CN105959404A - GPU virtualization platform based on cloud computing - Google Patents
GPU virtualization platform based on cloud computing Download PDFInfo
- Publication number
- CN105959404A CN105959404A CN201610474412.8A CN201610474412A CN105959404A CN 105959404 A CN105959404 A CN 105959404A CN 201610474412 A CN201610474412 A CN 201610474412A CN 105959404 A CN105959404 A CN 105959404A
- Authority
- CN
- China
- Prior art keywords
- gpu
- cloud computing
- server
- platform
- vitualization
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Debugging And Monitoring (AREA)
Abstract
The invention discloses a GPU virtualization platform based on cloud computing. The GPU virtualization platform based on the cloud computing is characterized in that a multi-GPU accelerated computing platform architecture is utilized by a virtual machine in a redirection manner; a GPU management end is used for improving the processing capability of the virtual machine; a component in the virtual machine calls a GPU by intercepting an application program, and redirects to a local privileged domain or a remote GPU server, so that the GPU call of the application program is executed on the local privileged domain or the remote GPU server, and a result is returned to the virtual machine after the execution is finished. The invention provides the GPU virtualization platform based on the cloud computing with high performance and low cost.
Description
Technical field
The present invention relates to a kind of virtual platform, particularly relate to a kind of GPU vitualization platform based on cloud computing.
Background technology
Virtualization is the core technology basis of cloud computing, and the advantage such as cost savings that it is brought, safety enhancing is gradually
Obtain the accreditation of people, be the study hotspot of computer science.Intel Virtualization Technology by the virtualization to hardware resource,
Multiple identical computer hardware platforms can be simulated on one computer such that it is able to run multiple operating system also simultaneously
Realize mutually isolated, improve the utilization ratio of server, in server merging, network security, calculating data protection, high-performance
Calculate and there is substantial amounts of application in the field such as credible.In recent years, performance and the function of Graphics Processing Unit (GPU) dramatically increased.
The function of GPU is no longer limited to image procossing, and develop into one has high calculating peak value and the height of memory bandwidth simultaneously
Parallel processor.Along with some support GPGPU(general graphical processing unit) the pushing away of technology (such as CUDA) that calculate
Going out, the application of GPGPU is more and more extensive.Powerful computation capability due to GPU so that increasing high property
Energy computing introduces the heterogeneous schemas of CPU+GPU.But, on the one hand the power consumption of GPU is relatively big, if each node
It is equipped with GPU, then may be greatly increased the power consumption of cluster；On the one hand, the computation capability of GPU is powerful, and in major part
In computing, GPU, as coprocessor, only accelerates the parallel section in code so that the utilization rate of GPU is the highest；Another
Aspect is due to the closure of GPU, and virtual machine cannot directly utilize GPU and carry out speed-up computation.This makes GPU at virtual machine
The application of aspect is limited by very large.
Summary of the invention
In order to solve the problems of the prior art, the present invention provides the GPU of the cloud computing of a kind of high-performance, low cost virtual
Change platform.
In order to solve the problems referred to above, the technical solution used in the present invention is:
A kind of GPU vitualization platform based on cloud computing, it is characterised in that: virtual machine utilizes many GPU to accelerate in redirection mode
The platform architecture calculated, utilizes GPU to manage the end lifting to virtual machine disposal ability, and the assembly in virtual machine is applied by interception
GPU is called by program, and is redirected to the machine privileged domain or long-range GPU server so that the GPU of application program is invoked at spy
Perform on power territory or remotely GPU server, and return results to virtual machine having performed.
Aforesaid a kind of GPU vitualization platform based on cloud computing, it is characterised in that: under virtual environment, use C/S side
Formula, it is achieved that many GPU cooperated computing.
Aforesaid a kind of GPU vitualization platform based on cloud computing, it is characterised in that: GPU management end is to GPU service end
The request of module and virtual-machine client module processes, the resource registering of response GPU server module, and supervises in real time
Control the load of each long-range GPU server.
Aforesaid a kind of GPU vitualization platform based on cloud computing, it is characterised in that: GPU management end is also to each GPU
The task of server is adjusted, and is managed the calculating resource of each GPU server, simultaneously according to load response virtual machine
The request of client modules, and the GPU server mated most for its distribution.
Aforesaid a kind of GPU vitualization platform based on cloud computing, it is characterised in that: described GPU is cloud computing engine
Platform architecture, implementation content includes the realization of cloud storage system, the realization of " C+G " cloud computing engine algorithms.
Aforesaid a kind of GPU vitualization platform based on cloud computing, it is characterised in that: described GPU server module node
Device is 2U two-way server.
Aforesaid a kind of GPU vitualization platform based on cloud computing, it is characterised in that: on GPU, application is for quickly
The page locking page in memory of storage.
Aforesaid a kind of GPU vitualization platform based on cloud computing, it is characterised in that: make GPU render by programming model
Calculate and both MapReduce model combine, it is achieved a complete high-performance renders calculating system.
The beneficial effect that the present invention is reached: the present invention implements three layers of SPI model of cloud platform under " CPU+GPU " heterogeneous schemas
Specialized structure, it is important that the breakthrough to many GPU cooperated computing, many GPU vitualization and Cross Platform Technology is studied and right
The development of games of " on-demand application " and management platform, the free of game products " zero puts into " accesses operating service platform, and support is many
The exploitation of the big application platform of game three of the cross-platform access of media mixing terminal, its objective is to be embodied as middle-size and small-size development of games factory
Business provides multifunctional integrated one-stop, the Gao Xing such as collection development of games, game operation, game security, game marketing, game payment
Energy, the comprehensive cloud service platform of low cost.
Detailed description of the invention
Following example are only used for clearly illustrating technical scheme, and can not limit the present invention with this
Protection domain.
A kind of GPU vitualization platform based on cloud computing, it is characterised in that: virtual machine utilizes many GPU in redirection mode
The platform architecture of speed-up computation, utilizes GPU to manage the end lifting to virtual machine disposal ability, and the assembly in virtual machine is by intercepting
GPU is called by application program, and is redirected to the machine privileged domain or long-range GPU server so that the GPU of application program calls
In privileged domain or remotely perform on GPU server, and return results to virtual machine having performed.
Under virtual environment, use C/S mode, it is achieved that many GPU cooperated computing.GPU management end is to GPU service end mould
The request of block and virtual-machine client module processes, the resource registering of response GPU server module, and monitors in real time
The load of each long-range GPU server.
The task of each GPU server is also adjusted by GPU management end, the calculating resource to each GPU server
It is managed, simultaneously according to the request of load response virtual-machine client module, and distributes, for it, the GPU server mated most.
Described GPU is the platform architecture of cloud computing engine, and implementation content includes the realization of cloud storage system, and " C+G " cloud computing engine is calculated
The realization of method.Described GPU server module node apparatus is 2U two-way server.On GPU, application is for quick storage
Page locking page in memory.Make GPU render calculating by programming model and both MapReduce model combine, it is achieved one complete
High-performance render calculating system.
The invention have the characteristics that:
(1) " CPU+GPU " heterogeneous schemas is applied to cloud platform of playing, it is proposed that in a kind of virtualized environment, virtual machine is to reset
The platform architecture of many GPU speed-up computation is utilized to mode.Utilize the GPU lifting to virtual machine disposal ability, the group in virtual machine
GPU is called by part by intercepting application program, and is redirected to the machine privileged domain or long-range GPU server so that application journey
The GPU of sequence is invoked at privileged domain or remotely performs on GPU server, and returns results to virtual machine having performed.
(2) have studied many GPU system under many GPU based on MapReduce parallel computational model and virtualized environment to calculate
Implementation.Under actual physics environment, make GPU render by programming model and calculate and the advantage phase of both MapReduce model
In conjunction with, it is achieved a complete high-performance renders calculating system.Under virtual environment, use C/S mode, it is achieved that many GPU assist
With calculating, and devise Monte Carlo (MC) and verify the acceleration effect of many GPU cooperated computing under virtualized environment.
(3) the cross-platform access technology of object game cloud is proposed.Can solve the problem that existing gaming platform compatibility deficiency
An industry difficult problem, makes the user of different terminals or different platform can pass through the unified game data of virtual machine terminal access and journey
Sequence, it is provided that greatly facilitate.Game cloud service middleware is absorbed on the basis of system hardware facility virtualizes, and builds virtual
The middleware game services changed, and high availability, elastic calculation, management, the service ability of intelligent strategy scheduling are provided, define
MySpice system.
The construction of this platform is broadly divided into three parts: one is the construction of hardware system, and we use with GPU as cloud computing
The platform architecture of engine, implementation content includes the realization of cloud storage system, the realization of " C+G " cloud computing engine algorithms, and service
The construction of device cluster, applies this technique construction cloud platform, on the premise of ensureing the same function of platform and calculated performance, and pole
The earth reduces the input of construction cost.Two is the exploitation of software system, subnetwork game high in the clouds development system, game operation support
And safeguards system, cross-platform access three digest journals.Finally carry out the integration of each big system, platform feature and the debugging of performance and excellent
Change.
The present invention is with GPU as computing engines, and it is a kind of application wound that " CPU+GPU " heterogeneous schemas is applied to game cloud platform
Newly, fill the domestic gaps.This pattern can solve the problem that and traditional builds cloud platform with pure cpu model, and hardware puts into and data height
Can process the contradiction limited mutually, the feature that GPU highly-parallelization calculates disclosure satisfy that in cloud environment game graph high-performance
The efficient process of large-scale data in demand, and game operation such as render, can effectively reduce IT hardware resource simultaneously and put into,
Realize cloud platform construction cost to overlap with the height of cloud platform performance.Additionally, by this project implementation, company completes from list
Pure game research and development company is to the transformation of the comprehensive game services business collecting game research and development, operation, marketing integration development；Complete
The all standing taking into account moving game operation is runed to from web game to mobile terminal development of games, web game.Nearly 2 years of company
Accumulative help 150 people achieve employment.
The ultimate principle of the present invention, principal character and advantage have more than been shown and described.The technical staff of the industry should
Understanding, the present invention is not restricted to the described embodiments, and the simply explanation present invention's described in above-described embodiment and description is former
Reason, without departing from the spirit and scope of the present invention, the present invention also has various changes and modifications, these changes and improvements
Both fall within scope of the claimed invention.Claimed scope is by appending claims and equivalent circle thereof
Fixed.
Claims (8)
1. a GPU vitualization platform based on cloud computing, it is characterised in that: virtual machine utilizes many GPU to add in redirection mode
The platform architecture that speed calculates, utilizes GPU to manage the end lifting to virtual machine disposal ability, and the assembly in virtual machine should by intercepting
By program, GPU is called, and be redirected to the machine privileged domain or long-range GPU server so that the GPU of application program is invoked at
Perform on privileged domain or remotely GPU server, and return results to virtual machine having performed.
2. want a kind of based on cloud computing the GPU vitualization platform described in 1 according to right, it is characterised in that: under virtual environment,
Use C/S mode, it is achieved that many GPU cooperated computing.
3. want a kind of based on cloud computing the GPU vitualization platform described in 2 according to right, it is characterised in that: GPU manages end pair
The request of GPU server module and virtual-machine client module processes, the resource note of response GPU server module
Volume, and monitor the load of each long-range GPU server in real time.
4. want a kind of based on cloud computing the GPU vitualization platform described in 3 according to right, it is characterised in that: GPU management end is the most right
The task of each GPU server is adjusted, and is managed the calculating resource of each GPU server, simultaneously according to load
The request of response virtual-machine client module, and distribute, for it, the GPU server mated most.
5. want a kind of based on cloud computing the GPU vitualization platform described in 4 according to right, it is characterised in that: described GPU is cloud meter
Calculating the platform architecture of engine, implementation content includes the realization of cloud storage system, the realization of " C+G " cloud computing engine algorithms.
6. want a kind of based on cloud computing the GPU vitualization platform described in 5 according to right, it is characterised in that: described GPU services
End Module nodes device is 2U two-way server.
7. want a kind of based on cloud computing the GPU vitualization platform described in 6 according to right, it is characterised in that: apply on GPU
Page locking page in memory for quick storage.
8. want a kind of based on cloud computing the GPU vitualization platform described in 7 according to right, it is characterised in that: pass through programming model
Make GPU render calculating and both MapReduce model combine, it is achieved a complete high-performance renders calculating system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610474412.8A CN105959404A (en) | 2016-06-27 | 2016-06-27 | GPU virtualization platform based on cloud computing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610474412.8A CN105959404A (en) | 2016-06-27 | 2016-06-27 | GPU virtualization platform based on cloud computing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105959404A true CN105959404A (en) | 2016-09-21 |
Family
ID=56904675
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610474412.8A Pending CN105959404A (en) | 2016-06-27 | 2016-06-27 | GPU virtualization platform based on cloud computing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105959404A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109304031A (en) * | 2018-09-19 | 2019-02-05 | 电子科技大学 | A kind of virtualization cloud game platform based on isomery intelligent terminal |
| CN109582425A (en) * | 2018-12-04 | 2019-04-05 | 中山大学 | A kind of GPU service redirection system and method merged based on cloud with terminal GPU |
| CN112162854A (en) * | 2020-09-21 | 2021-01-01 | 南开大学 | Method, system and medium for scheduling calculation tasks between CPU-GPU |
| CN114039921A (en) * | 2021-11-22 | 2022-02-11 | 北京计算机技术及应用研究所 | ICE load balancing system suitable for heterogeneous computing platform |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102650950A (en) * | 2012-04-10 | 2012-08-29 | 南京航空航天大学 | Platform architecture supporting multi-GPU (Graphics Processing Unit) virtualization and work method of platform architecture |
| CN102880785A (en) * | 2012-08-01 | 2013-01-16 | 北京大学 | Method for estimating transmission energy consumption of source code grade data directed towards GPU program |
| CN103279330A (en) * | 2013-05-14 | 2013-09-04 | 江苏名通信息科技有限公司 | MapReduce multiple programming model based on virtual machine GPU computation |
-
2016
- 2016-06-27 CN CN201610474412.8A patent/CN105959404A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102650950A (en) * | 2012-04-10 | 2012-08-29 | 南京航空航天大学 | Platform architecture supporting multi-GPU (Graphics Processing Unit) virtualization and work method of platform architecture |
| CN102880785A (en) * | 2012-08-01 | 2013-01-16 | 北京大学 | Method for estimating transmission energy consumption of source code grade data directed towards GPU program |
| CN103279330A (en) * | 2013-05-14 | 2013-09-04 | 江苏名通信息科技有限公司 | MapReduce multiple programming model based on virtual machine GPU computation |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109304031A (en) * | 2018-09-19 | 2019-02-05 | 电子科技大学 | A kind of virtualization cloud game platform based on isomery intelligent terminal |
| CN109582425A (en) * | 2018-12-04 | 2019-04-05 | 中山大学 | A kind of GPU service redirection system and method merged based on cloud with terminal GPU |
| CN112162854A (en) * | 2020-09-21 | 2021-01-01 | 南开大学 | Method, system and medium for scheduling calculation tasks between CPU-GPU |
| CN114039921A (en) * | 2021-11-22 | 2022-02-11 | 北京计算机技术及应用研究所 | ICE load balancing system suitable for heterogeneous computing platform |
| CN114039921B (en) * | 2021-11-22 | 2024-04-19 | 北京计算机技术及应用研究所 | ICE load balancing system suitable for heterogeneous computing platform |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102650950B (en) | Platform architecture supporting multi-GPU (Graphics Processing Unit) virtualization and work method of platform architecture | |
| CN105959404A (en) | GPU virtualization platform based on cloud computing | |
| Reaño et al. | Local and remote GPUs perform similar with EDR 100G InfiniBand | |
| CN101976200B (en) | Virtual machine system for input/output equipment virtualization outside virtual machine monitor | |
| CN109636592A (en) | High-performance intelligence Contract Design | |
| CN102567080B (en) | Virtual machine position selection system facing load balance in cloud computation environment | |
| CN108093652A (en) | The simulation of application | |
| Song et al. | A load balancing scheme using federate migration based on virtual machines for cloud simulations | |
| CN103927193B (en) | Loading method, service end virtual machine during the migration operation of java application function | |
| Wu et al. | Android unikernel: Gearing mobile code offloading towards edge computing | |
| CN104967121B (en) | A kind of tidal current computing method of large-scale electrical power system node | |
| CN103685564A (en) | Plug-in application ability layer introduced industry application online operation cloud platform architecture | |
| Madhavapeddy et al. | Reconfigurable data processing for clouds | |
| Wu et al. | GPU-based power converter transient simulation with matrix exponential integration and memory management | |
| CN103747439B (en) | Wireless controller equipment, wireless authentication processing method, system, networking | |
| Coppola et al. | From embedded multi-core socs to scale-out processors | |
| Huang et al. | A quantum computing simulator scheme using MPI technology on cloud platform | |
| Yang et al. | Using pci pass-through for gpu virtualization with cuda | |
| Mentone et al. | CUDA virtualization and remoting for GPGPU based acceleration offloading at the edge | |
| Wu et al. | Research and application of SOA and cloud computing model | |
| Zhang et al. | Research on resource scheduling algorithm in cloud computing data center | |
| Wu et al. | Cloud Computing Operation Management Platform for Internet of Things of Construction Machinery | |
| Panchal et al. | Performance Enhancement of cloud computing with clustering | |
| Guo et al. | Hybrid computing hierarchy based on-line analysis service for power dispatching and control system | |
| Tang et al. | Parallel optimization of stencil computation base on sunway taihulight |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160921 |
|
| RJ01 | Rejection of invention patent application after publication |