CN102904943A - Cluster computing system hybrid communication method based on embedded processor memory interface - Google Patents
Cluster computing system hybrid communication method based on embedded processor memory interface Download PDFInfo
- Publication number
- CN102904943A CN102904943A CN2012103724186A CN201210372418A CN102904943A CN 102904943 A CN102904943 A CN 102904943A CN 2012103724186 A CN2012103724186 A CN 2012103724186A CN 201210372418 A CN201210372418 A CN 201210372418A CN 102904943 A CN102904943 A CN 102904943A
- Authority
- CN
- China
- Prior art keywords
- network
- queue
- memory interface
- data
- flush bonding
- 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
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention provides a cluster computing system hybrid communication method based on an embedded processor memory interface. A network controller integrated with an embedded processor is connected to a first network through a network transceiver. A parallel memory interface is connected to a second network through a hybrid network communication module. When a parallel interface state machine receives data from a memory bus which is connected with the parallel memory interface, the data are stored in a transmit queue. A data packet is received from the second network through a receiving control logic and the data packet is disassembled, then the data packet is selectively stored in a receive queue or a forward queue according to a destination address of the data packet, and the receive queue buffers data to be forwarded to the parallel memory interface. When the receive queue is non-null and the memory bus which is connected with the parallel memory interface is not occupied, the data are sent to the memory bus. The transmit queue and the forward queue are arbitrated by a send control logic, and the oldest data packet is selected based on an age strategy to be sent.
Description
Technical field
The present invention relates to the computing technique field, more particularly, the present invention relates to a kind of cluster computing system mixed communication method based on the flush bonding processor memory interface.
Background technology
Along with based on ARM(Advanced RISC Machines) development of architecture processor technology, academia and industrial quarters have been considered at present and have been started to develop Novel low power consumption group system based on the ARM flush bonding processor.Issued the system of a " Mini supercomputer " by name on May 1st, 2011 such as Sandia National Laboratory, this system is comprised of 196 Gumstix Overo Tide computing nodes, per seven Gumstix Overo Tide are integrated on the Stagecoach motherboard, then realize node interconnection by Ethernet.Germany Ludwig-Maximilians-Universitat (LMU) university uses AppleTV to build a group system in April, 2011, each node of this system comprises Cortex-A8 processor, PowerVR SGX535GPU and the 256MB internal memory of a 1GHz, network using 100 m ethernet.
But, at first, technique scheme is subject to the restriction that the flush bonding processor network interface is single, performance is not high, specifically, the flush bonding processor of main flow mainly towards fields such as mobile application, does not usually possess wired network interface or only is equipped with the lower network interface of this class performance of 100 m ethernet at present.Therefore, when making up group system based on flush bonding processor, network performance can't satisfy the demand of high-performance data transmission.
Secondly, technique scheme is subject to the limited restriction of flush bonding processor extensive interface; Specifically, usually, as SOC (system on a chip), flush bonding processor can comprise a series of peripheral interface, and such as USB 2.0, GPIO, I2C etc., these interfaces all belong to the following low-speed interface of gigabit.Although partially embedded processor is arranged, ARM flush bonding processor such as Marvel company has comprised PCIE interface at a high speed, and can realize the express network interface by the PCIE Interface Expanding easily, but the flush bonding processor product of most companies does not comprise this interface, therefore, the method also lacks universality.
Summary of the invention
Technical problem to be solved by this invention is for there being defects in the prior art, provide a kind of and can solve the shortcoming that the flush bonding processor network interface is single, performance is weak, by realizing a hybrid network communication module based on parallel memory interface, for the cluster computing system of building based on flush bonding processor provides more at a high speed network interface and hybrid network group-net communication method more flexibly.
According to the present invention, a kind of cluster computing system mixed communication method based on the flush bonding processor memory interface is provided, it comprises: the network controller that flush bonding processor is integrated is connected to first network by network transceivers; And the parallel memory interface that flush bonding processor is integrated is connected to second network by the hybrid network communication module.
Preferably, first network is Ethernet, and network transceivers is ethernet transceiver, and network controller is ethernet controller.
Preferably, the hybrid network communication module comprises: parallel interface state machine, transmit queue, forwarding queue, receiving queue, transmission control logic, reception control logic and multiplexer; Wherein, the parallel interface state machine is connected to the data input pin of transmit queue and the data output end of receiving queue; The data output end of transmit queue is connected to the first data input pin of multiplexer; The data input pin of receiving queue is connected to the reception control logic; The data output end that sends control logic is connected to the second data input pin of multiplexer, and the data input pin that sends control logic is connected to the reception control logic; The output that sends control logic is connected to the control signal input of multiplexer; And the data output end of multiplexer is connected to second network.
Preferably, described cluster computing system mixed communication method based on the flush bonding processor memory interface also comprises: when the parallel interface state machine receives data from the memory bus that is connected with parallel memory interface, deposit data is arrived transmit queue.
Preferably, described cluster computing system mixed communication method based on the flush bonding processor memory interface also comprises: utilize the reception control logic to receive packet and disassemble packet from second network, then according to the destination address of packet, selection is stored in receiving queue or forwarding queue with packet, and wherein the receiving queue buffer memory waits to be forwarded to the data of parallel memory interface.
Preferably, described cluster computing system mixed communication method based on the flush bonding processor memory interface also comprises: when the receiving queue non-NULL, and the memory bus that is connected with parallel memory interface sends to data on the memory bus when vacant.
Preferably, described cluster computing system mixed communication method based on the flush bonding processor memory interface also comprises: utilize to send control logic transmit queue and forwarding queue arbitrated, based on the age policy selection the packet in old age send.
Preferably, the hybrid network communication module also comprises the credit management logic that is connected to the transmission control logic; And preferably, described cluster computing system mixed communication method based on the flush bonding processor memory interface also comprises: utilize the credit management logic to carry out credit management and maintenance, so that Internet data adopts Credit-based flow-control mechanism.
Preferably, the queue depth of transmit queue, forwarding queue and receiving queue is configurable.
Preferably, transmit queue, forwarding queue and receiving queue all have the full signal of a sky, are used for carrying out Flow Control.
The above embodiment of the present invention provides the hybrid network communication module of a special use.Expand the network communication interface of flush bonding processor by this module, be used for connecting the second cover network, overcome the single shortcoming of flush bonding processor network communication interface.Simultaneously, inside modules has realized packet storage forwarding function, can support more flexibly multihop network of structural texture.And, the above embodiment of the present invention is expanded based on the parallel memory interface of flush bonding processor, take full advantage of the fast characteristics of memory interface transmission rate, make the hybrid network communication module that the interference networks interface higher than Ethernet interface transmission rate can be provided, thereby solved the weak problem of flush bonding processor network interface performance.In addition, memory interface is the indispensable interface of flush bonding processor, expands based on this interface, has certain universality.Thus, the shortcoming of the invention provides and a kind ofly can solve that the flush bonding processor network interface is single, performance is weak, by realizing a hybrid network communication module based on parallel memory interface, for the cluster computing system of building based on flush bonding processor provides more at a high speed network interface and hybrid network group-net communication method more flexibly.
Description of drawings
By reference to the accompanying drawings, and by with reference to following detailed description, will more easily to the present invention more complete understanding be arranged and more easily understand its advantage of following and feature, wherein:
Fig. 1 schematically shows according to the meshed network interface of the embodiment of the invention and the configuration schematic diagram of data path.
Fig. 2 schematically shows the structure according to the hybrid network communication module of the embodiment of the invention.
Fig. 3 schematically shows the cluster computing system based on hybrid network according to the embodiment of the invention.
Fig. 4 schematically shows the computing node network data path based on the OMAP3530 flush bonding processor according to the embodiment of the invention.
Need to prove, accompanying drawing is used for explanation the present invention, and unrestricted the present invention.Note, the accompanying drawing of expression structure may not be to draw in proportion.And in the accompanying drawing, identical or similar element indicates identical or similar label.
Embodiment
In order to make content of the present invention more clear and understandable, below in conjunction with specific embodiments and the drawings content of the present invention is described in detail.
In the present invention, on the one hand, the network controller that flush bonding processor is integrated is connected to first network by network transceivers; On the other hand, the parallel memory interface that flush bonding processor is integrated is connected to second network by the hybrid network communication module.Like this, the present invention has expanded the network communication interface of flush bonding processor by this hybrid network communication module, is used for connecting the second cover network, has overcome the single shortcoming of flush bonding processor network communication interface.
The below will describe the preferred embodiments of the present invention, thus the example of the specific implementation structure of explanation hybrid network communication module N2 and the operation example of hybrid network communication module N2.
Fig. 1 schematically shows according to the meshed network interface of the embodiment of the invention and the configuration schematic diagram of data path.
As shown in Figure 1, flush bonding processor 100 integrated first processor nuclear M1, the second processor core M2, ethernet controller M3 and the parallel memory interface M4 of comprising.
Ethernet controller M3 is connected to the Ethernet W1 of standard by connecting outside ethernet transceiver N1.By expanding parallel memory interface M4, realize the hybrid network communication module N2 of an outside, and then be connected to express network W2 between node.Data D1 can transmit between two cover networks through the on-chip bus B of flush bonding processor 100.
The focusing on of the embodiment of the invention designed a hybrid network communication module N2 based on parallel memory interface.Figure 2 shows that the structure of hybrid network communication module N2, as shown in Figure 2, hybrid network communication module N2 mainly comprises: parallel interface state machine N11, transmit queue N13, forwarding queue N14, receiving queue N15, transmission control logic N16, credit management logic N17, reception control logic N18 and multiplexer N12.
Wherein, parallel interface state machine N11 is connected to the data input pin of transmit queue N13 and the data output end of receiving queue N15; The data output end of transmit queue N13 is connected to the first data input pin of multiplexer N12; The data input pin of receiving queue N15 is connected to and receives control logic N18; The data output end that sends control logic N16 is connected to the second data input pin of multiplexer N12, and the data input pin that sends control logic N16 is connected to and receives control logic N18; The output that sends control logic N16 is connected to the control signal input of multiplexer N12; And send control logic N16 and also be connected to credit management logic N17; In addition, the data output end of multiplexer N12 is connected to second network (being express network W2 between node in the present embodiment).
For example, parallel interface state machine N11 is a standard agreement state machine that meets parallel memory interface protocol, links to each other with the parallel memory interface M4 of processor N12 with 16 parallel-by-bit memory buses.This state machine is pressed memory interface protocol, the control data transmit-receive take 16 word lengths as unit.
When parallel interface state machine N11 receives data from the memory bus that is connected with parallel memory interface M4, deposit data is arrived transmit queue N13.When receiving queue N15 non-NULL, and memory bus sends to data on the memory bus when vacant.
Transmit queue N13, forwarding queue N14 and receiving queue N15 are used for the data of buffer memory transmission course.For example, each memory cell is 16 bit wides in these formations; In addition, for example, the queue depth of transmit queue N13, forwarding queue N14 and receiving queue is configurable, and Typical Disposition comprises 1024 memory cell for each formation.
Transmit queue N13, forwarding queue N14 and receiving queue N15 all have the full signal of a sky, are used for carrying out Flow Control.
And transmit queue N13 is used for the Internet data for the treatment of that buffer memory comes self processor (for example flush bonding processor 100); Receiving queue N15 buffer memory waits to be forwarded to the data of parallel memory interface M4, and more particularly, receiving queue N15 buffer memory is off line and wait to be forwarded to the data of the processor (for example flush bonding processor 100) under the parallel memory interface M4; Forwarding queue N14 buffer memory waits to be forwarded to the data of the hybrid network communication module of other node.
Send control logic N16(or, send control logic N16 and credit management logic N17 common) be responsible for control data sent online.In the situation that have credit management logic N17, Internet data adopts Credit-based flow-control mechanism, carries out credit management and maintenance by the credit management logic.
Send control logic N16 transmit queue N13 and forwarding queue N14 arbitrated, based on the age policy selection the packet in old age send.
Receive control logic N18 and receive packet and disassemble packet from network, then according to the destination address of packet, select packet is stored in receiving queue N15 or forwarding queue N14.
Therefore, according to Fig. 2 structure, form three data paths:
(1) data path 1: from processor receive data bag, storage forwarding is to network (second network);
(2) data path 2: from network (second network) receive data bag (Internet data), storage forwarding is to processor;
(3) data path 3: from network (second network) receive data bag (lower network data), storage forwarding is to network (second network).
Thus, the above embodiment of the present invention provides the hybrid network communication module of a special use.Expand the network communication interface of flush bonding processor by this module, be used for connecting the second cover network, overcome the single shortcoming of flush bonding processor network communication interface.Simultaneously, inside modules has realized packet storage forwarding function, can support more flexibly multihop network of structural texture.
And, the above embodiment of the present invention is expanded based on the parallel memory interface of flush bonding processor, take full advantage of the fast characteristics of memory interface transmission rate, make the hybrid network communication module that the interference networks interface higher than Ethernet interface transmission rate can be provided, thereby solved the weak problem of flush bonding processor network interface performance.In addition, memory interface is the indispensable interface of flush bonding processor, expands based on this interface, has certain universality.
Concrete application of the present invention is described below.
Fig. 3 schematically shows the cluster computing system based on hybrid network according to the embodiment of the invention.As shown in Figure 3, Fig. 3 has described a clustered computing system based on hybrid network, and this system adopts the described communication means of the above embodiment of the present invention, has realized the transmission of hybrid network and inter-network data.
Specifically, this group system comprises cover 100,000,000-gigabit-Wan million multilayer switched ethernets.Each node accesses this network by the 100 m ethernet interface.Wherein, per 8 nodes are a node group, have set up high speed looped network in the group.The node group is connected to a cover gigabit-Wan million switches by a cover 100,000,000-gigabit switch, and a plurality of gigabit-Wan million switches finally are connected to a cover 10,000,000,000 switches.
For example, the network data path of computing node can form as shown in Figure 4.Computing node adopts the OMAP3530 processor of TI company, connects LAN9221 100 m ethernet transceiver by the MII interface, is connected to Ethernet W11.By the GPMC interface, connect hybrid network communication module N2, be connected to high speed looped network W22.GPMC is a kind of general memory interface for flush bonding processor, and this interface can be supported the highest 1.3Gbps transmission rate.
The hybrid network communication module adopts fpga chip to realize, with 32 network interfaces that the GPIO interface configuration is high speed looped network W22 on the FPGA, can support at full speed bidirectional data communication.
Be understandable that, although the present invention with the preferred embodiment disclosure as above, yet above-described embodiment is not to limit the present invention.For any those of ordinary skill in the art, do not breaking away from the technical solution of the present invention scope situation, all can utilize the technology contents of above-mentioned announcement that technical solution of the present invention is made many possible changes and modification, or be revised as the equivalent embodiment of equivalent variations.Therefore, every content that does not break away from technical solution of the present invention according to any simple modification, equivalent variations and the modification that technical spirit of the present invention is done above embodiment, all still belongs in the scope of technical solution of the present invention protection.
Claims (10)
1. cluster computing system mixed communication method based on the flush bonding processor memory interface, it is characterized in that comprising: the network controller that flush bonding processor is integrated is connected to first network by network transceivers; And the parallel memory interface that flush bonding processor is integrated is connected to second network by the hybrid network communication module.
2. the cluster computing system mixed communication method based on the flush bonding processor memory interface according to claim 1 is characterized in that, first network is Ethernet, and network transceivers is ethernet transceiver, and network controller is ethernet controller.
3. the cluster computing system mixed communication method based on the flush bonding processor memory interface according to claim 1 and 2, it is characterized in that, the hybrid network communication module comprises: parallel interface state machine, transmit queue, forwarding queue, receiving queue, transmission control logic, reception control logic and multiplexer; Wherein, the parallel interface state machine is connected to the data input pin of transmit queue and the data output end of receiving queue; The data output end of transmit queue is connected to the first data input pin of multiplexer; The data input pin of receiving queue is connected to the reception control logic; The data output end that sends control logic is connected to the second data input pin of multiplexer, and the data input pin that sends control logic is connected to the reception control logic; The output that sends control logic is connected to the control signal input of multiplexer; And the data output end of multiplexer is connected to second network.
4. the cluster computing system mixed communication method based on the flush bonding processor memory interface according to claim 3, characterized by further comprising: when the parallel interface state machine receives data from the memory bus that is connected with parallel memory interface, deposit data is arrived transmit queue.
5. according to claim 3 or 4 described cluster computing system mixed communication methods based on the flush bonding processor memory interface, characterized by further comprising: utilize the reception control logic to receive packet and disassemble packet from second network, then according to the destination address of packet, selection is stored in receiving queue or forwarding queue with packet, and wherein the receiving queue buffer memory waits to be forwarded to the data of parallel memory interface.
6. according to claim 3 to one of 5 described cluster computing system mixed communication methods based on the flush bonding processor memory interface, characterized by further comprising: when the receiving queue non-NULL, and when the memory bus that is connected with parallel memory interface is vacant, data are sent on the memory bus.
7. according to claim 3 to one of 6 described cluster computing system mixed communication methods based on the flush bonding processor memory interface, characterized by further comprising: utilize to send control logic transmit queue and forwarding queue arbitrated, based on the age policy selection the packet in old age send.
8. according to claim 3 to one of 7 described cluster computing system mixed communication methods based on the flush bonding processor memory interface, it is characterized in that, the hybrid network communication module also comprises and is connected to the credit management logic that sends control logic; And described cluster computing system mixed communication method also comprises: utilize the credit management logic to carry out credit management and maintenance, so that Internet data adopts Credit-based flow-control mechanism.
9. according to claim 3 to one of 8 described cluster computing system mixed communication methods based on the flush bonding processor memory interface, it is characterized in that, the queue depth of transmit queue, forwarding queue and receiving queue is configurable.
10. according to claim 3 to one of 9 described cluster computing system mixed communication methods based on the flush bonding processor memory interface, it is characterized in that, transmit queue, forwarding queue and receiving queue all have the full signal of a sky, are used for carrying out Flow Control.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210372418.6A CN102904943B (en) | 2012-09-28 | 2012-09-28 | Cluster computing system hybrid communication method based on embedded processor memory interface |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210372418.6A CN102904943B (en) | 2012-09-28 | 2012-09-28 | Cluster computing system hybrid communication method based on embedded processor memory interface |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102904943A true CN102904943A (en) | 2013-01-30 |
| CN102904943B CN102904943B (en) | 2015-07-08 |
Family
ID=47576978
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210372418.6A Active CN102904943B (en) | 2012-09-28 | 2012-09-28 | Cluster computing system hybrid communication method based on embedded processor memory interface |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102904943B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107248944A (en) * | 2017-07-28 | 2017-10-13 | 杭州聪普智能科技有限公司 | A kind of performance optimization system of intelligent domestic system controlling bus |
| CN115809685A (en) * | 2023-02-09 | 2023-03-17 | 鹏城实验室 | NPU cluster network structure and network interconnection method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1529472A (en) * | 2003-10-13 | 2004-09-15 | 中国科学院计算技术研究所 | High-performance computer group internet adaptor |
| CN2659038Y (en) * | 2003-12-09 | 2004-11-24 | 海信集团有限公司 | Inserted network interface circuit |
| CN1933438A (en) * | 2006-10-17 | 2007-03-21 | 杭州华为三康技术有限公司 | System and parts for realizing communication between first part and secondary part |
| CN101227459A (en) * | 2007-01-16 | 2008-07-23 | 窦晓鸣 | Method and apparatus for interactive processing multimedia information |
-
2012
- 2012-09-28 CN CN201210372418.6A patent/CN102904943B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1529472A (en) * | 2003-10-13 | 2004-09-15 | 中国科学院计算技术研究所 | High-performance computer group internet adaptor |
| CN2659038Y (en) * | 2003-12-09 | 2004-11-24 | 海信集团有限公司 | Inserted network interface circuit |
| CN1933438A (en) * | 2006-10-17 | 2007-03-21 | 杭州华为三康技术有限公司 | System and parts for realizing communication between first part and secondary part |
| CN101227459A (en) * | 2007-01-16 | 2008-07-23 | 窦晓鸣 | Method and apparatus for interactive processing multimedia information |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107248944A (en) * | 2017-07-28 | 2017-10-13 | 杭州聪普智能科技有限公司 | A kind of performance optimization system of intelligent domestic system controlling bus |
| CN115809685A (en) * | 2023-02-09 | 2023-03-17 | 鹏城实验室 | NPU cluster network structure and network interconnection method |
| CN115809685B (en) * | 2023-02-09 | 2023-07-25 | 鹏城实验室 | NPU cluster network structure and network interconnection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102904943B (en) | 2015-07-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111104775B (en) | Network-on-chip topological structure and implementation method thereof | |
| CN106603358B (en) | A kind of high-speed bus system and implementation method based on MLVDS interface | |
| US9158882B2 (en) | Automatic pipelining of NoC channels to meet timing and/or performance | |
| CN101753388A (en) | Routing and interface device suitable for on-chip and inter-chip extension of multi-core processor | |
| CN110837486B (en) | FlexRay-CPCIe communication system based on FPGA | |
| CN102760111A (en) | FPGA-based (Field Programmable Gate Array) extended multi-serial port device and data receiving-transmitting method thereof | |
| CN204650513U (en) | Distributed structure/architecture equipment and serial port circuit thereof | |
| CN102866980B (en) | Network communication cell used for multi-core microprocessor on-chip interconnected network | |
| CN102035723A (en) | On-chip network router and realization method | |
| CN100421423C (en) | Central router based on serial Rapid 10 bus | |
| CN109582622A (en) | Dual link interconnection architecture based on three-dimensional Mesh network-on-chip | |
| CN102904943B (en) | Cluster computing system hybrid communication method based on embedded processor memory interface | |
| CN101764795A (en) | Link layer controller of IEEE1394 bus | |
| CN102004713B (en) | Method for converting LINK ports of Tiger SHARC digital signal processor (DSP) to a serial Rapid IO bus | |
| CN202309716U (en) | Optical fiber-based high-speed real-time communication card | |
| CN101702714B (en) | Method, system, and apparatus for a credit based flow control in a computer system | |
| CN206195819U (en) | Spatial information network link controlgear | |
| CN201355815Y (en) | IEEE1394 bus protocol controller applied in space technology | |
| CN110995604B (en) | SpaceWire router level connection structure for expanding SpaceWire port | |
| CN108449287A (en) | FC exchanger chip architectural frameworks | |
| CN103873330A (en) | RS422-CAN bus converter | |
| CN103744817B (en) | For Avalon bus to the communication Bridge equipment of Crossbar bus and communication conversion method thereof | |
| CN204145526U (en) | With the CAN repeater of dual master control module | |
| CN103957145B (en) | CAN repeater and its data relay method with dual master control module | |
| CN103064360A (en) | Data transmission long-range control system based on duel-port random-access memory (RAM) |
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 |