CN107480094A - A kind of pond server system architecture of fusion architecture - Google Patents
A kind of pond server system architecture of fusion architecture Download PDFInfo
- Publication number
- CN107480094A CN107480094A CN201710685507.9A CN201710685507A CN107480094A CN 107480094 A CN107480094 A CN 107480094A CN 201710685507 A CN201710685507 A CN 201710685507A CN 107480094 A CN107480094 A CN 107480094A
- Authority
- CN
- China
- Prior art keywords
- pcie
- pcie switch
- chip
- pieces
- architecture
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F15/00—Digital computers in general; Data processing equipment in general
- G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
- G06F15/161—Computing infrastructure, e.g. computer clusters, blade chassis or hardware partitioning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F15/00—Digital computers in general; Data processing equipment in general
- G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
- G06F15/163—Interprocessor communication
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F15/00—Digital computers in general; Data processing equipment in general
- G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
- G06F15/163—Interprocessor communication
- G06F15/17—Interprocessor communication using an input/output type connection, e.g. channel, I/O port
Abstract
The invention discloses a kind of pond server system architecture of fusion architecture, including calculation server node, PCIe Switch nodes, described calculation server node electrically connects with PCIe Switch nodes, and GPU computing hardwares Pooled resources or NVME storage hardwares Pooled resources carry out free distribution by the PCIe Switch plates on PCIe Switch nodes on server node.The present invention utilizes PCIe Switch ponds technology, the storage of Large Copacity is provided by NVME pondization extension, meets the needs of cold data, warm/hot data further separate, and extend by GPU pondization, the new hardware platform of AI applications is started, realizes the fast layout of AI applications.
Description
Technical field
The present invention relates to the pondization service of the internal cooling technical field of heat dissipation of server, more particularly to a kind of fusion architecture
Device system architecture.
Background technology
With the development of big data analysis application, increasing machine learning and AI applications need fast layout；Precipitation
The quick increase of data volume causes memory capacity to need fast lifting, and cold data, warm/hot data etc. to need further to divide
From server system architecture many of the prior art is limited by structure, cost, space etc., storage often be present and holds
Measure the problem of small, it is impossible to meet the needs of cold data, warm/hot data further separate well, whole server resource is not
Optimal efficiency can be played, cause the waste of pooling technology.
The content of the invention
For solving the above-mentioned problems of the prior art, it is an object of the present invention to provide a kind of pondization of fusion architecture clothes
Business device system architecture, the present invention utilize PCIe Switch ponds technology, and depositing for Large Copacity is provided by NVME pondization extension
Storage, meet the needs of cold data, warm/hot data further separate, and extended by GPU pondization, start AI (artificial intelligence
Can) application new hardware platform, realize AI application fast layout.
The present invention solve technical problem technical scheme be：
A kind of pond server system architecture of fusion architecture, including calculation server node, PCIe Switch nodes,
Described calculation server node electrically connects with PCIe Switch nodes, GPU computing hardwares Pooled resources or NVME storage hardwares pond
Resource carries out free distribution by the PCIe Switch plates on PCIe Switch nodes on server node.
Described PCIe Switch nodes include one block of PCIe Switch plate, two pieces of PCIe adapters, GPU card or
NVME hard disks.It is attached between the PCIe Switch plates and PCIe adapters by high speed connector.Described in two pieces
PCIe adapters are connected by PCIe x16 with 4 pieces of GPU cards or 16 pieces of NVME hard disks.
Chip one is installed on the PCIe Switch plates, described one end of chip one sets mini SAS HD ports,
Mini SAS HD ports are used for up connection calculation server, and the other end of chip one sets a PCIe x16 slot, made
PCIe Switch nodes pass through the descending carry out system cascade of Redriver cards or up connection calculation server；Chip one is in addition
Provided with four road PCIe x16, two pieces of PCIe adapters are connected respectively to by high speed connector, the PCIe adapters described in two pieces
It is connected by PCIe x16 with 4 pieces of GPU cards or 16 pieces of NVME hard disks.
Chip two is installed, chip two is connected with RJ45 interfaces, and chip two is drawn RJ45 and connect on the PCIe Switch plates
Mouth is used for network management and connected.
The chip one uses PEX9797 chips, and chip two uses AST2500 chips.
Described calculation server node is connected with PCIe Switch nodes by mini SAS HD cables.
Be connected using a calculation server with two PCIe Switch nodes, or using two calculation servers with
One PCIe Switch nodes connection, or be connected using a calculation server with one of PCIe Switch nodes,
Then saved by the Redriver cards of the PCIe Switch nodes by mini SAS HD cables and another PCIe Switch
The mini SAS HD ports connection of point.
Beneficial effects of the present invention：
1. the present invention can be provided GPU computing hardwares resource or NVME storage hardwares by the pond server system architecture
Source is integrated, and externally being sent and received by GPU computing hardwares resource or NVME storage hardwares resource needs task to be processed,
And to GPU computing hardwares resource or the dynamically distributes of NVME storage hardware resources so that whole pooling technology is accomplished optimal
Chi Hua.
Pass through 2. calculated in the pond server system architecture simultaneously between server node and PCIe Switch nodes
MiniSAS HD cables connect, and calculation server node can be made flexibly to be arranged in pairs or groups with PCIe Switch nodes, there is provided no
Same topological structure meets the needs of different scene.
3. the PCIe Switch nodes in the present invention can support 4 pieces of GPU cards or 16 NVME hard disks, pass through NVME's
Pondization extension provides the storage of Large Copacity, meets the needs of cold data, warm/hot data further separate, and deposition data amount
Requirement of the quick increase to server storage capacity, while extended by GPU pondization, the new hardware for starting AI applications is put down
Platform, the fast layout for realizing AI applications.
Brief description of the drawings
Fig. 1 is the structural representation one of embodiments of the invention 1；
Fig. 2 is the structural representation two of embodiments of the invention 1；
Fig. 3 is the structural representation of embodiments of the invention 2；
Fig. 4 is the structural representation two of embodiments of the invention 3.
Embodiment
For a better understanding of the present invention, embodiments of the present invention are explained in detail below in conjunction with the accompanying drawings.
Embodiment 1
It is connected using a calculation server main frame with two PCIe Switch nodes, such as Fig. 1, shown in Fig. 2, Yi Zhongrong
The pond server system architecture of framework is closed, including for depositing calculation server node, the PCIe of local computing resource
Switch nodes, described calculation server node are connected with PCIe Switch nodes by mini SAS HD cables, GPU meters
Calculate hardware resource or NVME storage hardwares carry out free distribution by PCIe Switch plates on server node.Pass through the pond
GPU computing hardwares resource or NVME storage hardwares can be integrated by changing server system architecture, and pass through certain management
GPU computing hardwares resource or NVME storage hardwares is allowed externally to send and receive need task to be processed so that whole pond
Resource accomplishes optimal pond, and provides different topological structures and meet the needs of different scene.
Described PCIe Switch nodes include one block of PCIe Switch plate, two pieces of PCIe adapters, GPU card or
NVME hard disks, it is attached between the PCIe Switch plates and PCIe adapters by high speed connector.Described in two pieces
PCIe adapters are connected by PCIe x16 with 4 pieces of GPU cards or 16 pieces of NVME hard disks.
PEX9797 chips are installed, described PEX9797 chips one end sets mini SAS on the PCIe Switch plates
HD ports, mini SAS HD ports are used for up connection calculation server, and the PEX9797 chips other end sets a PCIe
X16 slot, PCIe Switch nodes are made to be cascaded by a Redriver cards downlink system or up connection calculation server；
PEX9797 chips are additionally provided with four road PCIe x16, and two pieces of PCIe adapters, two pieces of institutes are connected respectively to by high speed connector
The PCIe adapters stated are connected by PCIe x16 with 4 pieces of GPU cards or 16 pieces of NVME hard disks.
AST2500 chips are installed, AST2500 chips are connected with RJ45 interfaces on the PCIe Switch plates.
AST2500 chips are connected by 1 RJ45 interface for network management.
Embodiment 2
As shown in figure 3, the present embodiment uses the structure that a calculation server is connected with two PCIe Switch nodes.
Wherein calculation server is connected with a PCIe Switch node, then passes through the Redriver cards of the PCIe Switch nodes
It is connected by mini SAS HD cables with the mini SAS HD ports of another PCIe Switch node.Other are with reference to implementation
Example 1, will not be repeated here.
Embodiment 3
As shown in figure 4, the present embodiment is so that using two calculation servers and a PCIe switch-box, (PCIe Switch are saved
Point) connection.Other will not be repeated here with reference to embodiment 1.
Although above-mentioned the embodiment of invention is described with reference to accompanying drawing, not to the scope of the present invention
Limitation, on the basis of technical scheme, those skilled in the art need not pay creative work and can make
Various modifications or deformation still within protection scope of the present invention.
Claims (7)
1. a kind of pond server system architecture of fusion architecture, it is characterized in that, including calculation server node, PCIe
Switch nodes, described calculation server node electrically connect with PCIe Switch nodes, GPU computing hardwares Pooled resources or
NVME storage hardwares Pooled resources are carried out freely by the PCIe Switch plates on PCIe Switch nodes on server node
Distribution.
2. a kind of pond server system architecture of fusion architecture as claimed in claim 1, it is characterized in that, described PCIe
Switch nodes include one block of PCIe Switch plate, two pieces of PCIe adapters, GPU card or NVME hard disks；The PCIe
It is attached between Switch plates and PCIe adapters by high speed connector, the PCIe adapters described in two pieces pass through PCIe
X16 is connected with 4 pieces of GPU cards or 16 pieces of NVME hard disks.
3. a kind of pond server system architecture of fusion architecture as claimed in claim 2, it is characterized in that, the PCIe
Chip one is installed, described one end of chip one sets mini SAS HD ports, and mini SAS HD ports are used on Switch plates
Up connection calculation server, the other end of chip one set a PCIe x16slot, pass through PCIe Switch nodes
The descending carry out system cascade of Redriver cards or up connection calculation server；Chip one is additionally provided with four road PCIe x16, leads to
Cross high speed connector and be connected respectively to two pieces of PCIe adapters, the PCIe adapters described in two pieces pass through PCIe x16 and 4 pieces of GPU
Card or 16 pieces of NVME hard disks connections.
4. a kind of pond server system architecture of fusion architecture as claimed in claim 3, it is characterized in that, the PCIe
Chip two is installed, chip two is connected with RJ45 interfaces, and chip two is drawn RJ45 interfaces and connected for network management on Switch plates
Connect.
5. a kind of pond server system architecture of fusion architecture as claimed in claim 1, it is characterized in that, the chip one is adopted
With PEX9797 chips, chip two uses AST2500 chips.
6. a kind of pond server system architecture of fusion architecture as claimed in claim 1, it is characterized in that, described calculating clothes
Business device node is connected with PCIe Switch nodes by mini SAS HD cables.
7. a kind of pond server system architecture of fusion architecture as claimed in claim 1, it is characterized in that, using a calculating
Server is connected with two PCIe Switch nodes, or is connected using two calculation servers and a PCIe Switch node
Connect, or be connected using a calculation server with one of PCIe Switch nodes, then pass through the PCIe Switch
The Redriver cards of node are connected by the mini SAS HD ports of mini SAS HD cables and another PCIe Switch node
Connect.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710685507.9A CN107480094A (en) | 2017-08-11 | 2017-08-11 | A kind of pond server system architecture of fusion architecture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710685507.9A CN107480094A (en) | 2017-08-11 | 2017-08-11 | A kind of pond server system architecture of fusion architecture |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107480094A true CN107480094A (en) | 2017-12-15 |
Family
ID=60600330
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710685507.9A Pending CN107480094A (en) | 2017-08-11 | 2017-08-11 | A kind of pond server system architecture of fusion architecture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107480094A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108710418A (en) * | 2018-08-06 | 2018-10-26 | 郑州云海信息技术有限公司 | A kind of GPU-Switch structures cabinet |
| CN108984309A (en) * | 2018-08-07 | 2018-12-11 | 郑州云海信息技术有限公司 | A kind of RACK server resource pond system and method |
| CN109933554A (en) * | 2019-03-20 | 2019-06-25 | 浪潮商用机器有限公司 | A kind of NVMe hard disk expansion apparatus based on GPU server |
| CN110069274A (en) * | 2019-05-08 | 2019-07-30 | 苏州浪潮智能科技有限公司 | Pond server ReDriver chip configures update method and device |
| CN110221999A (en) * | 2019-05-23 | 2019-09-10 | 深圳市同泰怡信息技术有限公司 | It is realized based on standard PCIE and supports GPU and M2 SSD extension adapter and implementation method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090167771A1 (en) * | 2007-12-28 | 2009-07-02 | Itay Franko | Methods and apparatuses for Configuring and operating graphics processing units |
| CN104202359A (en) * | 2014-08-12 | 2014-12-10 | 浪潮电子信息产业股份有限公司 | NVMe SSD virtualization design method based on blade server |
| CN104932837A (en) * | 2015-06-05 | 2015-09-23 | 浪潮电子信息产业股份有限公司 | Storage pool framework |
| CN204883525U (en) * | 2015-09-09 | 2015-12-16 | 浪潮电子信息产业股份有限公司 | External switching card |
| CN205229928U (en) * | 2015-12-18 | 2016-05-11 | 山东海量信息技术研究院 | Extension GPU's blade server |
| CN106649014A (en) * | 2016-12-28 | 2017-05-10 | 郑州云海信息技术有限公司 | Automatic testing method of calculating type server which supports multiple GPUs |
-
2017
- 2017-08-11 CN CN201710685507.9A patent/CN107480094A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090167771A1 (en) * | 2007-12-28 | 2009-07-02 | Itay Franko | Methods and apparatuses for Configuring and operating graphics processing units |
| CN104202359A (en) * | 2014-08-12 | 2014-12-10 | 浪潮电子信息产业股份有限公司 | NVMe SSD virtualization design method based on blade server |
| CN104932837A (en) * | 2015-06-05 | 2015-09-23 | 浪潮电子信息产业股份有限公司 | Storage pool framework |
| CN204883525U (en) * | 2015-09-09 | 2015-12-16 | 浪潮电子信息产业股份有限公司 | External switching card |
| CN205229928U (en) * | 2015-12-18 | 2016-05-11 | 山东海量信息技术研究院 | Extension GPU's blade server |
| CN106649014A (en) * | 2016-12-28 | 2017-05-10 | 郑州云海信息技术有限公司 | Automatic testing method of calculating type server which supports multiple GPUs |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108710418A (en) * | 2018-08-06 | 2018-10-26 | 郑州云海信息技术有限公司 | A kind of GPU-Switch structures cabinet |
| CN108710418B (en) * | 2018-08-06 | 2023-09-22 | 郑州云海信息技术有限公司 | GPU-Switch structure case |
| CN108984309A (en) * | 2018-08-07 | 2018-12-11 | 郑州云海信息技术有限公司 | A kind of RACK server resource pond system and method |
| CN109933554A (en) * | 2019-03-20 | 2019-06-25 | 浪潮商用机器有限公司 | A kind of NVMe hard disk expansion apparatus based on GPU server |
| CN110069274A (en) * | 2019-05-08 | 2019-07-30 | 苏州浪潮智能科技有限公司 | Pond server ReDriver chip configures update method and device |
| CN110221999A (en) * | 2019-05-23 | 2019-09-10 | 深圳市同泰怡信息技术有限公司 | It is realized based on standard PCIE and supports GPU and M2 SSD extension adapter and implementation method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107480094A (en) | A kind of pond server system architecture of fusion architecture | |
| CN106293508B (en) | Data-storage system and method | |
| CN109190276A (en) | FPGA prototype verification system | |
| US9904577B2 (en) | Hybrid heterogeneous host system, resource configuration method and task scheduling method | |
| CN107102964A (en) | A kind of method that GPU cluster expansion is carried out using high-speed connector | |
| CN104601684A (en) | Cloud server system | |
| CN109408451B (en) | Graphic processor system | |
| CN104699654A (en) | Interconnection adapting system and method based on CHI on-chip interaction bus and QPI inter-chip interaction bus | |
| CN104360982A (en) | Implementation method and system for host system directory structure based on reconfigurable chip technology | |
| CN104035525A (en) | Computational node | |
| CN110086854A (en) | A kind of privately owned cloud system of distribution | |
| CN105099776A (en) | Cloud server management system | |
| CN115586964A (en) | Resource sharing device, resource management device, and resource management method | |
| CN206489563U (en) | Heterogeneous computing system | |
| CN109410117B (en) | Graphics processor system | |
| CN106844263B (en) | Configurable multiprocessor-based computer system and implementation method | |
| CN209248436U (en) | A kind of expansion board clamping and server | |
| CN104641593B (en) | Web plate and communication equipment | |
| US11138146B2 (en) | Hyperscale architecture | |
| CN108363461A (en) | A method of it carrying out blade server IO using non-transparent bridge and shares and extend | |
| CN105224496A (en) | A kind of flow for dynamic reconfigurable system | |
| CN104580527A (en) | Multiple-I/O high-density multi-node server system designing method oriented to cloud server application | |
| US11055252B1 (en) | Modular hardware acceleration device | |
| CN210466253U (en) | Server with high-density GPU expansion capability | |
| CN109739802B (en) | Computing cluster and computing cluster configuration method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171215 |