CN107193657A - Low latency server based on SOLAFLARE network interface cards - Google Patents
Low latency server based on SOLAFLARE network interface cards Download PDFInfo
- Publication number
- CN107193657A CN107193657A CN201710351354.4A CN201710351354A CN107193657A CN 107193657 A CN107193657 A CN 107193657A CN 201710351354 A CN201710351354 A CN 201710351354A CN 107193657 A CN107193657 A CN 107193657A
- Authority
- CN
- China
- Prior art keywords
- network interface
- solaflare
- interface cards
- low latency
- server
- 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
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/5044—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering hardware capabilities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L12/00—Data switching networks
- H04L12/02—Details
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Multi Processors (AREA)
Abstract
The invention discloses a kind of low latency server based on SOLAFLARE network interface cards, in the server using SOLAFLARE network interface cards, and on-site programmable gate array FPGA is put into SOLAFLARE network interface cards, realizes load balancing.The present invention solves the problems, such as the parallel conflict of order protocol and CPU and Cache temperatures, maximum delay and average retardation can be reduced to greatest extent, user can be met to greatest extent to be actually needed, obtain up-to-date information, can be more fast and efficiently that company and numerous netizens service.
Description
Technical field
The present invention relates to server field, specifically a kind of low latency server based on SOLAFLARE network interface cards.
Background technology
The current network market is not only faced with the problem of marketing data amount increases substantially, while also being come by time-bands are performed
Pressure.In these environment, if it is possible to obtain up-to-date information and improve the speed taken action, it becomes possible to compare rival
Fast several nanoseconds, will imply that may create millions of RMB even more big trading volume each second more.Therefore clothes how to be reduced
The problem of time delay of business device is into people's urgent need to resolve instantly.
In order to reduce data delay, increasing service organization requires to shorten server and the department of computer science of oneself as far as possible
The physical distance of system, designs more complicated fine algorithm to reduce data delay, these modes can not be met to greatest extent
The demand of low latency.Because the network interface card of nowadays high capacity is the rich and influential family of traps, it is easy to form bottleneck, and high latency.And
In considerably long period, network interface card is interrupted and all handled by CPU0, causes CPU0 hypertonia, other CPU are relatively empty
It is not busy.The many queue technologies of network interface card had been occurred as soon as later, have realized that network card data request can be handled by multi -CPU.Other physical network card
Interruption is will also result in, high latency, RSS is formed（Aggregated content, is one kind letter that website is used for sharing content between other websites
Easy mode）It is the ardware feature of network interface card, if realizing many queues, different flow points is dealt on different CPU, same data flow
Through end on same CPU, it is possible to avoid TCP succession and CPU concurrency from occurring punching real.
Therefore, in order to pursue lightning transaction, in the market is solved to the demand of server low latency, is needed badly a scientific and effective
Low latency server help numerous netizens.
The content of the invention is it is an object of the invention to provide a kind of low latency server based on SOLAFLARE network interface cards, to solve
Prior art server network interface card interrupts the problem of easily forming high latency.
In order to achieve the above object, the technical solution adopted in the present invention is：
Low latency server based on SOLAFLARE network interface cards, it is characterised in that：SOLAFLARE network interface cards are used in the server, and
On-site programmable gate array FPGA is put into SOLAFLARE network interface cards, the interrupt distribution that each queue is produced is into server
CPU multiple cores, realize load balancing.
The described low latency server based on SOLAFLARE network interface cards, it is characterised in that：The interruption of each queue makes respectively
Use single interrupt line.
The described low latency server based on SOLAFLARE network interface cards, it is characterised in that：SOLAFLARE network interface cards pass through
Tcpip stack agreement bypasses LINUX kernels and directly invokes network, and from server loading algorithm and data, makes SOLAFLARE
Network interface card can directly access the database partly to realize the function of the webserver.
The described low latency server based on SOLAFLARE network interface cards, it is characterised in that：Adopted in SOLAFLARE network interface cards
The other precision of Microsecond grade is provided with agreement Precision Time Protocol.
The described low latency server based on SOLAFLARE network interface cards, it is characterised in that：The GUP cards used in server
Be designed to can many tasks of parallel processing, the serial section of program run on CPU, and parallel section is then transported on GPU
OK.
Server of the present invention carries brand-new engine --- Application Onload Engine SOLAFLARE network interface cards,
And different flow points is dealt on the different CPU core of server, load balancing is realized, order protocol and CPU is solved parallel
Conflict and Cache temperature problems, maximum delay and average retardation can be reduced to greatest extent, the actual need of user can be met to greatest extent
Will, up-to-date information is obtained, can be more fast and efficiently that company and numerous netizens service.CPU in server of the present invention
Processor most more top than equivalent specifications is also fast by more than 6 ~ 30%；Memory bandwidth is maximized, than conventional bandwidth lifting 30%；Network interface
40Gbps data bandwidths, postpone minimum to 1us microseconds；Hard disk IO, more than readwrite bandwidth 1.5GB/s, postpone minimum 64us microseconds,
IOPS values 700000；Motherboard circuit has high-performance designs, the protection of high frequency environment and high temperature.
Brief description of the drawings
Fig. 1 is appearance assumption diagram of the present invention.
Fig. 2 is hardware structure diagram of the present invention.
Fig. 3 is CPU card fundamental diagram of the present invention, wherein：
Fig. 3 a are serial operation schematic diagram, and Fig. 3 b are parallel operation schematic diagram.
Fig. 4 is SOLAFLARE network interface cards fundamental diagram of the present invention.
Fig. 5 is schematic diagram when the present invention is fabricated to pcb board.
Embodiment
Low latency server based on SOLAFLARE network interface cards, in the server using SOLAFLARE network interface cards, and by scene
Programmable gate array FPGA is put into SOLAFLARE network interface cards, the interrupt distribution that each queue is produced into server CPU it is many
Individual core, realizes load balancing.
The interruption of each queue uses single interrupt line respectively.
SOLAFLARE network interface cards bypass LINUX kernels by tcpip stack agreement and directly invoke network, and add from server
Algorithm and data are carried, make SOLAFLARE network interface cards to directly access the database partly to realize the function of the webserver.
In SOLAFLARE network interface cards the other precision of Microsecond grade is provided using agreement Precision Time Protocol.
The GUP cards used in server be designed to can many tasks of parallel processing, the serial section of program transports on CPU
OK, the parallel section then operation on GPU.
As shown in Figure 1, it is first determined function is to realize a low latency server, and outward appearance is a 777*437*43mm
（L*W*H）Cuboid, weight 21.8kg.
As shown in Fig. 2 server hardware, which is constituted, includes central processing unit, internal memory, chipset, I/O extensions, power supply and correlation
Software.Find supplier and determine hardware plan, the central processing unit of use is Intel Xeon Processor E5-
2687W V3 (GHz of 10Core/ 25M Cache/ 3.10) * 2, the memory size of use is 8*8GB 2133MHz DDR4
64G, the hard disk of use is Intel S3510 Series 480G*2, and the GUP cards of use are NIVIDIA GRID GPU K1, are adopted
Onboard network interface is 4x Gigabit Ethernet LAN ports and SolarFlare 7322f (containing module), is adopted
I/O extensions are 3.0 x16 1x PCI-E of 2x PCI-E 3.0 x8, Low-profile, the x8 of 1x PCI-E 3.0
SAS3 integrated, the remote management tool of use is the with KVM and of Integrated IPMI 2.0
Dedicated LAN, the power supply of use is 750W Redundant Power Supplies Platinum Level (94%
+)。
As shown in figure 3, be designed to can many tasks of parallel processing, the serial portion of program for the GUP cards used in server
Divide and run on CPU, and parallel section is then run on GPU.Consequently, it is possible to which program operation can be improved farthest
Efficiency, this is with regard to that can efficiently reduce the time of computing.
As shown in figure 4, programmable array（FPGA）Network adapter is put into, LINUX is bypassed by tcpip stack agreement
Kernel directly invokes network and uses special software loading algorithm and data from server, allows network interface card to realize（Realize part）Net
The function of network server, adds the ability directly accessed the database, quickly can import and export data from database, shortens
The disposed of in its entirety time.
As shown in figure 5, making pcb board, the drawing come out by PCB Software for Design takes the professional PCB factories of outside
Business goes production, such packing density height, electronic product small volume, lightweight.Ready-made PCB is taken back into assembling.Will
The device assembled test modifications repeatedly, while sending survey.
What is finally illustrated is：Above example is merely to illustrate technical scheme, rather than its limitations；Although ginseng
Invention is described in detail according to previous embodiment, those skilled in the art should be understood：It can still be implemented to foregoing
Technical scheme described in example is modified, or to which part technical characteristic equivalent substitution；And these modifications or replacement,
The essence of technical scheme is not set to depart from the spirit and scope of technical scheme of the embodiment of the present invention.
Claims (5)
1. the low latency server based on SOLAFLARE network interface cards, it is characterised in that：SOLAFLARE network interface cards are used in the server,
And on-site programmable gate array FPGA is put into SOLAFLARE network interface cards, the interrupt distribution that each queue is produced to server
Middle CPU multiple cores, realize load balancing.
2. the low latency server according to claim 1 based on SOLAFLARE network interface cards, it is characterised in that：Each queue
Interruption respectively use single interrupt line.
3. the low latency server according to claim 1 based on SOLAFLARE network interface cards, it is characterised in that：SOLAFLARE
Network interface card bypasses LINUX kernels by tcpip stack agreement and directly invokes network, and from server loading algorithm and data, makes
SOLAFLARE network interface cards can directly access the database partly to realize the function of the webserver.
4. the low latency server according to claim 1 based on SOLAFLARE network interface cards, it is characterised in that：
In SOLAFLARE network interface cards the other precision of Microsecond grade is provided using agreement Precision Time Protocol.
5. the low latency server according to claim 1 based on SOLAFLARE network interface cards, it is characterised in that：In server
The GUP cards of use be designed to can many tasks of parallel processing, the serial section of program is run on CPU, and parallel section is then
Run on GPU.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710351354.4A CN107193657A (en) | 2017-05-18 | 2017-05-18 | Low latency server based on SOLAFLARE network interface cards |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710351354.4A CN107193657A (en) | 2017-05-18 | 2017-05-18 | Low latency server based on SOLAFLARE network interface cards |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107193657A true CN107193657A (en) | 2017-09-22 |
Family
ID=59874158
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710351354.4A Pending CN107193657A (en) | 2017-05-18 | 2017-05-18 | Low latency server based on SOLAFLARE network interface cards |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107193657A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109284192A (en) * | 2018-09-29 | 2019-01-29 | 网宿科技股份有限公司 | Method for parameter configuration and electronic equipment |
| CN111614651A (en) * | 2020-05-14 | 2020-09-01 | 南京艾科朗克信息科技有限公司 | Low-delay TDS protocol analysis method based on FPGA |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101827029A (en) * | 2010-04-21 | 2010-09-08 | 烽火通信科技股份有限公司 | Method for dynamically marking RTP flow by Linux network equipment |
| CN201681397U (en) * | 2009-12-18 | 2010-12-22 | 燕山大学 | GPU-based accounting electronic document encrypting system under C/S framework and device |
| CN102710424A (en) * | 2012-05-30 | 2012-10-03 | 曙光信息产业(北京)有限公司 | Gigabit/10-gigabit multifunctional network card and implementation method for same |
| CN102904729A (en) * | 2012-10-26 | 2013-01-30 | 曙光信息产业(北京)有限公司 | Intelligent boost network card supporting multiple applications according to protocol and port shunt |
| CN103616934A (en) * | 2013-12-06 | 2014-03-05 | 江南大学 | FPGA core circuit board structure |
| CN104714850A (en) * | 2015-03-02 | 2015-06-17 | 心医国际数字医疗系统(大连)有限公司 | Heterogeneous joint account balance method based on OPENCL |
| CN105899268A (en) * | 2015-06-23 | 2016-08-24 | 中国科学院深圳先进技术研究院 | GPU-based parallel electrocardiosignal analyzing method |
| CN106101019A (en) * | 2016-06-22 | 2016-11-09 | 浪潮电子信息产业股份有限公司 | Interrupt binding-based multi-queue network card performance tuning method |
| US9525500B2 (en) * | 2011-06-13 | 2016-12-20 | Mediatek Inc. | Low-cost test/calibration system and calibrated device for low-cost test/calibration system |
| CN106598752A (en) * | 2016-11-15 | 2017-04-26 | 北京大学深圳研究生院 | Remote zero-copy method |
| CN110177087A (en) * | 2019-05-05 | 2019-08-27 | 方一信息科技（上海）有限公司 | A kind of end Target protocol hardware analytic method based on TOE network interface card |
-
2017
- 2017-05-18 CN CN201710351354.4A patent/CN107193657A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201681397U (en) * | 2009-12-18 | 2010-12-22 | 燕山大学 | GPU-based accounting electronic document encrypting system under C/S framework and device |
| CN101827029A (en) * | 2010-04-21 | 2010-09-08 | 烽火通信科技股份有限公司 | Method for dynamically marking RTP flow by Linux network equipment |
| US9525500B2 (en) * | 2011-06-13 | 2016-12-20 | Mediatek Inc. | Low-cost test/calibration system and calibrated device for low-cost test/calibration system |
| CN102710424A (en) * | 2012-05-30 | 2012-10-03 | 曙光信息产业(北京)有限公司 | Gigabit/10-gigabit multifunctional network card and implementation method for same |
| CN102904729A (en) * | 2012-10-26 | 2013-01-30 | 曙光信息产业(北京)有限公司 | Intelligent boost network card supporting multiple applications according to protocol and port shunt |
| CN103616934A (en) * | 2013-12-06 | 2014-03-05 | 江南大学 | FPGA core circuit board structure |
| CN104714850A (en) * | 2015-03-02 | 2015-06-17 | 心医国际数字医疗系统(大连)有限公司 | Heterogeneous joint account balance method based on OPENCL |
| CN105899268A (en) * | 2015-06-23 | 2016-08-24 | 中国科学院深圳先进技术研究院 | GPU-based parallel electrocardiosignal analyzing method |
| CN106101019A (en) * | 2016-06-22 | 2016-11-09 | 浪潮电子信息产业股份有限公司 | Interrupt binding-based multi-queue network card performance tuning method |
| CN106598752A (en) * | 2016-11-15 | 2017-04-26 | 北京大学深圳研究生院 | Remote zero-copy method |
| CN110177087A (en) * | 2019-05-05 | 2019-08-27 | 方一信息科技（上海）有限公司 | A kind of end Target protocol hardware analytic method based on TOE network interface card |
Non-Patent Citations (1)
| Title |
|---|
| 廖惠如: "10Gb以太网络新时代 专访Solarflare 营销副总裁Bruce Tolley", 《电子与电脑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109284192A (en) * | 2018-09-29 | 2019-01-29 | 网宿科技股份有限公司 | Method for parameter configuration and electronic equipment |
| CN111614651A (en) * | 2020-05-14 | 2020-09-01 | 南京艾科朗克信息科技有限公司 | Low-delay TDS protocol analysis method based on FPGA |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ajima et al. | The tofu interconnect d | |
| US10540584B2 (en) | Queue management for direct memory access | |
| Huang et al. | High-performance design of hbase with rdma over infiniband | |
| Agrawal et al. | Rhythm: Harnessing data parallel hardware for server workloads | |
| TW201933081A (en) | Bridge device and method for providing near storage compute | |
| Daglis et al. | Manycore network interfaces for in-memory rack-scale computing | |
| US20210075745A1 (en) | Methods and apparatus for improved polling efficiency in network interface fabrics | |
| Lockwood et al. | Implementing ultra low latency data center services with programmable logic | |
| EP3482294A1 (en) | Methods and modules relating to allocation of host machines | |
| CN107193657A (en) | Low latency server based on SOLAFLARE network interface cards | |
| CN107194811A (en) | A kind of high frequency transaction quantization system based on FPGA | |
| Qiu et al. | Full-kv: Flexible and ultra-low-latency in-memory key-value store system design on cpu-fpga | |
| Sanka et al. | Efficient high-performance FPGA-Redis hybrid NoSQL caching system for blockchain scalability | |
| Li et al. | Understanding hot interconnects with an extensive benchmark survey | |
| Ziegler et al. | Efa: A viable alternative to rdma over infiniband for dbmss? | |
| CN114138476A (en) | Processing method and device of pooled resources, electronic equipment and medium | |
| WO2016138657A1 (en) | Techniques for storing or accessing key-value item | |
| CN115687708B (en) | Transaction quotation data processing method and device and data processing board card | |
| Dang et al. | Network hardware-accelerated consensus | |
| Vaidyanathan et al. | Exploiting RDMA operations for Providing Efficient Fine-Grained Resource Monitoring in Cluster-based Servers | |
| Chung et al. | Fpga-based accelerator platform for big data matrix processing | |
| CN216927600U (en) | Network data computing system and server with built-in network data computing system | |
| Istvan et al. | Active pages 20 years later: Active storage for the cloud | |
| TW202008172A (en) | Memory system | |
| CN113079152B (en) | Data transmission method, device and medium |
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: 20170922 |