CN101727351A - Multicore platform-orientated asymmetrical dispatcher for monitor of virtual machine and dispatching method thereof - Google Patents
Multicore platform-orientated asymmetrical dispatcher for monitor of virtual machine and dispatching method thereof Download PDFInfo
- Publication number
- CN101727351A CN101727351A CN200910241837A CN200910241837A CN101727351A CN 101727351 A CN101727351 A CN 101727351A CN 200910241837 A CN200910241837 A CN 200910241837A CN 200910241837 A CN200910241837 A CN 200910241837A CN 101727351 A CN101727351 A CN 101727351A
- Authority
- CN
- China
- Prior art keywords
- service
- nuclear
- module
- request
- processor
- 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.)
- Granted
Links
Images
Landscapes
- Hardware Redundancy (AREA)
Abstract
The invention discloses an asymmetrical dispatcher for a multicore platform-orientated asymmetrical dispatcher for a monitor of a virtual machine and a dispatching method thereof, which provides a specialized service core meeting the special high performance demand is provided for a client operation system based on the conventional virtual machine monitor VMM. The dispatcher comprises a service core request resolution module in a management domain, a Domain specific service core configuration module in VMM, a system top dispatcher module, a service dispatcher module, a shared memory mechanism module and a request inquiry module, and a hybrid virtual module in an HVM Domain. The asymmetrical dispatcher in the invention not only can provide different service cores according to the demands of upper layer Domain, but also can provide higher flexibility and expandability than the traditional dispatcher due to the consideration of the topological structure of a physical platform during the distribution of the service cores. In addition, the dispatcher provides an acceleration mechanism of a specific application type for the client operation system and improves the service efficiency of the entire platform.
Description
Technical field
The present invention relates to a kind of multinuclear or many nuclear dispatching platforms technology, is towards asymmetric scheduler of the monitor of virtual machine of multi-core platform and dispatching method thereof specifically, belongs to computer operating system and technical field of virtualization.
Background technology
The system virtualization technology is meant utilizes virtual machine monitor (Virtual Machine Monitor, be called for short VMM) this system software layer, resources such as the processor peripheral hardware of management system platform, for client operating system provides a virtual hardware execution environment, these virtual hardware execution environments become a virtual machine (Virtual Machine is called for short VM).Each virtual machine (VM) all comprises virtual hardware resources such as processor, storer and equipment, and can support a plurality of VM to move simultaneously on same physical host.This technique functions comes from the sixties in 20th century, IBM Corporation has used Intel Virtualization Technology in 360 family computer systems, its objective is and allow the user can share computational resource, and subsequently three during the decade, it is perfect that Intel Virtualization Technology on the large scale computer constantly is tending towards, and be the basic technology of efficiently utilizing the powerful computational resource of large scale computer always.Arrived twentieth century nineties, along with improving constantly of x86 processor performance, some are more and more stronger based on the computing power of low and middle-grade servers of x86 processor even PC, and it is low that resource utilization ratio has all appearred in lot of data center and PC, the problem of computational resource serious waste.Under this background, the system virtualization technology that was mainly used in large scale computer in the past is subjected to the extensive concern of industry member and academia, Intel Virtualization Technology based on the x86 processor has obtained significant progress, and in all over the world data center, has obtained on the PC user using widely.Intel Virtualization Technology not only can improve usage factor of system resource, the installation amount and the corresponding power supply cost of data center have been reduced, and for some conventional requirement provide new simple realization means, such as the load balance based on the system migration technology, redundancy with failover capabilities.In addition, Intel Virtualization Technology also provides the method for using diversity operating system more easily for the PC user, for systemic software developments such as many operating systems and tester provide unprecedented facility.
With the closely related another one background technology of the present invention be the sustainable development of multi-core technology in recent years.Traditional multipath server is a plurality of concurrent physical processor slots, just the symmetric multiprocessor system of directly supporting on a mainboard.And along with SMT (Simultaneous Multi-Threading, i.e. multithreading simultaneously), CMP (Chip multiprocessors, be on-chip multi-processor nuclear) development, processor core integrated on the single chip can get more and more, and succeeds in developing the single chip of integrated 80 nuclears in 2007 in the Intel laboratory.In following a period of time, four nuclears, eight core processors etc. all can be widely used gradually, and hardware platform is entering the multinuclear epoch.Correspondingly be, the development of software, especially the development of the development of system software and hardware platform is compared, some hysteresis seems, the processing power that under a lot of situations all is hardware has improved, and perhaps provide new function, but the innovation speed of software stack is comparatively slow, make many new ardware features to be widely used, caused a large amount of wastes of computational resource, power supply etc.Along with the development of multi-core platform, operating system and application program press for quick innovation, adapt to new powerful computing platform.In addition, heterogeneous processor nuclear also can be used on low and middle-end server and PC in a large number, most typical example is present very popular graphic process unit GPU (GraphicProcessing Unit), network processing unit NP (Network Processor, be called for short NP), even the GPU Larrabee of the Intel that will go on the market within the next few days.
The Intel Virtualization Technology of a new round is to develop rapidly because of the computational resource of efficient using system surplus at first, and now along with processor core integrated on the single hardware platform continues to increase, concern that the system virtualization technology will obtain continuing and development, Intel Virtualization Technology will more importantly act in the powerful calculating ability performance that abundant usage platform provides.Manufacturer server has been arranged at present with among the virtual BIOS of being placed on, for system software provides more transparent virtual platform, be similar to the partitioning technique on the large scale computer, the most typical use of Intel Virtualization Technology is carried out subregion to a large amount of processor cores exactly, and traditional operating system can be installed in each subregion, so both can fully use newer hardware platform, made full use of long operating system and application program of update cycle again, for the sustainable development of computer system platform provides feasible solution.In a word, the development of system virtualization technology and the development of multi-core technology are the most significant different technology in present hardware and the system software development, this different technology is interacting simultaneously, promote mutually, will consider of the use of system virtualization technology in the design of multinuclear and the realization to platform, and the also inevitable constantly innovation along with the development of multinuclear of system virtualization technology, more isomorphism processor core and various heterogeneous processor integrated on the most effective usage platform are examined.
How the system virtualization technical research is used resources allocations such as processor, internal memory, peripheral hardware to client computer, it is a kind of of Intel Virtualization Technology, abstract to as if whole computing machine.The system virtualization technology is abstract to processor, and be important component part in the Intel Virtualization Technology to the scheduling of virtual processor, present monitor of virtual machine VMM mainly contains three kinds of modes from the virtualized technical division of processor, a kind of is binary translation, under this main conditions, the privileged instruction in the client operating system is carried out by explaining; Second kind is that class is virtual, promptly revise client operating system the privileged instruction that needs in the client operating system to carry out is modified as the calling of monitor of virtual machine, thereby certain this technology is general and client operating system is carried out privileged instruction and caused being trapped in unusually then and take over execution by VMM among the VMM and combine; The third is exactly the auxiliary complete Intel Virtualization Technology of hardware, use has the processor that hardware virtualization is supported, SVM such as Intel VT-x or AMD, client operating system all can be trapped among the VMM when carrying out all sensitivity instructions and privileged instruction, VMM can turn back in the client operating system on behalf of after disposing.Above several method cuts both ways, wherein the third hardware is assisted the present performance of virtual method and bad, reason VM exit (virtual machine withdraws from) quantity that to be client computer cause because carry out privileged instruction is too much, and it is long to handle these times of withdrawing among the VMM, causes the performance of HVM (hardware virtual machine) Domain (territory) not high.Substantially all be to have continued traditionally Linux to the support of multinuclear during present in addition VMM realizes to the support pattern of SMP platform, scheduling to vcpu (virtual processor) also is the method that adopts the symmetry scheduling, be that scheduler is not distinguished the processor core that is in the different encapsulation, also do not support to be specifically designed to the VMM task of certain specific demand and the binding of processor core.
Summary of the invention
The present invention is directed to that symmetrical dispatching method can not fully use the multinuclear processing power among the present multi-core platform VMM, and the problem of the service of customization flexibly can not be provided for different client computer, propose asymmetric scheduler of a kind of monitor of virtual machine and dispatching method thereof towards multi-core platform.
The asymmetric scheduler of monitor of virtual machine towards multi-core platform of the present invention's design, it structurally the existing general scheduler module, also comprises the module of following new interpolation in comprising VMM: the service nuclear request analysis module in the management domain; Module, the top Scheduler module of system, service dispatch device module, the machine-processed module of shared drive and request enquiry module are put in Domain specific service caryogamy among the VMM, wherein comprise topological identification of platform processor nuclear and analysis submodule and the distribution of service nuclear in the top Scheduler module of system and reclaim submodule; Hybrid virtualization module among the HVM Domain.VMM operates on multinuclear or the many karyonides system platform, is managing the hardware platform of bottom downwards, is moving management domain and HVMDomain to upper support.
The request analysis module is examined in service, is used for serving the configuration of examining when HVM Domain creates, and the request to create of serving nuclear is sent in the type and the quantity parsing of the service nuclear that indicates in the configuration file when being responsible for HVM Domain client computer establishment just to VMM;
The caryogamy of Domain specific service is put module and then is responsible for the service nuclear request to create of the HVMDomain that preliminary analysis service nuclear request analysis module sent out and sends to the top Scheduler module of system then, and the information that its service nuclear phase of management closes in the life cycle of whole HVM Domain;
The top Scheduler module of system is mainly used to the processor core of energy usefulness in the Management Calculation machine platform, and in global scope to general Scheduler module and service the adjusting of Scheduler module with processor nuclear resource bitmap, be responsible for finishing caryogamy and put request in the module from the Domain specific service, according to platform processor nuclear topology identification with analyze topology information in the submodule, the service nuclear of use distributes and reclaims submodule and create service nuclear or reclaim service nuclear; Topology identification of platform processor nuclear and analysis submodule are responsible for the processor core resource that has in the system is discerned, processor core after being identified, add top scheduler use and manage can be with in the processor nuclear resource bitmap, top scheduler begins to move on these processor cores after the initialization; Distribution of service nuclear and recovery submodule are responsible for serving the establishment or the recovery of nuclear, can be from system with distinguishing service nuclear the processor nuclear resource bitmap, and insert service dispatch device module can be with in the processor nuclear resource bitmap, perhaps can examine from service dispatch device module with deletion service the processor nuclear resource bitmap;
Service dispatch device module is responsible for serving the management and the scheduling of nuclear;
Shared drive mechanism module is responsible for responding the mapping request of HVM Domain, and after shared drive is set up, the trigger request enquiry module;
The hybrid virtualization module is at first to VMM request mapping shared drive, and the mapping request of shared drive mechanism module responds HVM Domain is set up the shared drive zone in VMM; After shared drive was set up, the hybrid virtualization module was sent the services request of HVMDomain to service nuclear, and these requests are written in the shared drive zone;
The request enquiry module is responsible for inquiring about the services request to service nuclear in the affiliated shared drive zone of HVM Domain, and these requests are sent to service dispatch device module.
The top Scheduler module of described system just can be triggered in the occasions such as load adjustment that system initialization, HVM Domain create, request to create is examined in service, service is examined in recovery, HVM Domain destruction, general purpose core and the adjustment of service nuclear ratio, the global scope.
The asymmetric dispatching method of monitor of virtual machine towards multi-core platform of the present invention's design, concrete steps comprise:
Step 1, client computer are created the request to create of startup and service nuclear and are sent;
When HVM Domain client computer is created, in management domain, have the establishment configuration file of a corresponding client machine, can indicate the quantity of the vcpu of client computer use in this document, and with the binding information of concurrent physical processor nuclear, also can indicate the type and the quantity of required service nuclear simultaneously in this configuration file, concrete type is by special appointment and realization among the different embodiment; Service nuclear request analysis module in the management domain is resolved the type and the quantity of the nuclear of the service in this configuration file, and will create the overshoot Hypercall of the paired VMM of shift in demand, and this overshoot put module with the Domain specific service caryogamy that is transferred among the VMM, and the caryogamy of Domain specific service is put module after adapter overshoot is with Hypercall, adopt all information of a new HVM Domain who creates of structure management, the configuration information that comprises service nuclear, and tentatively resolve overshoot Hypercall, trigger the top Scheduler module operation of system among the VMM then; Wherein vcpu represents virtual processor, and Hypercall represents hypercalls;
Step 2, identification of processor core topology and the distribution of analyzing and serving nuclear;
Identification of processor core topology and analysis submodule in the top Scheduler module of system can be discerned the processor core resource that has in the system, what processor core top Scheduler module of adding system after being identified was used and managed can use in the processor nuclear resource bitmap, the top Scheduler module of system begins to move on these processor cores after the initialization, is allocated to the general Scheduler module among the VMM or is allocated to the scheduling that the service stone grafting is subjected to the service dispatch device module among the VMM up to these processor cores;
During system initialization, the top Scheduler module of system is initialised, after the top Scheduler module of system is received the request of creating service nuclear, at first analyze the topology information that submodule obtains platform processor nuclear according to platform processor nuclear topological sum, the concurrent physical processor nuclear phase neighbour's of the vcpu binding of selection of employing nearby principle and HVM Domain processor core, if setting, vcpu do not need binding concurrent physical processor nuclear, then according to the whole strategy of system, be performance preferentially or the predetermined policy of energy conservation priority select, when performance is preferential, the preferential processor core of selecting the free time then preferentially selects the processor core in hyperthread or the same encapsulation to examine as service as service nuclear during energy conservation priority;
After service nuclear is selected, be responsible for serving the final constructive process of nuclear by distribution of the nuclear of the service in the top Scheduler module of system and recovery submodule, service nuclear distributes and reclaims submodule can be with distinguishing service nuclear the processor nuclear resource bitmap from system, and insert service dispatch device module can be with in the processor nuclear resource bitmap, examine by interrupting the IPI call service between processor then, and make it begin to carry out from the function of appointment; Service nuclear is created and when beginning to carry out, really do not begin to provide service, after the shared drive zone was created in the HVMDomain application, service nuclear just can really be activated, initiatively arrive shared drive zone polling tasks, perhaps serviced Scheduler module just is activated when dispatching service is provided; The recovery of service nuclear is to trigger when the affiliated HVMDomain of service nuclear is destroyed, similar with the process of creating, the top Scheduler module of system is called the distribution of service nuclear and is reclaimed submodule and reclaim, and the service that will reclaim nuclear is redistributed, collocation strategy according to system, with the service of free time nuclear, join general Scheduler module can usefulness processor nuclear resource bitmap in, or join in system's idling-resource;
The VMM initial start stage, all processor cores in the computer platform are all by top processor module management of system and scheduling, along with the establishment of HVM Domain and corresponding vcpu, general Scheduler module has had scheduler object, is initialised and moves on the general purpose core of being managed;
After service nuclear is created, service dispatch device module begins service nuclear is dispatched, it understands the services request in the shared drive zone of each HVM Domain of poll and VMM, the call request enquiry module obtains new service role and sends to service dispatch device module, service dispatch device module is examined according to the service under the task, task is inserted into the task queue of this service nuclear, and triggers the operation of service nuclear; If service nuclear is designated as direct poll pattern when creating, then do not need service dispatch device module that task is distributed to service nuclear, the direct poll shared drive of service nuclear zone obtains services request;
Step 4, client computer are sent the request of services request and corresponding with service nuclear customer in response machine;
After HVM Domain is created and starts, load the hybrid virtualization module among the HVM Domain, the hybrid virtualization module is sent the request of mapping shared drive to VMM, shared drive mechanism module responds request among the VMM, mapping shared drive zone in the space of VMM, and the start address in shared drive zone and size information are write the caryogamy of Domain specific service put in the information structure of module management, the service code that shared drive shines upon on the corresponding service nuclear in successful back brings into operation;
The hybrid virtualization module sends services request by the mode of writing shared drive, services request also can be sent by the application-specific that operates in user's attitude, by having operating system user attitude now, kernel state module data and control method are transferred to hybrid virtualization module agency and are sent, after services request sends, request enquiry module among the VMM will be inquired about these requests according to certain frequency, and these requests are transferred to the service nuclear that service dispatch device module is distributed correspondence, service nuclear can be carried out the task queue of oneself according to the scheduling strategy in the service dispatch device module, also can adopt a kind of mode of serving the shared drive zone of the direct poll correspondence of nuclear, in case request is arranged, directly carry out, this situation is applicable to that real-time requires perhaps frequently to carry out and determining the key service of entire system performance than higher service; After service nuclear is finished services request, result is write back shared drive by shared drive mechanism module, other kernel modules among hybrid virtualization module or the HVM Domain just can be read the service execution result who returns.
The asymmetric dispatching method of described monitor of virtual machine towards multi-core platform, it is characterized in that, identification of processor core topology and analysis submodule can be discerned the processor core resource that has in the platform in the described step 2, the main basis of identification is, whether in an encapsulation, whether whether share L2 cache, be the hyperthread on the same nuclear.
The asymmetric dispatching method of described monitor of virtual machine towards multi-core platform, it is characterized in that, nearby principle is adopted in the distribution of the nuclear of service described in the step 2, be specially: if vcpu is a hyperthread, then service nuclear is preferentially selected from the most contiguous hyperthread, if vcpu has been tied to a processor core, then service nuclear preferentially distributes from other processor core that is in together in the encapsulation, further, if the processor core of some processor core and vcpu binding is shared L2 cache in the same encapsulation, other are not shared, the then preferential processor core of selecting to share L2 cache.
The asymmetric dispatching method of described monitor of virtual machine towards multi-core platform, it is characterized in that, the shared drive of setting up between HVMDomain and the VMM in the described step 4, unified address space is arranged in VMM, the mapping space that oneself is also arranged in HVM Domain internal system, be unified management to the request zone in these shared drives among the VMM, the synchronous communication mode of poll or the mode that adopts message to transmit are adopted in the management of request.
Asymmetric scheduler of the monitor of virtual machine towards multi-core platform and dispatching method thereof that the present invention proposes have following technological merit and effect:
(1) the client operating system special demand has been satisfied in the existence of service nuclear
The performance bottleneck of present most of monitor of virtual machine is all at IO, reason mainly is exactly the demand that the framework of VMM can not adapt to the concurrent visit of a plurality of client operating systems IO equipment, the acceleration that the existence of service nuclear just can provide specific IO to visit for client operating system.In addition, the specific demand of client operating system not only is embodied on some performance requirement, also be embodied on some functional demand, belong to the auxiliary virtual machine of hardware such as certain client operating system, but want to adopt the hybrid virtualization method, quicken some driving, and service nuclear can assist also client operating system to finish these tasks.Service nuclear is a kind of a kind of mechanism of using virtualization architecture in a word, and the virtualized environment that exists for of this mechanism provides new client operating system to the interactive mode between the VMM, by the model of this request and service mode, can satisfy the interaction demand of a variety of client operating systems and VMM.
(2) distribution of service nuclear adapts to the processor topology of platform
Traditional scheduler at smp system can not adapt to the complicated day by day platform based on the x86 processor, and the monitor of virtual machine software of current trend mainly is towards popular processor configuration, does not also fully take into account the topological structure of numerous processor cores of platform.Asymmetric scheduler among the present invention has taken into full account service nuclear when distribution services is examined and the topological relation of the concurrent physical processor nuclear that the client operating system virtual processor vcpu that served is moved, some advantages have been demonstrated for present computer platform, such as reducing the cache miss rate, improved the overall performance of system, for following many nuclear platforms and the platform that has a large amount of isomeries nuclears, this service nuclear allocation model can show more obvious advantage, be fit to carry out matrix operation such as graphic process unit GPU, can provide GPU for the client operating system that needs the dense matrix computing as service nuclear.
(3) asymmetrical two-level scheduler device model provides dirigibility and configurability for total system
The asymmetric scheduler of the two-stage of using among the present invention, not only the specific demand for client operating system provides more flexible services, also makes things convenient for simultaneously the abundant processor core resources in multinuclear or the following many nuclear platforms are utilized.With asymmetrical two-level scheduler device is the center, to the individual demand of last adaptation client operating system, to making full use of the computational resource of platform down, and can expand the scheduling of carrying out heterogeneous processor nuclear easily and use.The symmetrical scheduling model that the compatible existing most monitor of virtual machine of general scheduler adopts, and the service dispatch device provides available support easily for the various specific demands of client operating system.The two is worked in coordination with, and does the distribution of the overall situation and the dynamic adjustment in when operation, the dirigibility that has improved total system greatly by top scheduler.
The raising of configurability mainly is because providing of nuclear is provided can be disposed based on the single virtual machine by management domain, and the assigning process of service nuclear also is the topological structure optimization distribution according to platform.In addition the dynamic adjustment strategy of top scheduler also is to be specified by management domain, when disposing respectively such as or energy conservation priority strategy preferential according to performance, top scheduler is exactly different to the adjustment of the available processors nuclear bitmap of general scheduler and service scheduler.During energy conservation priority, no service nuclear can enter low energy consumption state, and when performance was preferential, the service nuclear of recovery can be joined in the available processors resource bitmap of general scheduler by top scheduler, is used for dispatching the vcpu of client operating system by general scheduler.The new trend that among the present invention the improvement of dirigibility and configurability also is being fit to the hardware platform in the constantly innovation.
Description of drawings
Fig. 1 is the modular structure synoptic diagram towards the asymmetric scheduler of monitor of virtual machine of multi-core platform;
Fig. 2 is the flow chart of steps towards the asymmetric dispatching method of monitor of virtual machine of multi-core platform.
Embodiment
In following concrete exemplifying embodiment, in conjunction with the accompanying drawings the present invention is further described in detail.
As shown in Figure 1, VMM 10 operates on multinuclear or the many karyonides system platform, is managing the hardware platform 12 of bottom downwards, is moving management domain 9 and HVM Domain11 to upper support.The asymmetric scheduler of monitor of virtual machine towards multi-core platform of the present invention's design, it structurally the existing general scheduler module 4, also comprises the module of following new interpolation in comprising VMM 10: the service nuclear request analysis module 1 in the management domain 9; Module 2, the top Scheduler module 3 of system, service dispatch device module 5, the machine-processed module 7 of shared drive and request enquiry module 8 are put in Domain specific service caryogamy among the VMM10, wherein comprise topological identification of platform processor nuclear and analysis submodule 301 and the distribution of service nuclear in the top Scheduler module 3 of system and reclaim submodule 302; Hybrid virtualization module 6 among the HVM Domain11.
HVM Domain11 is the Domain of carrying calculation task, and management domain 9 is controlled HVM Domain11 according to the mechanism that VMM10 provides.The type and the quantity of the service nuclear that indicates in the configuration file when service nuclear request analysis module 1 is responsible for client computer created in the management domain 9 are resolved, and send the request to create of service nuclear to VMM10.Domain specific service caryogamy among the VMM10 is put 2 of modules and is responsible for the service nuclear request to create that preliminary analysis request parsing modules 1 send and sends to the top Scheduler module 3 of system then, and the information that its service nuclear phase of management closes in the life cycle of whole HVM Domain11.The top Scheduler module 3 of system is mainly used to the available processor core in the management system, and in global scope can adjust general Scheduler module 4 and service Scheduler module 5 with processor nuclear resource bitmap, be responsible for specifically finishing caryogamy and put request in the module 2 from the Domain specific service, according to platform processor nuclear topology identification with analyze topology information in the submodule 301, the service nuclear of use distributes and reclaims submodule 302 and create service nuclear or reclaim service nuclear; Topology identification of platform processor nuclear and analysis submodule 301 are responsible for the processor core resource that has in the system is discerned, service nuclear distributes and reclaims the establishment or the recovery of submodule 302 responsible service nuclears, can be from system with distinguishing service nuclear the processor nuclear resource bitmap, and insert service dispatch device module 5 can be with in the processor nuclear resource bitmap, perhaps can examine from service dispatch device module 5 with deletion service the processor nuclear resource bitmap.Mainly be the system under the normal operating condition, general Scheduler module 4 is responsible for the scheduling of virtual processor vcpu on general purpose core of HVMDomain11, and service dispatch device module 5 is responsible for the management and the scheduling of service nuclear.
Shared drive mechanism module 7 is responsible for the mapping request of response HVM Domain, and after shared drive is set up, trigger request enquiry module 8.
Hybrid virtualization module 6 among the HVM Domain11 is responsible for to VMM10 request mapping shared drive, and the mechanism of the shared drive among VMM10 module 7 is responsible for the mapping request of response HVM Domain11; After shared drive was set up, HVM Domain11 sent by hybrid virtualization module 6 services request of service nuclear, simultaneously shared drive mechanism module 7 trigger request enquiry modules 8.
The general purpose core Scheduler module 4 and service nuclear Scheduler module 5 two-stages that comprise top Scheduler module 3 of system and subordinate thereof towards the asymmetric scheduler of the monitor of virtual machine of multi-core platform, the role that the top Scheduler module 3 of system is responsible for processor core divides, the establishment and the recovery of service nuclear, variation according to running load, adjust the quantity of the service nuclear of general purpose core that general Scheduler module 4 managed and 5 management of service dispatch device module, realize that computer processor nuclear is as general purpose core with as the role transforming of service nuclear.
The asymmetric scheduler of the monitor of virtual machine towards multi-core platform that the present invention proposes, mainly showing in the asymmetry scheduling of the difference of it and existing scheduler: at first, processor core to computer platform does not use traditional symmetrical dispatching method, but adopted some nuclear is examined as service, moved special VMM service; Secondly, the demand of upper strata HVM Domain11 had both been considered in the distribution of service nuclear, also the processor core topological structure of computer platform is considered to some extent, the top Scheduler module 3 of system is according to the information such as number of types of vcpu binding information in the configuration file of creating HVM Domain11 and service nuclear, adopt nearby principle distribution services nuclear, if setting, vcpu do not need binding concurrent physical processor nuclear, then according to the whole strategy of system, be performance preferentially or the predetermined policy of energy conservation priority select, when performance is preferential, the preferential processor core of selecting the free time then preferentially selects the processor core in hyperthread or the same encapsulation to examine as service as service nuclear during energy conservation priority.
The asymmetric dispatching method of the monitor of virtual machine towards multi-core platform that the present invention proposes mainly is divided into 4 steps in system implementation, promptly client computer is created and started and the request to create of service nuclear is sent, processor core topology identification and analyzing and request that the distribution of service nuclear, two-level scheduler device are sent services request and corresponding with service nuclear customer in response machine to general purpose core and the scheduling of service nuclear and client computer.Fig. 2 in conjunction with the accompanying drawings, concrete steps are as follows:
Step 1: client computer is created the request to create of startup and service nuclear and is sent;
When HVM Domain11 creates, in management domain 9, have the establishment configuration file of a corresponding client machine, can indicate the quantity of the vcpu of client computer use in this document, and with the binding information of concurrent physical processor nuclear, the type and the quantity that also can indicate required service nuclear simultaneously in this configuration file (are served such as file system service, debugging services, IO, the perhaps acceleration request of the concrete operations as page table traversal), concrete type can be by special appointment and realization among the different embodiment.Service nuclear request analysis module 1 in the management domain 9 is resolved the type and the quantity of the nuclear of the service in this configuration file, and will create the overshoot Hypercall of the paired VMM10 of shift in demand, and this overshoot put module 2 with being transferred to the caryogamy of Domain specific service, and the caryogamy of Domain specific service is put module 2 after adapter overshoot is with Hypercall, adopt all information of a new HVM Domain who creates of structure management, the configuration information that comprises service nuclear, and tentatively resolve overshoot Hypercall, the top Scheduler module 3 of triggering system is moved then.
Step 2, identification of processor core topology and the distribution of analyzing and serving nuclear;
Identification of processor core topology and analysis submodule 301 can be discerned the processor core resource that has in the system, main according to whether in an encapsulation, whether share L2 cache, whether be the hyperthread on the same nuclear, processor core top Scheduler module 3 of adding system after being identified use and manage can usefulness processor nuclear resource bitmap in; Top Scheduler module 3 beginnings of system move on these processor cores after the initialization, are allocated to general Scheduler module 4 or are allocated to the scheduling that the service stone grafting is subjected to service dispatch device module 5 up to these processor cores.
During system initialization, the top Scheduler module 3 of system among the VMM10 is initialised, the top Scheduler module 3 of system is when accepting request distribution services nuclear, from system can with the processor core resource find suitable processor nuclear, and to this processor core initialization make it to become the function that operation HVM Domain needs (such as file system service, IO device service, debugging services service, the IO service of simulation) service nuclear, service nuclear is created then.After the top Scheduler module 3 of system is received the request of creating service nuclear, at first analyze the topology information that submodule 301 obtains platform processor nuclear according to platform processor nuclear topological sum, the concurrent physical processor nuclear phase neighbour's of the vcpu binding of selection and HVM Domain11 processor core, be specially: the top Scheduler module 3 of system is according to the information such as number of types of vcpu binding information in the configuration file of creating HVM Domain11 and service nuclear, adopt nearby principle distribution services nuclear, if such as vcpu is a hyperthread, then service nuclear is preferentially selected from the most contiguous hyperthread, if vcpu has been tied to a processor core, then service nuclear preferentially distributes from other nuclear that is in together in the encapsulation, further, if the processor core of some processor core and vcpu binding is shared L2 cache in the same encapsulation, other are not shared, the then preferential processor core of selecting to share L2 cache.If setting, vcpu do not need binding concurrent physical processor nuclear, then according to the whole strategy of system, be performance preferentially or the predetermined policy of energy conservation priority select, when performance is preferential, the preferential processor core of selecting the free time then preferentially selects the processor core in hyperthread or the same encapsulation to examine as service as service nuclear during energy conservation priority.
After service nuclear is selected, distribute and reclaim the final constructive process of submodule 302 responsible service nuclears by service nuclear.Service nuclear distributes and reclaims submodule 302 mainly is can be with distinguishing service nuclear the processor nuclear resource bitmap from system, and insert service dispatch device module 5 can be with in the processor nuclear resource bitmap, examine by interrupting IPI (inter processorinterrupt) call service between processor then, and make it begin to carry out from the function of appointment.Service nuclear is created and when beginning to carry out, really do not begin to provide service, after the shared drive zone is created in HVM Domain11 application, service nuclear just can really be activated, initiatively arrive shared drive zone polling tasks, perhaps serviced Scheduler module 5 just is activated when dispatching service is provided.
The recovery of service nuclear is to trigger when the affiliated HVM Domain11 of service nuclear is destroyed, similar with the process of creating, the top Scheduler module 3 of system is called the distribution of service nuclear and is reclaimed submodule 302 and reclaim, and the service that will reclaim nuclear is redistributed, collocation strategy according to system, with the service nuclear of free time, can join also can joining in system's idling-resource of general Scheduler module 4 with in the processor nuclear resource bitmap.
Step 3: the two-level scheduler device is to general purpose core and the scheduling of service nuclear;
The VMM10 initial start stage, all processor cores in the platform are all by top processor module 3 management of system and scheduling, along with the establishment of HVM Domain11 and corresponding vcpu, general Scheduler module 4 has had scheduler object, is initialised and moves on the general purpose core of being managed.
After service nuclear was created, service dispatch device module 5 began service nuclear is dispatched, and the scheduling of traditional vcpu then is responsible for by general purpose core Scheduler module 4.General Scheduler module 4 has the function of load between the processor core under scheduling vcpu and the balance, and it can be with all can be as the code of the general Scheduler module 4 of the periodic operation of conventional scheduler on each processor core in the processor nuclear resource bitmap.The services request in the shared drive zone of service dispatch device module 5 meeting each HVM Domain11 of poll and VMM10, be that the new service role of call request enquiry module 8 acquisitions sends to service dispatch device module 5, service dispatch device module 5 is examined according to the service under the task, task is inserted into the task queue of this service nuclear, and triggers the operation of service nuclear.If service nuclear is designated as direct poll pattern when creating, then do not need service dispatch device module 5 that task is distributed to service nuclear, the direct poll shared drive of service nuclear zone obtains services request.
Step 4: client computer is sent the request of services request and corresponding with service nuclear customer in response machine;
Step 1 and step 2 have been finished the establishment to service nuclear, and to the initialization of service nuclear, and service nuclear is dispatched by the scheduler of two-stage.After HVM Domain11 was created and starts up, the service under it was at this moment examined also just ready, and service this moment nuclear is in waiting status, and wait service dispatch device module 5 is received request back dispatch service nuclear and carried out.It is after services request is authorized out in 6 pairs of services of hybrid virtualization module in HVM Domain11 that service nuclear really puts into operation.After HVM Domain11 is created and starts, load hybrid virtualization module 6, hybrid virtualization module 6 is sent the request of mapping shared drive to VMM10, shared drive mechanism module 7 among the VMM10 is shone upon shared drive in the space of VMM10, and information such as the start address of shared region and size are write the caryogamy of Domain specific service put module 2 information of managing structures, the service code that shared drive shines upon on the corresponding service nuclear in successful back brings into operation.
Hybrid virtualization module 6 among the HVM Domain11 sends services request by the mode of writing shared drive, services request also can be sent by the application-specific that operates in user's attitude, transfers to the 6 agency's transmissions of hybrid virtualization module by having operating system user attitude, kernel state module data and control method now.After services request sent, the request enquiry module 8 among the VMM10 will be inquired about these requests according to certain frequency, and the service nuclear that service dispatch device module 5 is distributed correspondence is transferred in these requests.Service nuclear can be carried out the task queue of oneself according to the scheduling strategy in the service dispatch device module 5, after service nuclear is finished services request, result is write back shared drive by shared drive mechanism module 7.Hybrid virtualization module 6 or other kernel modules among the HVM Domain11 just can be read the service execution result who returns.
Client computer is for the service that the service nuclear that uses VMM10 provides, at first need and VMM10 between set up shared drive.The hybrid virtualization module 6 that is arranged in HVM Domain11 sends the request that shared drive shines upon to VMM10, and VMM10 handles request, and has set up VMM10 and examined a region of memory of sharing with service.
The request of client computer and data all pass to VMM10 by shared drive mechanism, before shared drive is created, service nuclear is in ready state, really do not begin to provide service, in case and shared drive is created successfully, the service dispatch device will be inquired about the request of newly arriving by the services request enquiry module, can distribute suitable request on the service nuclear simultaneously.Hybrid virtualization module 6 among the HVMDomain11 not only is responsible for the mapping shared drive, and the responsible services request that sends reality, can expand this mechanism among the HVM Domain11, the user's attitude program that is HVM Domain11 can be mutual with the hybrid virtualization module 6 that is installed in kernel state, makes hybrid virtualization module 6 on behalf of the special service request of carrying out from user's attitude.The shared drive of setting up between HVM Domain11 and the VMM10, unified address space is arranged in VMM 10, oneself mapping space is also arranged operating system inside separately in client computer, be unified management to the request zone in these shared drives among the VMM10, among the present invention the synchronous communication mode that has adopted poll, the mode that also can adopt message to transmit.
After HVM Domain11 sends services request by shared drive, service dispatch device module 5 can be inquired about the request in each shared drive zone by request enquiry module 8, and distribute suitable service according to zone under the request and type and examine, so services request can be by corresponding service nuclear execution.In most cases, request is distributed to the scheduling of service nuclear by service dispatch device module 5, but also can adopt a kind of shared drive zone of serving the direct poll correspondence of nuclear in the present invention's practice, in case request is arranged, directly carry out, this situation is applicable to that real-time requires perhaps frequently to carry out and determining the key service of entire system performance than higher service.
Claims (6)
1. towards the asymmetric scheduler of the monitor of virtual machine of multi-core platform, comprise the general Scheduler module among the VMM, it is characterized in that, also comprise the service nuclear request analysis module in the management domain, module, the top Scheduler module of system, service dispatch device module, shared drive mechanism module and request enquiry module are put in Domain specific service caryogamy among the VMM, hybrid virtualization module among the HVMDomain wherein comprises topological identification of platform processor nuclear and analysis submodule and the distribution of service nuclear again and reclaims submodule in the top Scheduler module of system;
The request analysis module is examined in service, is used for serving the configuration of examining when HVM Domain creates, and the request to create of serving nuclear is sent in the type and the quantity parsing of the service nuclear that indicates in the configuration file when being responsible for HVM Domain client computer establishment just to VMM;
The caryogamy of Domain specific service is put module and then is responsible for the service nuclear request to create of the HVMDomain that preliminary analysis service nuclear request analysis module sends and sends to the top Scheduler module of system then, and the information that its service nuclear phase of management closes in the life cycle of whole HVM Domain;
The top Scheduler module of system is mainly used to the processor core of energy usefulness in the Management Calculation machine platform, and in global scope to general Scheduler module and service the adjusting of Scheduler module with processor nuclear resource bitmap, be responsible for finishing caryogamy and put request in the module from the Domain specific service, according to platform processor nuclear topology identification with analyze topology information in the submodule, the service nuclear of use distributes and reclaims submodule and create service nuclear or reclaim service nuclear; Topology identification of platform processor nuclear and analysis submodule are responsible for the processor core resource that has in the system is discerned, processor core after being identified, add top scheduler use and manage can be with in the processor nuclear resource bitmap, top scheduler begins to move on these processor cores after the initialization; Distribution of service nuclear and recovery submodule are responsible for serving the establishment or the recovery of nuclear, can be from system with distinguishing service nuclear the processor nuclear resource bitmap, and insert service dispatch device module can be with in the processor nuclear resource bitmap, perhaps can examine from service dispatch device module with deletion service the processor nuclear resource bitmap;
Service dispatch device module is responsible for serving the management and the scheduling of nuclear;
Shared drive mechanism module is responsible for responding the mapping request of HVM Domain, and after shared drive is set up, the trigger request enquiry module;
The hybrid virtualization module is at first to VMM request mapping shared drive, and the mapping request of shared drive mechanism module responds HVM Domain is set up the shared drive zone in VMM; After shared drive was set up, the hybrid virtualization module was sent the services request of HVMDomain to service nuclear, and these requests are written in the shared drive zone;
The request enquiry module is responsible for inquiring about the services request to service nuclear in the affiliated shared drive zone of HVM Domain, and these requests are sent to service dispatch device module;
Wherein, VMM represents the virtual monitor device, and HVM represents the hardware virtual manager, and Domain represents the territory.
2. the asymmetric scheduler of the monitor of virtual machine towards multi-core platform according to claim 1, it is characterized in that the load of the top Scheduler module of described system in system initialization, HVM Domain create, serve nuclear request to create, the recovery of service nuclear, HVM Domain destruction, general purpose core and the adjustment of service nuclear ratio, global scope is triggered when adjusting occasion.
3. application rights requires the dispatching method of the 1 described asymmetric scheduler of monitor of virtual machine towards multi-core platform, it is characterized in that, comprises the steps:
Step 1, client computer are created the request to create of startup and service nuclear and are sent;
When HVM Domain client computer is created, in management domain, have the establishment configuration file of a corresponding client machine, can indicate the quantity of the vcpu of client computer use in this document, and with the binding information of concurrent physical processor nuclear, also can indicate the type and the quantity of required service nuclear simultaneously in this configuration file, concrete type is by special appointment and realization among the different embodiment; Service nuclear request analysis module in the management domain is resolved the type and the quantity of the nuclear of the service in this configuration file, and will create the overshoot Hypercall of the paired VMM of shift in demand, and this overshoot put module with the Domain specific service caryogamy that is transferred among the VMM, and the caryogamy of Domain specific service is put module after adapter overshoot is with Hypercall, adopt all information of a new HVM Domain who creates of structure management, the configuration information that comprises service nuclear, and tentatively resolve overshoot Hypercall, trigger the top Scheduler module operation of system among the VMM then; Wherein vcpu represents virtual processor, and Hypercall represents hypercalls;
Step 2, identification of processor core topology and the distribution of analyzing and serving nuclear;
Identification of processor core topology and analysis submodule in the top Scheduler module of system can be discerned the processor core resource that has in the system, what processor core top Scheduler module of adding system after being identified was used and managed can use in the processor nuclear resource bitmap, the top Scheduler module of system begins to move on these processor cores after the initialization, is allocated to the general Scheduler module among the VMM or is allocated to the scheduling that the service stone grafting is subjected to the service dispatch device module among the VMM up to these processor cores;
During system initialization, the top Scheduler module of system is initialised, after the top Scheduler module of system is received the request of creating service nuclear, at first analyze the topology information that submodule obtains platform processor nuclear according to platform processor nuclear topological sum, the concurrent physical processor nuclear phase neighbour's of the vcpu binding of selection of employing nearby principle and HVM Domain processor core, if setting, vcpu do not need binding concurrent physical processor nuclear, then according to the whole strategy of system, be performance preferentially or the predetermined policy of energy conservation priority select, when performance is preferential, the preferential processor core of selecting the free time then preferentially selects the processor core in hyperthread or the same encapsulation to examine as service as service nuclear during energy conservation priority;
After service nuclear is selected, be responsible for serving the final constructive process of nuclear by distribution of the nuclear of the service in the top Scheduler module of system and recovery submodule, service nuclear distributes and reclaims submodule can be with distinguishing service nuclear the processor nuclear resource bitmap from system, and insert service dispatch device module can be with in the processor nuclear resource bitmap, examine by interrupting the IPI call service between processor then, and make it begin to carry out from the function of appointment; Service nuclear is created and when beginning to carry out, really do not begin to provide service, after the shared drive zone was created in the HVMDomain application, service nuclear just can really be activated, initiatively arrive shared drive zone polling tasks, perhaps serviced Scheduler module just is activated when dispatching service is provided; The recovery of service nuclear is to trigger when the affiliated HVMDomain of service nuclear is destroyed, similar with the process of creating, the top Scheduler module of system is called the distribution of service nuclear and is reclaimed submodule and reclaim, and the service that will reclaim nuclear is redistributed, collocation strategy according to system, with the service of free time nuclear, join general Scheduler module can usefulness processor nuclear resource bitmap in, or join in system's idling-resource;
Step 3, the two-level scheduler device is to general purpose core and the scheduling of service nuclear;
The VMM initial start stage, all processor cores in the computer platform are all by top processor module management of system and scheduling, along with the establishment of HVM Domain and corresponding vcpu, general Scheduler module has had scheduler object, is initialised and moves on the general purpose core of being managed;
After service nuclear is created, service dispatch device module begins service nuclear is dispatched, it understands the services request in the shared drive zone of each HVM Domain of poll and VMM, the call request enquiry module obtains new service role and sends to service dispatch device module, service dispatch device module is examined according to the service under the task, task is inserted into the task queue of this service nuclear, and triggers the operation of service nuclear; If service nuclear is designated as direct poll pattern when creating, then do not need service dispatch device module that task is distributed to service nuclear, the direct poll shared drive of service nuclear zone obtains services request;
Step 4, client computer are sent the request of services request and corresponding with service nuclear customer in response machine;
After HVM Domain is created and starts, load the hybrid virtualization module among the HVM Domain, the hybrid virtualization module is sent the request of mapping shared drive to VMM, shared drive mechanism module responds request among the VMM, mapping shared drive zone in the space of VMM, and the start address in shared drive zone and size information are write the caryogamy of Domain specific service put in the information structure of module management, the service code that shared drive shines upon on the corresponding service nuclear in successful back brings into operation;
The hybrid virtualization module sends services request by the mode of writing shared drive, services request also can be sent by the application-specific that operates in user's attitude, by having operating system user attitude now, kernel state module data and control method are transferred to hybrid virtualization module agency and are sent, after services request sends, request enquiry module among the VMM will be inquired about these requests according to certain frequency, and these requests are transferred to the service nuclear that service dispatch device module is distributed correspondence, service nuclear can be carried out the task queue of oneself according to the scheduling strategy in the service dispatch device module, also can adopt a kind of mode of serving the shared drive zone of the direct poll correspondence of nuclear, in case request is arranged, directly carry out, this situation is applicable to that real-time requires perhaps frequently to carry out and determining the key service of entire system performance than higher service; After service nuclear is finished services request, result is write back shared drive by shared drive mechanism module, other kernel modules among hybrid virtualization module or the HVM Domain just can be read the service execution result who returns.
4. according to the asymmetric dispatching method of the monitor of virtual machine towards multi-core platform described in the claim 3, it is characterized in that, identification of processor core topology and analysis submodule can be discerned the processor core resource that has in the platform in the described step 2, the main basis of identification is, whether in an encapsulation, whether whether share L2 cache, be the hyperthread on the same nuclear.
5. according to the asymmetric dispatching method of the monitor of virtual machine towards multi-core platform described in the claim 3, it is characterized in that, nearby principle is adopted in the distribution of the nuclear of service described in the step 2, be specially: if vcpu is a hyperthread, then service nuclear is preferentially selected from the most contiguous hyperthread, if vcpu has been tied to a processor core, then service nuclear preferentially distributes from other processor core that is in together in the encapsulation, further, if the processor core of some processor core and vcpu binding is shared L2 cache in the same encapsulation, other are not shared, the then preferential processor core of selecting to share L2 cache.
6. according to the asymmetric dispatching method of the monitor of virtual machine towards multi-core platform described in the claim 3, it is characterized in that, the shared drive of setting up between HVM Domain and the VMM in the described step 4, unified address space is arranged in VMM, the mapping space that oneself is also arranged in HVM Domain internal system, be unified management to the request zone in these shared drives among the VMM, the synchronous communication mode of poll or the mode that adopts message to transmit are adopted in the management of request.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200910241837A CN101727351B (en) | 2009-12-14 | 2009-12-14 | Multicore platform-orientated asymmetrical dispatcher for monitor of virtual machine and dispatching method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200910241837A CN101727351B (en) | 2009-12-14 | 2009-12-14 | Multicore platform-orientated asymmetrical dispatcher for monitor of virtual machine and dispatching method thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101727351A true CN101727351A (en) | 2010-06-09 |
| CN101727351B CN101727351B (en) | 2012-09-05 |
Family
ID=42448280
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN200910241837A Expired - Fee Related CN101727351B (en) | 2009-12-14 | 2009-12-14 | Multicore platform-orientated asymmetrical dispatcher for monitor of virtual machine and dispatching method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101727351B (en) |
Cited By (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101968749A (en) * | 2010-09-26 | 2011-02-09 | 华中科技大学 | Method for receiving message passing interface (MPI) information under circumstance of over-allocation of virtual machine |
| CN102223254A (en) * | 2011-06-17 | 2011-10-19 | 南京中兴新软件有限责任公司 | Monitoring system and method |
| CN102508696A (en) * | 2011-10-30 | 2012-06-20 | 北京方物软件有限公司 | Asymmetrical resource scheduling method and device |
| CN102622245A (en) * | 2012-02-24 | 2012-08-01 | 中国人民解放军国防科学技术大学 | Virtual machine automatic-starting control method under sun4v architecture |
| CN102831016A (en) * | 2012-08-01 | 2012-12-19 | 浪潮(北京)电子信息产业有限公司 | Physical machine recycle method of cloud computing and device thereof |
| CN102831015A (en) * | 2012-08-01 | 2012-12-19 | 华为技术有限公司 | Method and equipment for scheduling multi-core processor |
| CN103235738A (en) * | 2013-05-17 | 2013-08-07 | 浙江大学 | Energy-saving integration planning method of virtual machine based on application program performance constraint |
| CN103294540A (en) * | 2013-05-17 | 2013-09-11 | 北京航空航天大学 | Method of improving performance of Erlang virtual machines by Xeon Phi coprocessor |
| CN103365784A (en) * | 2013-06-27 | 2013-10-23 | 华为技术有限公司 | Method and device for memory recycle and allocation |
| CN103392314A (en) * | 2010-12-29 | 2013-11-13 | 思杰***有限公司 | Systems and methods for scalable n-core statistics aggregation |
| CN103778017A (en) * | 2012-10-19 | 2014-05-07 | 国际商业机器公司 | Affinity of virtual processor dispatching |
| CN103853597A (en) * | 2014-02-21 | 2014-06-11 | 北京神舟航天软件技术有限公司 | Method for realizing high-frequency data collection of embedded virtual machine platform |
| CN103870313A (en) * | 2012-12-17 | 2014-06-18 | ***通信集团公司 | Virtual machine task scheduling method and system |
| CN103945548A (en) * | 2014-04-29 | 2014-07-23 | 西安电子科技大学 | Resource distribution system and task/service scheduling method in C-RAN |
| 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 |
| US9026695B2 (en) | 2012-12-11 | 2015-05-05 | Gemtek Technology Co., Ltd. | Asymmetrical processing multi-core system and network device |
| 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 |
| CN104714849A (en) * | 2013-12-12 | 2015-06-17 | 国际商业机器公司 | System and method achieving optimum performance in comprehensive work load environment |
| CN104793924A (en) * | 2014-01-21 | 2015-07-22 | 中兴通讯股份有限公司 | Calculation mission processing method and device |
| CN104834565A (en) * | 2014-02-12 | 2015-08-12 | 华为技术有限公司 | Method and device for dynamically deploying system services |
| CN105849670A (en) * | 2014-11-17 | 2016-08-10 | 联发科技股份有限公司 | Energy efficient multi-cluster system and its operations |
| CN106850565A (en) * | 2016-12-29 | 2017-06-13 | 河北远东通信***工程有限公司 | A kind of network data transmission method of high speed |
| CN107329822A (en) * | 2017-01-15 | 2017-11-07 | 齐德昱 | Towards the multi-core dispatching method based on super Task Network of multi-source multiple nucleus system |
| CN107391245A (en) * | 2017-07-18 | 2017-11-24 | 致象尔微电子科技（上海）有限公司 | A kind of software systems of multi core chip |
| CN107479943A (en) * | 2017-07-03 | 2017-12-15 | 北京东土科技股份有限公司 | Multiple operating system operation method and device based on industry internet operating system |
| US9990306B2 (en) | 2013-04-26 | 2018-06-05 | Huawei Technologies Co., Ltd. | Inter-manycore communications method and system |
| CN108255622A (en) * | 2018-01-17 | 2018-07-06 | 郑州云海信息技术有限公司 | The method and system that a kind of SMASH CLP functions are realized |
| CN108459906A (en) * | 2017-02-20 | 2018-08-28 | 华为技术有限公司 | A kind of dispatching method and device of VCPU threads |
| CN108694152A (en) * | 2017-04-11 | 2018-10-23 | 上海福赛特机器人有限公司 | Communication system between multinuclear, communication control method and server based on the system |
| CN108874500A (en) * | 2018-04-26 | 2018-11-23 | 南京大学 | A kind of application security guard method based on hardware virtualization technology |
| CN112153116A (en) * | 2020-08-27 | 2020-12-29 | 上海赫千电子科技有限公司 | Data sharing method for multiple virtual machines of central computing platform based on vehicle-mounted Ethernet |
| CN113238832A (en) * | 2021-05-20 | 2021-08-10 | 元心信息科技集团有限公司 | Scheduling method, device and equipment of virtual processor and computer storage medium |
| CN113382030A (en) * | 2020-03-10 | 2021-09-10 | 本无链科技(深圳)有限公司 | Design method and system of safety miniature server for block chain |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7620953B1 (en) * | 2004-10-05 | 2009-11-17 | Azul Systems, Inc. | System and method for allocating resources of a core space among a plurality of core virtual machines |
| CN101398768B (en) * | 2008-10-28 | 2011-06-15 | 北京航空航天大学 | Construct method of distributed virtual machine monitor system |
-
2009
- 2009-12-14 CN CN200910241837A patent/CN101727351B/en not_active Expired - Fee Related
Cited By (57)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101968749A (en) * | 2010-09-26 | 2011-02-09 | 华中科技大学 | Method for receiving message passing interface (MPI) information under circumstance of over-allocation of virtual machine |
| CN101968749B (en) * | 2010-09-26 | 2013-01-02 | 华中科技大学 | Method for receiving message passing interface (MPI) information under circumstance of over-allocation of virtual machine |
| CN103392314A (en) * | 2010-12-29 | 2013-11-13 | 思杰***有限公司 | Systems and methods for scalable n-core statistics aggregation |
| CN103392314B (en) * | 2010-12-29 | 2016-06-15 | 思杰***有限公司 | For the system and method that the N nuclear statistics information that can expand is polymerized |
| CN102223254A (en) * | 2011-06-17 | 2011-10-19 | 南京中兴新软件有限责任公司 | Monitoring system and method |
| CN102508696A (en) * | 2011-10-30 | 2012-06-20 | 北京方物软件有限公司 | Asymmetrical resource scheduling method and device |
| CN102508696B (en) * | 2011-10-30 | 2016-06-08 | 北京方物软件有限公司 | A kind of asymmetric resource regulating method and device |
| CN102622245A (en) * | 2012-02-24 | 2012-08-01 | 中国人民解放军国防科学技术大学 | Virtual machine automatic-starting control method under sun4v architecture |
| CN102622245B (en) * | 2012-02-24 | 2015-04-08 | 中国人民解放军国防科学技术大学 | Virtual machine automatic-starting control method under sun4v architecture |
| CN102831016A (en) * | 2012-08-01 | 2012-12-19 | 浪潮(北京)电子信息产业有限公司 | Physical machine recycle method of cloud computing and device thereof |
| CN102831015A (en) * | 2012-08-01 | 2012-12-19 | 华为技术有限公司 | Method and equipment for scheduling multi-core processor |
| CN102831016B (en) * | 2012-08-01 | 2014-10-01 | 浪潮(北京)电子信息产业有限公司 | Physical machine recycle method of cloud computing and device thereof |
| CN102831015B (en) * | 2012-08-01 | 2015-09-23 | 华为技术有限公司 | The dispatching method of polycaryon processor and equipment |
| CN103778017A (en) * | 2012-10-19 | 2014-05-07 | 国际商业机器公司 | Affinity of virtual processor dispatching |
| CN103778017B (en) * | 2012-10-19 | 2017-08-15 | 国际商业机器公司 | Improve the compatibility of virtual processor scheduling |
| US10013264B2 (en) | 2012-10-19 | 2018-07-03 | International Business Machines Corporation | Affinity of virtual processor dispatching |
| US9026695B2 (en) | 2012-12-11 | 2015-05-05 | Gemtek Technology Co., Ltd. | Asymmetrical processing multi-core system and network device |
| CN103870313B (en) * | 2012-12-17 | 2017-02-08 | ***通信集团公司 | Virtual machine task scheduling method and system |
| CN103870313A (en) * | 2012-12-17 | 2014-06-18 | ***通信集团公司 | Virtual machine task scheduling method and system |
| 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 |
| US9990306B2 (en) | 2013-04-26 | 2018-06-05 | Huawei Technologies Co., Ltd. | Inter-manycore communications method and system |
| CN103235738B (en) * | 2013-05-17 | 2016-04-27 | 浙江大学 | A kind of energy-conservation Conformity planning method of virtual machine based on application program capacity constraint |
| CN103235738A (en) * | 2013-05-17 | 2013-08-07 | 浙江大学 | Energy-saving integration planning method of virtual machine based on application program performance constraint |
| CN103294540A (en) * | 2013-05-17 | 2013-09-11 | 北京航空航天大学 | Method of improving performance of Erlang virtual machines by Xeon Phi coprocessor |
| CN103365784A (en) * | 2013-06-27 | 2013-10-23 | 华为技术有限公司 | Method and device for memory recycle and allocation |
| CN103365784B (en) * | 2013-06-27 | 2016-03-30 | 华为技术有限公司 | The method of Memory recycle and distribution and device |
| CN104714849A (en) * | 2013-12-12 | 2015-06-17 | 国际商业机器公司 | System and method achieving optimum performance in comprehensive work load environment |
| CN104714849B (en) * | 2013-12-12 | 2018-08-24 | 国际商业机器公司 | System and method for realizing optimum performance in synthetic workload environment |
| CN104793924B (en) * | 2014-01-21 | 2019-03-15 | 中兴通讯股份有限公司 | The processing method and processing device of calculating task |
| CN104793924A (en) * | 2014-01-21 | 2015-07-22 | 中兴通讯股份有限公司 | Calculation mission processing method and device |
| CN104834565B (en) * | 2014-02-12 | 2018-08-14 | 华为技术有限公司 | A kind of system service dynamic deployment method and device |
| CN104834565A (en) * | 2014-02-12 | 2015-08-12 | 华为技术有限公司 | Method and device for dynamically deploying system services |
| CN103853597B (en) * | 2014-02-21 | 2017-04-19 | 北京神舟航天软件技术有限公司 | Method for realizing high-frequency data collection of embedded virtual machine platform |
| CN103853597A (en) * | 2014-02-21 | 2014-06-11 | 北京神舟航天软件技术有限公司 | Method for realizing high-frequency data collection of embedded virtual machine platform |
| CN103945548A (en) * | 2014-04-29 | 2014-07-23 | 西安电子科技大学 | Resource distribution system and task/service scheduling method in C-RAN |
| CN103945548B (en) * | 2014-04-29 | 2018-12-14 | 西安电子科技大学 | Resource allocation system and task/business scheduling method in a kind of C-RAN network |
| CN105849670A (en) * | 2014-11-17 | 2016-08-10 | 联发科技股份有限公司 | Energy efficient multi-cluster system and its operations |
| 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 |
| CN106850565B (en) * | 2016-12-29 | 2019-06-18 | 河北远东通信***工程有限公司 | A kind of network data transmission method of high speed |
| CN106850565A (en) * | 2016-12-29 | 2017-06-13 | 河北远东通信***工程有限公司 | A kind of network data transmission method of high speed |
| CN107329822A (en) * | 2017-01-15 | 2017-11-07 | 齐德昱 | Towards the multi-core dispatching method based on super Task Network of multi-source multiple nucleus system |
| CN107329822B (en) * | 2017-01-15 | 2022-01-28 | 齐德昱 | Multi-core scheduling method based on hyper task network and oriented to multi-source multi-core system |
| CN108459906A (en) * | 2017-02-20 | 2018-08-28 | 华为技术有限公司 | A kind of dispatching method and device of VCPU threads |
| CN108694152B (en) * | 2017-04-11 | 2021-07-13 | 实时侠智能控制技术有限公司 | Communication system among multiple cores, communication control method based on system and server |
| CN108694152A (en) * | 2017-04-11 | 2018-10-23 | 上海福赛特机器人有限公司 | Communication system between multinuclear, communication control method and server based on the system |
| CN107479943A (en) * | 2017-07-03 | 2017-12-15 | 北京东土科技股份有限公司 | Multiple operating system operation method and device based on industry internet operating system |
| CN107479943B (en) * | 2017-07-03 | 2020-02-21 | 北京东土科技股份有限公司 | Multi-operating-system operation method and device based on industrial Internet operating system |
| US10656961B2 (en) | 2017-07-03 | 2020-05-19 | Kyland Technology Co., Ltd | Method and apparatus for operating a plurality of operating systems in an industry internet operating system |
| CN107391245A (en) * | 2017-07-18 | 2017-11-24 | 致象尔微电子科技（上海）有限公司 | A kind of software systems of multi core chip |
| CN108255622A (en) * | 2018-01-17 | 2018-07-06 | 郑州云海信息技术有限公司 | The method and system that a kind of SMASH CLP functions are realized |
| CN108874500A (en) * | 2018-04-26 | 2018-11-23 | 南京大学 | A kind of application security guard method based on hardware virtualization technology |
| CN108874500B (en) * | 2018-04-26 | 2022-10-18 | 南京大学 | Application program safety protection method based on hardware virtualization technology |
| CN113382030A (en) * | 2020-03-10 | 2021-09-10 | 本无链科技(深圳)有限公司 | Design method and system of safety miniature server for block chain |
| CN112153116A (en) * | 2020-08-27 | 2020-12-29 | 上海赫千电子科技有限公司 | Data sharing method for multiple virtual machines of central computing platform based on vehicle-mounted Ethernet |
| CN112153116B (en) * | 2020-08-27 | 2023-03-24 | 上海赫千电子科技有限公司 | Data sharing method for multiple virtual machines of central computing platform based on vehicle-mounted Ethernet |
| CN113238832A (en) * | 2021-05-20 | 2021-08-10 | 元心信息科技集团有限公司 | Scheduling method, device and equipment of virtual processor and computer storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101727351B (en) | 2012-09-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101727351B (en) | Multicore platform-orientated asymmetrical dispatcher for monitor of virtual machine and dispatching method thereof | |
| CN102096596B (en) | Cloud computing service Cache system based on internal memory template of virtual machine | |
| CN100570565C (en) | Operating system service method and system based on strategy are provided in supervisory routine | |
| Daniels | Server virtualization architecture and implementation | |
| CN100573456C (en) | A kind of paralleling multi-processor virtual machine system | |
| US9405585B2 (en) | Management of heterogeneous workloads | |
| EP2718813B1 (en) | Operating system decoupled heterogeneous computing | |
| US8667500B1 (en) | Use of dynamic entitlement and adaptive threshold for cluster process balancing | |
| US20140130057A1 (en) | Scheduling jobs in a cluster | |
| CN103744716B (en) | A kind of dynamically interruption Well-Balanced Mapping method based on current VCPU dispatch state | |
| US11579908B2 (en) | Containerized workload scheduling | |
| CN101788920A (en) | CPU virtualization method based on processor partitioning technology | |
| CN101751284A (en) | I/O resource scheduling method for distributed virtual machine monitor | |
| CN101149728A (en) | Multiple core processing system and its management method | |
| Xie et al. | The performance analysis of Docker and rkt based on Kubernetes | |
| Ballesteros et al. | Nix: A case for a manycore system for cloud computing | |
| Hu et al. | GPGPU cloud: A paradigm for general purpose computing | |
| CN105100180A (en) | Cluster node dynamic loading method, device and system | |
| Sun et al. | Green challenges to system software in data centers | |
| Ventroux et al. | Scmp architecture: an asymmetric multiprocessor system-on-chip for dynamic applications | |
| Liu et al. | BSPCloud: A hybrid distributed-memory and shared-memory programming model | |
| Grushin et al. | On effective scheduling in computing clusters | |
| CN113626148B (en) | Terminal virtual machine generation system and method based on hybrid virtualization | |
| Cencerrado et al. | Support for urgent computing based on resource virtualization | |
| Panneerselvam et al. | Dynamic processors demand dynamic operating systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120905 Termination date: 20151214 |
|
| EXPY | Termination of patent right or utility model |