CN106445068A - Fuzzy control energy-saving method based on GPU (Graphics Processing Unit) data packet classification - Google Patents
Fuzzy control energy-saving method based on GPU (Graphics Processing Unit) data packet classification Download PDFInfo
- Publication number
- CN106445068A CN106445068A CN201610785680.1A CN201610785680A CN106445068A CN 106445068 A CN106445068 A CN 106445068A CN 201610785680 A CN201610785680 A CN 201610785680A CN 106445068 A CN106445068 A CN 106445068A
- Authority
- CN
- China
- Prior art keywords
- gpu
- packet classification
- energy
- fuzzy control
- fuzzy
- 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
- G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
- G06F1/26—Power supply means, e.g. regulation thereof
- G06F1/32—Means for saving power
- G06F1/3203—Power management, i.e. event-based initiation of a power-saving mode
- G06F1/3234—Power saving characterised by the action undertaken
- G06F1/329—Power saving characterised by the action undertaken by task scheduling
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a fuzzy control energy-saving method based on GPU (Graphics Processing Unit) data packet classification. The method comprises the steps of designing a fuzzy control energy-saving model based on a GPU and defining a fuzzy rule; setting GPU initial operation parameters for a data packet classification task; monitoring operation power consumption and a throughput performance index state of the GPU; and continuously carrying out fuzzy adjustment control by employing the fuzzy control energy-saving model based on the GPU, thereby achieving an energy-saving effect. The invention provides the fuzzy control method which is designed according to a variable frequency air conditioner energy-saving principle. When the GPU operates the task, the balance problem between high performance and low energy consumption can be taken into consideration, and the method has important practical application value. The high-performance and low-energy-consumption fuzzy control model for carrying out the packet classification based on the GPU is provided in the field of data packet classification, and the method has important application value for a network device to realize green calculation.
Description
Technical field
The present invention relates to the energy-saving field of GPU application, particularly a kind of fuzzy control energy-conservation based on GPU packet classification
Method.
Background technology
Network packet classification refers to that router forwards packet to corresponding port.At present, most widely used bag classification
Technology is based on TCAM (Ternary Content Addressable Memory) hardware platform, although this platform performance is high,
But there is high power consumption, high price.In recent years, based on FPGA (Field-Programmable Gate Array),
The bag of other hardware platform such as NPU (Network Processing Unit), GPU (Graphics Processing Unit)
Sorting technique is also constantly developed.Wherein, the bag sorting technique based on GPU is good because of programmability, extensibility, versatility
By force, product category and price are various and have preferable scientific research and actual application value.However, similarly there is work(in GPU platform
Consume higher problem, therefore study and important using value is had based on the power-economizing method of GPU packet classification.
In supercomputer field, speed ability is a most important evaluation index, and TOP500 is then to list to work as previous existence
Calculating speed front 500 supercomputers the fastest in boundary.But it has been found that with the continuous lifting calculating performance, power consumption
Problem also constantly projects, thus leading to energy consumption wall problem.So also specially with the calculating performance of unit work consumptiom as index, listing generation
The Green500 supercomputer list of energy-conservation in boundary.In in June, 2015 TOP500 top ten list, only two employ
The supercomputer of Tesla K20GPU enters front ten, but has 6 supercomputers in the top ten list of same period Green500
Employ Tesla K20/40/80 series GPU.
As can be seen that in high-performance computing sector, energy consumption is an important evaluation index.Network packet is classified
It is a kind of task of execution high-performance calculation, we are also using Tesla K20 hardware platform based on the packet classification of GPU,
So research high-performance has important actual application value with the balance of low energy consumption.
Content of the invention
The technical problem to be solved is, not enough for prior art, provides one kind to be based on GPU packet classification
Fuzzy control power-economizing method.
For solving above-mentioned technical problem, the technical solution adopted in the present invention is：A kind of mould based on GPU packet classification
Paste controls power-economizing method, comprises the following steps：
1) the fuzzy control energy saving model based on GPU for the design, ambiguity in definition rule；
2) the GPU initial operational parameter of packet classification task is set, and including distribution calculating task, packet classification is calculated
Method, GPU thread scheduling policies, GPU core frequency；
3) monitoring GPU runs power consumption, throughput performance indications state；
4) use GPU energy-saving fuzzy Controlling model, constantly carry out fuzzy adjustment control, to reach energy-saving effect.
Step 1) the process that implements include：
1) measure GPU in four kinds of computation schema Default, Exclusive Thread, Exclusive Process,
Operation power P under Prohibited；
2) measure power under several running frequencies F that this model is supported for the GPU；
3) electricity energy when measurement is run using the GPU under multiple simultaneous resource Grid, Block, Thread allocation strategy A
Consumption；
4) measurement algorithms of different run time O, the GPU electricity energy consumption under calculating task amount N；
5) running temperature that throughput performance S, the GPU self-sensor device of measurement GPU processing data per second bag quantity obtains
T；
6) find out throughput performance S of various indexs and processing data bag quantity per second in power consumption P, network packet classification
Corresponding relation, design fuzzy rule.
According to following formula ambiguity in definition rule：
Wherein, N represents calculating task amount, and A represents the thread allocation strategy of GPU simultaneous resource Grid, Block, Thread,
O represents the run time using different packet classifications；
Define a series of IF-THEN fuzzy rule, the rule of definition is as follows：
If GPU is idle condition, then setting GPU is disabling pattern；
If GPU is running status, then setting GPU is default mode；
If GPU power is too high, then reduce GPU running frequency；The too high threshold interval of power is 65～70W；
If GPU power is higher, then reduce GPU simultaneous resource Grid, Block, Thread quantity；Power higher thresholds
Interval is 60～67W；
If throughput is too low, then improve GPU running frequency；It is 48～50Mpps that throughput crosses Low threshold；
If throughput is relatively low, then improve GPU simultaneous resource Grid, Block, Thread quantity；The relatively low threshold of throughput
It is worth for 49～55Mpps.
Compared with prior art, the present invention had the advantage that for：The invention provides one kind is according to convertible frequency air-conditioner
The fuzzy control method that energy-saving principle designs, so that GPU is in operation task, can take into account high-performance and low energy consumption simultaneously
Equilibrium problem, there is important actual application value；In packet classification field, there is provided one kind carries out bag point based on GPU
The high-performance of class, low energy consumption fuzzy control model, realize green calculating and have important practical value to the network equipment.
Brief description
Fig. 1 is the energy-saving fuzzy control flow chart based on the classification of GPU bag；
Fig. 2 is that the router based on GPU forwards packet schematic diagram；
Fig. 3 is fuzzy control principle figure；
Fig. 4 is the power diagram of GPU under different conditions；
Fig. 5 is GPU performance and energy consumption figure under different running frequencies；
Fig. 6 is GPU performance and energy consumption figure under different threads allocation strategy.
Specific embodiment
With reference to Fig. 1, the present invention comprises the following steps：
Step 1, starts based on the packet classification task of GPU, and initializes operational factor.The computation schema setting of GPU
For default mode；For Tesla K20GPU, initial launch set of frequency is system default value.
Step 2, is adjusted to GPU operational factor using fuzzy rule.Wherein, fuzzy rule is verified according to from experiment
And the derivation of equation drawing.
Step 3, carries out bag classification using GPU platform and accelerates.
Step 4, every operational factor of monitoring GPU.In accelerator, to GPU running state parameter, power, temperature
Etc. carrying out monitor in real time record.
Step 5, whether statistic mixed-state achieves the optimization of high-performance and low-power consumption.Monitoring parameter is counted, compares
Systematic function and the degree of optimization of power consumption.
Step 6, without reaching target, proceeds to step 2, proceeds fuzzy control adjustment.Similar to convertible frequency air-conditioner
The principle of energy-conservation, can be according to the loading level of work at present task, and the mode using fuzzy control actively changes running frequency,
If low-load, then carry out energy-conservation using relatively low running frequency.EFCGPU model proposed by the present invention is also had using GPU
The characteristic of frequency conversion function, dynamically adjusts GPU operational factor according to calculating task, for the target of high-performance and low-power consumption, carries out
One process constantly balancing, constantly adjusting, thus reach the energy-saving effect of optimization.
With reference to Fig. 2, it is the bag classification routing forwarding schematic diagram based on GPU, explains the technical spirit of bag classification.According to reaching
To the data packet header field feature forwarding engine, carry out rule match, finding next step needs the forwarding behavior of execution, thus
Classification transmission is carried out to packet.
With reference to Fig. 3, it is the schematic diagram of fuzzy control, represent that the fuzzy control of system is one and constantly adjusts, constantly balances
Process.The state variable of input system first, then carries out fuzzy rule coupling, according to fuzzy inference result, obtains non-mould
The control variables of paste, is adjusted to system.
With reference to Fig. 4, it is power diagram under different mode state for the GPU, the Tesla K20GPU that for example present invention uses exists
Idle period, power is 27W；Operationally section, power is 65W；But GPU is set to disabling pattern, its work(
Rate only 15W about, this just can carry out energy-conservation in the inoperative period of GPU by Adjustable calculation pattern.
With reference to Fig. 5, it is performance under different running frequencies for the GPU and energy consumption figure.The Tesla K20GPU that the present invention uses,
Support 6 kinds of different running frequencies.Different running frequencies is compared using identical calculating task, it can be found that GPU exists
Under the running frequency of 705MHz, reach an optimum level that energy consumption is minimum, throughput performance is also higher.
With reference to Fig. 6, it is performance under different threads allocation strategy for the GPU and energy consumption figure, it can be found that when thread distributes plan
During slightly 8blocks-256threads, reach the results of a high-performance and the double optimization of low-power consumption, compare other allocation strategies,
Highest can save 15.51% energy consumption.
The invention provides a kind of fuzzy control power-economizing method based on GPU packet classification, the frequency conversion being had using GPU
Characteristic, in conjunction with the fuzzy control energy-saving principle of convertible frequency air-conditioner, is a kind of new thinking, examples of implementation also demonstrate that having of the method
Effect property and feasibility.
Claims (4)
1. a kind of fuzzy control power-economizing method based on GPU packet classification is it is characterised in that comprise the following steps：
1) the fuzzy control energy saving model based on GPU for the design, ambiguity in definition rule；
2) the GPU initial operational parameter of packet classification task is set, including distribution calculating task, packet classification algorithm, GPU
Thread scheduling policies, GPU core frequency；
3) monitoring GPU runs power consumption, throughput performance indications state；
4) use GPU energy-saving fuzzy Controlling model, constantly carry out fuzzy adjustment control, to reach energy-saving effect.
2. the fuzzy control power-economizing method based on GPU packet classification according to claim 1 is it is characterised in that step
1) the process that implements includes：
1) measure GPU in four kinds of computation schema Default, Exclusive Thread, Exclusive Process,
Operation power P under Prohibited；
2) measure power under several running frequencies F that this model is supported for the GPU；
3) electricity energy consumption when measurement is run using the GPU under multiple simultaneous resource Grid, Block, Thread allocation strategy A；
4) measurement algorithms of different run time O, the GPU electricity energy consumption under calculating task amount N；
5) running temperature T that throughput performance S, the GPU self-sensor device of measurement GPU processing data per second bag quantity obtains；
6) various indexs and power consumption P are found out, throughput performance S of processing data bag quantity per second is right in network packet classification
Should be related to, design fuzzy rule.
3. the fuzzy control power-economizing method based on GPU packet classification according to claim 2 it is characterised in that according to
Following formula ambiguity in definition rule：
Wherein, N represents calculating task amount, and A represents the thread allocation strategy of GPU simultaneous resource Grid, Block, Thread, O generation
Table is using the run time of different packet classifications.
4. the fuzzy control power-economizing method based on GPU packet classification according to claim 3 is it is characterised in that define
A series of IF-THEN fuzzy rule, the rule of definition is as follows：
If GPU is idle condition, then setting GPU is disabling pattern；
If GPU is running status, then setting GPU is default mode；
If GPU power is too high, then reduce GPU running frequency；The too high threshold interval of power is 65～70W；
If GPU power is higher, then reduce GPU simultaneous resource Grid, Block, Thread quantity；
Power higher thresholds are interval to be 60～67W；
If throughput is too low, then improve GPU running frequency；It is 48～50Mpps that throughput crosses Low threshold；
If throughput is relatively low, then improve GPU simultaneous resource Grid, Block, Thread quantity；Throughput lower threshold is
49～55Mpps.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610785680.1A CN106445068A (en) | 2016-08-31 | 2016-08-31 | Fuzzy control energy-saving method based on GPU (Graphics Processing Unit) data packet classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610785680.1A CN106445068A (en) | 2016-08-31 | 2016-08-31 | Fuzzy control energy-saving method based on GPU (Graphics Processing Unit) data packet classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106445068A true CN106445068A (en) | 2017-02-22 |
Family
ID=58090779
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610785680.1A Pending CN106445068A (en) | 2016-08-31 | 2016-08-31 | Fuzzy control energy-saving method based on GPU (Graphics Processing Unit) data packet classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106445068A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108733195A (en) * | 2018-05-29 | 2018-11-02 | 郑州易通众联电子科技有限公司 | Computer operation method and device based on equipment operational energy efficiency |
| CN110928587A (en) * | 2019-11-14 | 2020-03-27 | 联想(北京)有限公司 | Control method and control device |
| CN117608389A (en) * | 2023-12-13 | 2024-02-27 | 摩尔线程智能科技(北京)有限责任公司 | GPU power consumption control method, device and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103401635A (en) * | 2013-06-28 | 2013-11-20 | 国家超级计算深圳中心(深圳云计算中心) | Method and device for analyzing throughout collapse behavior in cluster storage system |
| US8700925B2 (en) * | 2009-09-01 | 2014-04-15 | Nvidia Corporation | Regulating power using a fuzzy logic control system |
| CN103986622A (en) * | 2014-05-27 | 2014-08-13 | 重庆邮电大学 | Network throughput rate parallelized measuring method based on multi-core technology |
| CN105897587A (en) * | 2016-03-31 | 2016-08-24 | 湖南大学 | Method for classifying data packets |
-
2016
- 2016-08-31 CN CN201610785680.1A patent/CN106445068A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8700925B2 (en) * | 2009-09-01 | 2014-04-15 | Nvidia Corporation | Regulating power using a fuzzy logic control system |
| CN103401635A (en) * | 2013-06-28 | 2013-11-20 | 国家超级计算深圳中心(深圳云计算中心) | Method and device for analyzing throughout collapse behavior in cluster storage system |
| CN103986622A (en) * | 2014-05-27 | 2014-08-13 | 重庆邮电大学 | Network throughput rate parallelized measuring method based on multi-core technology |
| CN105897587A (en) * | 2016-03-31 | 2016-08-24 | 湖南大学 | Method for classifying data packets |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108733195A (en) * | 2018-05-29 | 2018-11-02 | 郑州易通众联电子科技有限公司 | Computer operation method and device based on equipment operational energy efficiency |
| CN110928587A (en) * | 2019-11-14 | 2020-03-27 | 联想(北京)有限公司 | Control method and control device |
| CN117608389A (en) * | 2023-12-13 | 2024-02-27 | 摩尔线程智能科技(北京)有限责任公司 | GPU power consumption control method, device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10274970B2 (en) | Method, and device for controlling the output of the air volume and memory medium | |
| US6513124B1 (en) | Method and apparatus for controlling operating speed of processor in computer | |
| CN106445068A (en) | Fuzzy control energy-saving method based on GPU (Graphics Processing Unit) data packet classification | |
| US20200159297A1 (en) | Hybrid cooling control of a computing system | |
| TWI516886B (en) | Intelligent learning energy-saving control system and method thereof | |
| KR101842016B1 (en) | Method for dynamically controlling power in multicore environment | |
| US9274585B2 (en) | Combined dynamic and static power and performance optimization on data centers | |
| ATE400006T1 (en) | CASCADED CONTROL OF AN AVERAGE VALUE OF A PROCESS PARAMETER TO A DESIRED VALUE | |
| CN104457069B (en) | Capacity regulating method for refrigerating system | |
| CN105305468B (en) | Thermal power generation unit primary frequency modulation parameter optimization method based on particle cluster algorithm | |
| CN109062668A (en) | A kind of virtual network function moving method of the multipriority based on 5G access network | |
| Liao et al. | Energy consumption optimization scheme of cloud data center based on SDN | |
| Masrom et al. | Hybridization of particle swarm optimization with adaptive genetic algorithm operators | |
| CN113434286B (en) | Energy efficiency optimization method suitable for mobile application processor | |
| CN110107391A (en) | Progress control method, system and electronic equipment after a kind of engine blower | |
| WO2021253706A1 (en) | Intermittent characteristic-based demand-side resource coordination control method and system | |
| CN105354054A (en) | Electronic product and adjusting method for performance parameter thereof | |
| CN104991443A (en) | Fuzzy control method based on adaptive domain partitioning | |
| Zhao et al. | Energy-efficient Edge Association in Digital Twin empowered 6G Networks | |
| CN106482281B (en) | It is a kind of for the control device of water cooler, control method and water cooler | |
| Reddy et al. | A Modified clustering for LEACH algorithm in WSN | |
| CN112887221B (en) | Periodic energy-saving method based on generalized predictive control | |
| CN115163540A (en) | Fan rotating speed control method, device and equipment | |
| CN205353675U (en) | Cloud calculates center computer room energy -saving control system | |
| CN114531665A (en) | Wireless sensor network node clustering method and system based on Laiwei flight |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170222 |