 3rd G-Core subregion and the 4th G-Core subregion; Virtualized tasks in each virtualized tasks subclass can be the application-specific among the independent Domain, also can be shared among the Domain and deposit a plurality of application, or operate in the service on the virtual machine monitor.As shown in Figure 2, virtualized tasks subclass A comprises application virtual machine A1, virtual machine A2, virtualized tasks subclass B comprises application virtual machine B 1, virtual machine B2, virtual machine B3, virtualized tasks subclass C comprises application virtual machine C1, virtual machine C2, virtual machine C3, and the virtualized tasks subset D comprises application virtual machine D1, virtual machine D2, virtual machine D3, virtual machine D4.
In addition, franchise Domain and management role are deployed in the H-Core subregion, as shown in Figure 2, franchise DomainH1 and management role H2 are deployed in the H-Core subregion.
After the virtualized tasks deployment finishes, carry out task scheduling independent of each other and resources allocation in the VMM control CPU subregion, and work in coordination on the whole, thereby the performance bottleneck of having avoided centralized CPU Intel Virtualization Technology to bring, both satisfy the individual demand of dissimilar application, also avoided the interfering with each other of variety classes application.

Claims (4)

1. CPU virtual method based on processor partitioning technology is characterized in that concrete steps comprise:
Step 1: virtual machine monitor VMM provides H-Core subregion and a plurality of G-Core subregion to the processor resource subregion of multinuclear or many nuclear platforms; During system start-up, virtual machine monitor VMM is with the processor core in multinuclear or the many nuclear platforms, be divided into management core H-Core and general purpose core G-Core two class CPU subregions, general purpose core G-Core class subregion wherein, dissimilar according to the constraint condition of the Universal Client Domain of system expection needs operation, be divided into the general purpose core G-Core subregion of respective amount, VMM is that each CPU subregion makes up data structure CPU subregion BITMAP, the concurrent physical processor ID that this subregion of record has in CPU subregion BITMAP, BITMAP creates different CPU subregions by traversal CPU subregion, the different CPU subregion is assigned the core cpu that quantity does not wait, by the VMM unified management, wherein Domain represents the territory, and BITMAP represents bitmap;
Step 2: from the management correlation function of VMM and Domain and code, isolate management role H-Task and franchise Domain H-Domain among the VMM, they are deployed to management core H-Core subregion in the step 1;
Step 3: the Universal Client Domain that moves on multinuclear or the many nuclear platforms is classified, and be deployed to corresponding general purpose core G-Core subregion; When Universal Client Domain starts,, be divided in the different virtualized tasks subclass according to the constraint condition of himself, then, in the general purpose core G-Core subregion of the correspondence that VMM is deployed to each virtualized tasks subclass in the step 1 to be divided;
Step 4: make up the scheduling sublayer module in the general purpose core G-Core subregion, and design the scheduling strategy that meets; In system's operational process, VMM makes up the scheduling sublayer module in each general purpose core G-Core subregion, the scheduling sublayer module of each general purpose core G-Core subregion, have separately independently scheduling queue and scheduling strategy, be responsible for the operation of Universal Client Domain in the scheduling subregion, and be responsible for safeguarding the resources allocation of the Universal Client Domain in this subregion, and configuration meets the various strategies of the Universal Client Domain constraint condition of moving it on;
Step 5: in the operational process of system, VMM carries out load balancing to general purpose core G-Core subregion and virtualized tasks subclass; Task scheduling, resources allocation work in each general purpose core G-Core subregion, finish by the scheduling sublayer module in each subregion, when computational resource in the general purpose core G-Core subregion breaks down or virtualized tasks subclass when changing, VMM adjusts the virtualized tasks subclass in general purpose core G-Core by stages, perhaps repartition general purpose core G-Core subregion, the virtualized tasks subclass is deployed in the new general purpose core G-Core subregion.
2. a kind of CPU virtual method based on processor partitioning technology according to claim 1 is characterized in that described constraint condition comprises: response time requirement, device access characteristic, reliability properties, security feature.
3. a kind of CPU virtual method according to claim 1 based on processor partitioning technology, it is characterized in that, the method that the subclass of virtualized tasks described in the step 3 is disposed, specifically when Universal Client Domain starts, set up mapping relations between the general purpose core G-Core subregion of VMM with the Universal Client Domain in this virtualized tasks subclass and its distribution, the general purpose core G-Core subregion BITMAP that specifies this client computer Domain to move by CPU mapping parameters in the Domain structure, thus a class virtualized tasks subclass and corresponding general purpose core G-Core subregion binding made.
4. a kind of CPU virtual method according to claim 1 based on processor partitioning technology, it is characterized in that, the operation of Universal Client Domain in the module schedules of scheduling sublayer described in the step 4 subregion, specifically the VCPU at the Universal Client Domain correspondence of moving in the general purpose core G-Core subregion carries out, VCPU is come it is dispatched by the scheduling code on each concurrent physical processor by in the scheduling queue of the scheduling sublayer module assignment of place general purpose core G-Core subregion each physical cpu correspondence in the general purpose core G-Core subregion of place then; Wherein, VCPU represents virtual processor.
CN201010108081A 2010-02-05 2010-02-05 CPU virtualization method based on processor partitioning technology Pending CN101788920A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201010108081A CN101788920A (en) 2010-02-05 2010-02-05 CPU virtualization method based on processor partitioning technology

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201010108081A CN101788920A (en) 2010-02-05 2010-02-05 CPU virtualization method based on processor partitioning technology

Publications (1)

Publication Number Publication Date
CN101788920A true CN101788920A (en) 2010-07-28

Family

ID=42532143

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201010108081A Pending CN101788920A (en) 2010-02-05 2010-02-05 CPU virtualization method based on processor partitioning technology

Country Status (1)

Country Link
CN (1) CN101788920A (en)

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102063331A (en) * 2011-01-07 2011-05-18 同济大学 Self-adaptive load balancing method for parallelization of spatial computation
WO2012103728A1 (en) * 2011-06-30 2012-08-09 华为技术有限公司 Method for dispatching central processing unit of hotspot domain virtual machine and virtual machine system
CN102662740A (en) * 2012-03-29 2012-09-12 迈普通信技术股份有限公司 Asymmetric multi-core system and realization method thereof
CN102831011A (en) * 2012-08-10 2012-12-19 上海交通大学 Task scheduling method and device based on multi-core system
CN102955685A (en) * 2011-08-17 2013-03-06 上海贝尔股份有限公司 Multicore DSP (digital signal processor), system with multicore DSP and scheduler
CN103019850A (en) * 2011-09-28 2013-04-03 中兴通讯股份有限公司 Method and device for binding tasks
WO2013097124A1 (en) * 2011-12-28 2013-07-04 华为技术有限公司 Method and device for guaranteeing quality of service of cloud application
CN103744735A (en) * 2014-01-09 2014-04-23 福建星网锐捷网络有限公司 Method and device for scheduling multi-core resource
CN104035808A (en) * 2013-03-04 2014-09-10 中兴通讯股份有限公司 Virtual machine set deployment method and device
CN104050042A (en) * 2014-05-30 2014-09-17 北京先进数通信息技术股份公司 Resource allocation method and resource allocation device for ETL (Extraction-Transformation-Loading) jobs
WO2014139379A1 (en) * 2013-03-13 2014-09-18 华为技术有限公司 Method and device for kernel running in heterogeneous operating system
CN104378420A (en) * 2014-11-12 2015-02-25 东软集团股份有限公司 Data transmission method and device based on environment perception
CN104615549A (en) * 2015-01-19 2015-05-13 杭州华三通信技术有限公司 Domain management method and device in virtual system
CN105653347A (en) * 2014-11-28 2016-06-08 杭州华为数字技术有限公司 Server, resource management method and virtual machine manager
CN105808328A (en) * 2014-12-31 2016-07-27 杭州华为数字技术有限公司 Task scheduling method, device and system
CN105959090A (en) * 2016-06-16 2016-09-21 国网信息通信产业集团有限公司 Service processing method and device of power wireless private network
CN106095589A (en) * 2016-06-30 2016-11-09 浪潮软件集团有限公司 Partition allocation method, device and system
CN107273188A (en) * 2017-07-19 2017-10-20 郑州云海信息技术有限公司 A kind of virtual machine central processing unit CPU binding method and device
CN108021441A (en) * 2016-11-03 2018-05-11 中国移动通信集团广东有限公司 A kind of resources of virtual machine collocation method and device based on cloud computing
CN109542831A (en) * 2018-10-28 2019-03-29 西南电子技术研究所(中国电子科技集团公司第十研究所) Airborne platform multi-core virtual multidomain treat-ment system
CN109840149A (en) * 2019-02-14 2019-06-04 百度在线网络技术(北京)有限公司 Method for scheduling task, device, equipment and storage medium
CN109933429A (en) * 2019-03-05 2019-06-25 北京达佳互联信息技术有限公司 Data processing method, device, electronic equipment and storage medium
CN110489230A (en) * 2019-08-05 2019-11-22 上海航天计算机技术研究所 The resource regulating method and equipment of safety-critical and time-critical
CN110489229A (en) * 2019-07-17 2019-11-22 长沙学院 A kind of multiple target method for scheduling task and system
CN111858073A (en) * 2020-04-30 2020-10-30 中科院计算所西部高等技术研究院 Processor with OODA multi-partition replaceable mechanism
CN113469837A (en) * 2021-06-28 2021-10-01 国网上海市电力公司 Twin terminal-based electric power automation system management and control platform
CN113608861A (en) * 2021-06-25 2021-11-05 西安空间无线电技术研究所 Software load computing resource virtualization distribution method and device
CN113778612A (en) * 2021-07-14 2021-12-10 中移物联网有限公司 Embedded virtualization system implementation method based on microkernel mechanism
CN117687799A (en) * 2024-02-02 2024-03-12 中国科学院空天信息创新研究院 Distributed stream type acceleration method and computing terminal for remote sensing interpretation application

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101149728A (en) * 2007-10-29 2008-03-26 中国科学院计算技术研究所 Multiple core processing system and its management method
CN101169731A (en) * 2007-12-05 2008-04-30 华为技术有限公司 Multiple-path multiple-core server and its CPU virtualization processing method
CN101460923A (en) * 2006-06-01 2009-06-17 微软公司 Virtual machine for operating n-core application on m-core processor

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101460923A (en) * 2006-06-01 2009-06-17 微软公司 Virtual machine for operating n-core application on m-core processor
CN101149728A (en) * 2007-10-29 2008-03-26 中国科学院计算技术研究所 Multiple core processing system and its management method
CN101169731A (en) * 2007-12-05 2008-04-30 华为技术有限公司 Multiple-path multiple-core server and its CPU virtualization processing method

Cited By (45)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102063331A (en) * 2011-01-07 2011-05-18 同济大学 Self-adaptive load balancing method for parallelization of spatial computation
CN102063331B (en) * 2011-01-07 2013-04-17 同济大学 Self-adaptive load balancing method for parallelization of spatial computation
WO2012103728A1 (en) * 2011-06-30 2012-08-09 华为技术有限公司 Method for dispatching central processing unit of hotspot domain virtual machine and virtual machine system
US9519499B2 (en) 2011-06-30 2016-12-13 Huawei Technologies Co., Ltd. Method for dispatching central processing unit of hotspot domain virtual machine and virtual machine system
CN102955685B (en) * 2011-08-17 2016-10-05 上海贝尔股份有限公司 Multi-core DSP and system thereof and scheduler
CN102955685A (en) * 2011-08-17 2013-03-06 上海贝尔股份有限公司 Multicore DSP (digital signal processor), system with multicore DSP and scheduler
CN103019850A (en) * 2011-09-28 2013-04-03 中兴通讯股份有限公司 Method and device for binding tasks
CN103019850B (en) * 2011-09-28 2019-02-12 中兴通讯股份有限公司 A kind of method and apparatus of task binding
CN103562852A (en) * 2011-12-28 2014-02-05 华为技术有限公司 Method and device for guaranteeing quality of service of cloud application
WO2013097124A1 (en) * 2011-12-28 2013-07-04 华为技术有限公司 Method and device for guaranteeing quality of service of cloud application
CN102662740A (en) * 2012-03-29 2012-09-12 迈普通信技术股份有限公司 Asymmetric multi-core system and realization method thereof
CN102831011B (en) * 2012-08-10 2015-11-18 上海交通大学 A kind of method for scheduling task based on many core systems and device
CN102831011A (en) * 2012-08-10 2012-12-19 上海交通大学 Task scheduling method and device based on multi-core system
CN104035808A (en) * 2013-03-04 2014-09-10 中兴通讯股份有限公司 Virtual machine set deployment method and device
CN104035808B (en) * 2013-03-04 2018-03-27 中兴通讯股份有限公司 The dispositions method and device of virtual machine set
WO2014139379A1 (en) * 2013-03-13 2014-09-18 华为技术有限公司 Method and device for kernel running in heterogeneous operating system
CN104050137B (en) * 2013-03-13 2018-02-09 华为技术有限公司 The method and device that kernel is run in a kind of OS
CN103744735A (en) * 2014-01-09 2014-04-23 福建星网锐捷网络有限公司 Method and device for scheduling multi-core resource
CN103744735B (en) * 2014-01-09 2017-07-25 福建星网锐捷网络有限公司 The dispatching method and device of a kind of multi-core resource
CN104050042B (en) * 2014-05-30 2017-06-13 北京先进数通信息技术股份公司 The resource allocation methods and device of ETL operations
CN104050042A (en) * 2014-05-30 2014-09-17 北京先进数通信息技术股份公司 Resource allocation method and resource allocation device for ETL (Extraction-Transformation-Loading) jobs
CN104378420A (en) * 2014-11-12 2015-02-25 东软集团股份有限公司 Data transmission method and device based on environment perception
CN104378420B (en) * 2014-11-12 2017-12-29 东软集团股份有限公司 Data transmission method and device based on environment sensing
CN105653347A (en) * 2014-11-28 2016-06-08 杭州华为数字技术有限公司 Server, resource management method and virtual machine manager
CN105808328B (en) * 2014-12-31 2019-05-17 杭州华为数字技术有限公司 The methods, devices and systems of task schedule
CN105808328A (en) * 2014-12-31 2016-07-27 杭州华为数字技术有限公司 Task scheduling method, device and system
CN104615549A (en) * 2015-01-19 2015-05-13 杭州华三通信技术有限公司 Domain management method and device in virtual system
CN105959090A (en) * 2016-06-16 2016-09-21 国网信息通信产业集团有限公司 Service processing method and device of power wireless private network
CN106095589B (en) * 2016-06-30 2019-04-09 浪潮卓数大数据产业发展有限公司 A kind of method, apparatus and system for distributing subregion
CN106095589A (en) * 2016-06-30 2016-11-09 浪潮软件集团有限公司 Partition allocation method, device and system
CN108021441A (en) * 2016-11-03 2018-05-11 中国移动通信集团广东有限公司 A kind of resources of virtual machine collocation method and device based on cloud computing
CN107273188B (en) * 2017-07-19 2020-08-18 苏州浪潮智能科技有限公司 Virtual machine Central Processing Unit (CPU) binding method and device
CN107273188A (en) * 2017-07-19 2017-10-20 郑州云海信息技术有限公司 A kind of virtual machine central processing unit CPU binding method and device
CN109542831A (en) * 2018-10-28 2019-03-29 西南电子技术研究所(中国电子科技集团公司第十研究所) Airborne platform multi-core virtual multidomain treat-ment system
CN109840149A (en) * 2019-02-14 2019-06-04 百度在线网络技术(北京)有限公司 Method for scheduling task, device, equipment and storage medium
CN109933429A (en) * 2019-03-05 2019-06-25 北京达佳互联信息技术有限公司 Data processing method, device, electronic equipment and storage medium
CN110489229A (en) * 2019-07-17 2019-11-22 长沙学院 A kind of multiple target method for scheduling task and system
CN110489229B (en) * 2019-07-17 2020-06-30 长沙学院 Multi-target task scheduling method and system
CN110489230A (en) * 2019-08-05 2019-11-22 上海航天计算机技术研究所 The resource regulating method and equipment of safety-critical and time-critical
CN111858073A (en) * 2020-04-30 2020-10-30 中科院计算所西部高等技术研究院 Processor with OODA multi-partition replaceable mechanism
CN113608861A (en) * 2021-06-25 2021-11-05 西安空间无线电技术研究所 Software load computing resource virtualization distribution method and device
CN113608861B (en) * 2021-06-25 2023-11-10 西安空间无线电技术研究所 Virtualized distribution method and device for software load computing resources
CN113469837A (en) * 2021-06-28 2021-10-01 国网上海市电力公司 Twin terminal-based electric power automation system management and control platform
CN113778612A (en) * 2021-07-14 2021-12-10 中移物联网有限公司 Embedded virtualization system implementation method based on microkernel mechanism
CN117687799A (en) * 2024-02-02 2024-03-12 中国科学院空天信息创新研究院 Distributed stream type acceleration method and computing terminal for remote sensing interpretation application

Similar Documents

Publication Publication Date Title
CN101788920A (en) CPU virtualization method based on processor partitioning technology
US11797327B2 (en) Dynamic virtual machine sizing
CN100511151C (en) Multiple-path multiple-core server and CPU virtualization processing method thereof
Vaishnav et al. Resource elastic virtualization for FPGAs using OpenCL
Zhuravlev et al. Survey of scheduling techniques for addressing shared resources in multicore processors
Sridharan et al. Holistic run-time parallelism management for time and energy efficiency
Sridharan et al. Adaptive, efficient, parallel execution of parallel programs
Kalé et al. A malleable-job system for timeshared parallel machines
US20150067356A1 (en) Power manager for multi-threaded data processor
CN102270159B (en) Access controlling and load balancing method for virtualized environment
CN103744716B (en) A kind of dynamically interruption Well-Balanced Mapping method based on current VCPU dispatch state
Polo et al. Performance management of accelerated mapreduce workloads in heterogeneous clusters
Navarro et al. Heterogeneous parallel_for template for CPU–GPU chips
Cheng et al. vScale: Automatic and efficient processor scaling for SMP virtual machines
Kang et al. ConVGPU: GPU management middleware in container based virtualized environment
Vaishnav et al. Heterogeneous resource-elastic scheduling for CPU+ FPGA architectures
CN107003713B (en) Event driven method and system for logical partitioning for power management
Alvarruiz et al. An energy manager for high performance computer clusters
Peter et al. Design principles for end-to-end multicore schedulers
Emeneker et al. Dynamic virtual clustering
Funk et al. A global optimal scheduling algorithm for multiprocessor low-power platforms
Chen et al. Smcompactor: a workload-aware fine-grained resource management framework for gpgpus
Suzuki et al. Gloop: An event-driven runtime for consolidating gpgpu applications
Shih et al. Fairness scheduler for virtual machines on heterogonous multi-core platforms
Liu et al. CPS: A Cooperative Para-virtualized Scheduling Framework for Manycore Machines

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C02 Deemed withdrawal of patent application after publication (patent law 2001)
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20100728