CN112463329A - Linux system network interface performance optimization method suitable for Loongson 3A - Google Patents
Linux system network interface performance optimization method suitable for Loongson 3A Download PDFInfo
- Publication number
- CN112463329A CN112463329A CN202011355830.8A CN202011355830A CN112463329A CN 112463329 A CN112463329 A CN 112463329A CN 202011355830 A CN202011355830 A CN 202011355830A CN 112463329 A CN112463329 A CN 112463329A
- Authority
- CN
- China
- Prior art keywords
- loongson
- interrupt
- processor
- linux system
- dpdk
- 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
Images
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
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4812—Task transfer initiation or dispatching by interrupt, e.g. masked
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F8/00—Arrangements for software engineering
- G06F8/70—Software maintenance or management
- G06F8/76—Adapting program code to run in a different environment; Porting
-
- 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
- G06F9/505—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 the load
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Debugging And Monitoring (AREA)
Abstract
The invention relates to a method for optimizing the performance of a Linux system network interface suitable for Loongson 3A, which is technically characterized by comprising the following steps: the invention sets an interrupt balance algorithm; running an interrupt balance algorithm, and carrying out interrupt transfer on a corresponding processor currently positioned on the MaxX when executing interrupt balance operation; implementing multi-core dynamic optimization service in Linux system to implement multi-core dynamic optimization service of interrupt processing; the MSI interruption mode is changed into a Legacy interruption mode; modifying the content of the atomic operation library in the DPDK EAL component; adjusting the library function; optimizing the transplanted DPDK EAL component to modify the DPDK EAL component and transplanting the DPDK EAL component to a Linux system of the Loongson 3A processor, thereby optimizing the performance of the network interface. The method fills the blank of related research in China, has universality, and can be used for improving the throughput rate, the packet loss rate and the time delay index of a network interface when all mainstream gigabit and tera network cards and Loongson 3A are suitable for a Linux system.
Description
Technical Field
The invention belongs to the field of network interface performance optimization, and particularly relates to a Linux system network interface performance optimization method suitable for Loongson 3A.
Background
The Loongson 3A series processor is a domestic multi-core general processor with proprietary intellectual property, which is jointly developed by a Chinese academy of sciences and a Loongson department company. There are four types of processors in volume production, 3a1000, 3a2000, 3a3000 and 3a4000, all of which use a 64-bit 4-core design. Because the Loongson uses the MIPS framework, the software ecology is not sound, the expanded network interface adopts open source drive of igb, ixgb, e1000 and the like under the Linux system, the open source software ecologically transplanted from x86 software is not optimized for the Loongson platform, the coupling with the Loongson hardware framework is low, and the running performance of the network interface on the Loongson platform is low.
The Loongson 3A supports at most 32 interrupt sources and manages in a unified mode, and all the interrupt sources can select a target processor core expected to be interrupted through software configuration. When the interface development is suitable, a user can optimize the interrupt processing according to the application requirements, interrupt resources are distributed reasonably, and the four-core performance of the Loongson 3A is exerted to the maximum extent.
By monitoring the processor occupation condition of a process during network performance test, the situation that 100% of processor occupation condition occurs in a single-core soft interrupt queue of a Loongson 3A processor for a long time when the throughput rate is high is found, and an interrupt imbalance phenomenon exists, while the same network card chip is tested under an x86 platform, although the situation that the single-core occupation rate approaches 100% along with the increase of the throughput rate also exists, other processor cores of an x86 processor effectively share interrupt load in the processing process, and no obvious interrupt imbalance phenomenon exists; therefore, the throughput rate performance of the network interface can be directly influenced when the single core in the system is in the 100% occupancy rate for a long time, and the network throughput rate can be improved to a certain extent by designing a reasonable interrupt balance algorithm.
DPDK is an open source network development kit developed by Intel corporation, is designed aiming at x86, Power and ARM general processors, can reduce unnecessary I/O (input/output) expenditure of the processors, better utilizes the operational capability of the processors, realizes high-performance network data packet processing, and is a network performance optimization tool which is widely applied and verified. How to change the bottom-layer function of the DPDK to adapt to the Loongson hardware architecture effectively utilizes the mature network optimization tool of the DPDK, and becomes a problem to be considered and solved by network technical researchers under a Loongson platform.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a Linux system network interface performance optimization method suitable for a Loongson 3A, can realize the optimization of the Linux system network interface of the Loongson 3A, and improves the throughput rate, the packet loss rate and the time delay index of the network interface.
The technical problem to be solved by the invention is realized by adopting the following technical scheme:
a performance optimization method for a Linux system network interface suitable for a Loongson 3A comprises the following steps:
step 1, constructing a multi-core dynamic optimization flow for interrupt processing of a Loongson 3A processor;
and 2, modifying the DPDK EAL component and transplanting the DPDK EAL component into a Linux system of the Loongson 3A processor.
Further, the step 1 includes the steps of:
step 1.1, setting an interrupt balance algorithm;
step 1.2, running an interrupt balance algorithm, transferring the interrupt of the corresponding processor currently positioned on the MaxX to the processor on the MaxX when executing the interrupt balance operation, and judging the load condition of the processor corresponding to the interrupt;
and step 1.3, realizing multi-core dynamic optimization service in the Linux system.
Moreover, the interrupt balancing algorithm of step 1.1 is:
ΔS＝S-Sa
wherein S is the standard deviation of the occupancy rate of the Loongson 3A processor, N is 4,
the real-time occupancy of the processor is the mean value of X1 to Xn, the value range of the real-time occupancy of the processor is 0 to 4, Sa is a standard deviation threshold value, and Delta S is an equilibrium index.
Moreover, the range of the standard deviation threshold value Sa is 3.5 to 4.5; when the balance rate index delta S is less than or equal to 0, judging that the occupancy rate of the current system processor is relatively uniform; when the balance rate index Δ S >0, it is determined that the current processor is in a situation with a large dispersion, and it is necessary to perform an interrupt balance operation.
Moreover, said step 1.3 comprises the steps of:
step 1.3.1, judging the current interrupt condition of the Linux system and the load condition of a corresponding processor, and running an interrupt balancing algorithm to make a decision;
step 1.3.2, judging whether the balance rate index delta S of the optimization condition is more than 0, if so, carrying out interrupted affinity change in real time, counting and carrying out step 1.3.4, otherwise, carrying out step 1.3.3;
step 1.3.3, if the execution frequency is increased, reducing the execution frequency to the common frequency, otherwise, performing step 1.3.4;
step 1.3.4, if the number of times of optimizing the condition reaches three times, increasing the execution frequency of the service and performing step 1.3.5, otherwise, directly performing step 1.3.5;
and 1.3.5, delaying and returning to the step 1.3.1.
Moreover, the specific implementation method of the step 2 is as follows:
2.1, aiming at an igb _ uio module in a DPDK EAL assembly, changing an MSI interruption mode into a Legacy interruption mode;
step 2.2, modifying the memory barrier function, the atomic addition and subtraction and the CAS function of the atomic operation library in the DPDK EAL component to adapt to the Linux system platform of the Loongson 3A;
step 2.3, adjusting library functions related to the hardware architecture in the DPDK EAL component according to the modification mode of the atomic operation library;
and 2.4, optimizing the transplanted DPDK EAL component.
Moreover, the adjustment of the library function includes adjustments to endianness, processor time tag counters, memory copy, prefetch operations, read-write locks, spin locks, vector instructions, and I/O reads and writes.
The invention has the advantages and positive effects that:
the invention sets an interrupt balance algorithm; running an interrupt balance algorithm, and transferring the interrupt of the corresponding processor currently positioned on the MaxX to the processor on the MaxX when executing the interrupt balance operation; implementing multi-core dynamic optimization service in Linux system to implement multi-core dynamic optimization service of interrupt processing; the MSI interruption mode is changed into a Legacy interruption mode; modifying the content of the atomic operation library in the DPDK EAL component; adjusting library functions related to a hardware architecture in a DPDK EAL component; optimizing the transplanted DPDK EAL component to modify the DPDK EAL component and transplanting the DPDK EAL component to a Linux system of the Loongson 3A processor, thereby optimizing the performance of the network interface. The method fills the blank of related research in China, has universality, and can be used for improving the throughput rate, the packet loss rate and the time delay index of a network interface when all mainstream gigabit and tera network cards and Loongson 3A are suitable for a Linux system.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a flow chart of the break-even algorithm of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
In this embodiment, the Loongson platform selects A3A 3000 processor, the chipset selects a Loongson 7A1000 bridge chip, the Loongson 7A1000 bridge chip is connected with the processor through an HT3.0 bus interface, the test network card selects an Intel I350 chip, and the test network card is connected to the Loongson 7A1000 through a PCIE2.0x4 bus; the system selects a standard-winning kylin desktop operating system (Loongson edition) V5.0, a Linux kernel edition is 3.10.0, and a transplanted DPDK edition is 17.11.4.
A method for optimizing the performance of a Linux system network interface suitable for a loongson 3A, as shown in fig. 1, includes the following steps:
step 1, constructing a multi-core dynamic optimization flow for interrupt processing of the Loongson 3A processor. The step 1 comprises the following steps:
step 1.1, setting an interrupt balance algorithm: Δ S-Sa
Wherein S is the standard deviation of the occupancy rate of the Loongson 3A processor, N is 4,
the real-time occupancy of the processor is the mean value of X1 to Xn, the real-time occupancy of the processor is Xi%, the value range of i is 0 to 4i and is 0 to 4, Sa is the value range of a standard deviation threshold value and is 3.5 to 4.5, and Delta S is an equilibrium rate index, and when the equilibrium rate index Delta S is less than or equal to 0, the occupancy of the current system processor is judged to be relatively uniform; when the balance rate index Δ S>And when the value is 0, judging that the current processor is in a situation with larger dispersion and needs to execute interrupt balance operation.
Step 1.2, running an interrupt balance algorithm, transferring the interrupt of a corresponding processor currently positioned on MaxX to the processor on MaxX when executing interrupt balance operation, and judging the load condition of the processor corresponding to the interrupt, wherein when Max { Xi }, 0< i < 4; representing the maximum processor occupancy among all cores currently; when MinX ═ Min { Xi }, 0< i < 4; representing the minimum processor occupancy among all cores currently.
And step 1.3, realizing multi-core dynamic optimization service in the Linux system. As shown in fig. 2, step 1.3 comprises the following steps:
step 1.3.1, judging the current interrupt condition of the Linux system and the load condition of a corresponding processor, and running an interrupt balancing algorithm to make a decision;
step 1.3.2, judging whether the balance rate index delta S of the optimization condition is more than 0, if so, operating/proc/irq/irqNum/smp _ affinity, carrying out interruption affinity change in real time, counting and carrying out step 1.3.4, otherwise, carrying out step 1.3.3;
step 1.3.3, if the execution frequency is increased, reducing the execution frequency to the common frequency, otherwise, performing step 1.3.4;
step 1.3.4, if the number of times of optimizing the condition reaches three times, increasing the execution frequency of the service and performing step 1.3.5, otherwise, directly performing step 1.3.5;
and 1.3.5, delaying and returning to the step 1.3.1.
The optimized service name in the step is NetAccerService, the optimized service name runs in a system service form, and after the optimized service is started, the optimized service is automatically loaded in a background without being processed by a user. And when updating, performing online updating through the self-development script. The pseudo code for the final implementation of the service is as follows:
do
{
collecting the current/proc/stat interrupt condition;
collecting the load condition of the corresponding processor core of the interrupt;
running an interrupt balance algorithm to perform interrupt optimization decision;
if (satisfy the optimization condition)
Performing real-time interruption affinity change by operation/proc/irq/irqNum/smp _ affinity;
else if (execution frequency of current service is increased)
Reducing the execution frequency to normal;
if (three times satisfying the optimization condition)
Increasing the execution frequency of the service;
waiting for a certain time;
whisle (NetAccerService open)
And 2, modifying the DPDK EAL component and transplanting the DPDK EAL component into a Linux system of the Loongson 3A processor. The specific implementation method of the step 2 is as follows:
2.1, aiming at an igb _ uio module in a DPDK EAL assembly, changing an MSI interruption mode into a Legacy interruption mode;
the initialization function corresponding to the interrupt mode in this step is:
igbuio_config_intr_mode(intr_mode)；
the module parameter is intr _ mode, three interrupts of msix, msi and legacy are provided, the default is msix, and the mode needs to be changed into the legacy mode.
Step 2.2, modifying the memory barrier function, the atomic addition and subtraction and the CAS function of the atomic operation library in the DPDK EAL component to adapt to the Linux system platform of the Loongson 3A;
in this step, the Loongson 3A3000 is based on an MIPS framework, a small end program is adopted, the page size of a large-page memory is 32M, and a large difference exists between an instruction set and X86. The code implementation mode under the x86 platform is characterized in that in a winning Chinese unicorn edition, a lib/library _ eal/common/include/arch/mips directory is newly built, rte _ atomic.h (atomic operation) implementation codes are added, wherein the implementation codes comprise memory barrier functions, 16/32/64-bit atomic addition and subtraction and CAS function codes, and the contents are according to a Loongson 3A3000 processor user manual.
Step 2.3, adjusting library functions related to a hardware architecture in the DPDK EAL component according to a modification mode of an atomic operation library, wherein the adjustment comprises adjustment of byte order, a processor time tag counter, memory copy, pre-fetch operation, a read-write lock, a spin lock, a vector instruction and I/O read-write as shown in Table 1;
TABLE 1 EAL component hardware-related code List
And 2.4, optimizing the transplanted DPDK component, wherein the method comprises the modes of using user mode driving, mixed interrupt polling mode and the like, and belongs to the characteristics of a DPDK tool.
It should be emphasized that the embodiments described herein are illustrative rather than restrictive, and thus the present invention is not limited to the embodiments described in the detailed description, but also includes other embodiments that can be derived from the technical solutions of the present invention by those skilled in the art.
Claims (7)
1. A performance optimization method for a Linux system network interface suitable for a Loongson 3A is characterized by comprising the following steps:
step 1, constructing a multi-core dynamic optimization flow for interrupt processing of a Loongson 3A processor;
and 2, modifying the DPDK EAL component and transplanting the DPDK EAL component into a Linux system of the Loongson 3A processor.
2. The method for optimizing the performance of the network interface of the Linux system suitable for the Loongson 3A according to claim 1, wherein: the step 1 comprises the following steps:
step 1.1, setting an interrupt balance algorithm;
step 1.2, running an interrupt balance algorithm, transferring the interrupt of the corresponding processor currently positioned on the MaxX to the processor on the MaxX when executing the interrupt balance operation, and judging the load condition of the processor corresponding to the interrupt;
and step 1.3, realizing multi-core dynamic optimization service in the Linux system.
3. The method for optimizing the performance of the network interface of the Linux system applied to the Loongson 3A according to claim 2, wherein: the interrupt balancing algorithm of the step 1.1 is as follows:
ΔS＝S-Sa
wherein S is the standard deviation of the occupancy rate of the Loongson 3A processor, N is 4,
the real-time occupancy of the processor is the mean value of X1 to Xn, the value range of the real-time occupancy of the processor is 0 to 4, Sa is a standard deviation threshold value, and Delta S is an equilibrium index.
4. The method for optimizing the performance of the network interface of the Linux system applied to the Loongson 3A according to claim 3, wherein: the value range of the standard deviation threshold Sa is 3.5-4.5; when the balance rate index delta S is less than or equal to 0, judging that the occupancy rate of the current system processor is relatively uniform; when the balance rate index Δ S >0, it is determined that the current processor is in a situation with a large dispersion, and it is necessary to perform an interrupt balance operation.
5. The method for optimizing the performance of the network interface of the Linux system applied to the Loongson 3A according to claim 3, wherein: the step 1.3 comprises the following steps:
step 1.3.1, judging the current interrupt condition of the Linux system and the load condition of a corresponding processor, and running an interrupt balancing algorithm to make a decision;
step 1.3.2, judging whether the balance rate index delta S of the optimization condition is more than 0, if so, carrying out interrupted affinity change in real time, counting and carrying out step 1.3.4, otherwise, carrying out step 1.3.3;
step 1.3.3, if the execution frequency is increased, reducing the execution frequency to the common frequency, otherwise, performing step 1.3.4;
step 1.3.4, if the number of times of optimizing the condition reaches three times, increasing the execution frequency of the service and performing step 1.3.5, otherwise, directly performing step 1.3.5;
and 1.3.5, delaying and returning to the step 1.3.1.
6. The method for optimizing the performance of the network interface of the Linux system suitable for the Loongson 3A according to claim 1, wherein: the specific implementation method of the step 2 comprises the following steps:
2.1, aiming at an igb _ uio module in a DPDK EAL assembly, changing an MSI interruption mode into a Legacy interruption mode;
step 2.2, modifying the memory barrier function, the atomic addition and subtraction and the CAS function of the atomic operation library in the DPDK EAL component to adapt to the Linux system platform of the Loongson 3A;
step 2.3, adjusting library functions related to the hardware architecture in the DPDK EAL component according to the modification mode of the atomic operation library;
and 2.4, optimizing the transplanted DPDK EAL component.
7. The method for optimizing the performance of the network interface of the Linux system applied to the Loongson 3A according to claim 6, wherein: the adjustments to the library functions include adjustments to endianness, processor time tag counters, memory copy, prefetch operations, read-write locks, spin locks, vector instructions, and I/O reads and writes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011355830.8A CN112463329A (en) | 2020-11-27 | 2020-11-27 | Linux system network interface performance optimization method suitable for Loongson 3A |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011355830.8A CN112463329A (en) | 2020-11-27 | 2020-11-27 | Linux system network interface performance optimization method suitable for Loongson 3A |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112463329A true CN112463329A (en) | 2021-03-09 |
Family
ID=74809789
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011355830.8A Pending CN112463329A (en) | 2020-11-27 | 2020-11-27 | Linux system network interface performance optimization method suitable for Loongson 3A |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112463329A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115705194A (en) * | 2021-08-13 | 2023-02-17 | 华为技术有限公司 | Code processing method under hardware memory order architecture and corresponding device |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111107053A (en) * | 2019-11-15 | 2020-05-05 | 北京宇航系统工程研究所 | Method and system for firewall to automatically detect bound network port based on Loongson DPDK |
-
2020
- 2020-11-27 CN CN202011355830.8A patent/CN112463329A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111107053A (en) * | 2019-11-15 | 2020-05-05 | 北京宇航系统工程研究所 | Method and system for firewall to automatically detect bound network port based on Loongson DPDK |
Non-Patent Citations (2)
| Title |
|---|
| 胡炜 等: ""DPDK 在国产龙芯平台的移植及评估"", 《通信技术》 * |
| 龚俊: ""基于龙芯3A的Linux系统网络设备性能优化设计与实现"", 《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115705194A (en) * | 2021-08-13 | 2023-02-17 | 华为技术有限公司 | Code processing method under hardware memory order architecture and corresponding device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US5862400A (en) | Formula processor having cache means for storing and retrieving results of previously computed formulas based on identification and dependency information | |
| US7370326B2 (en) | Prerequisite-based scheduler | |
| US9886242B2 (en) | Methods to optimize a program loop via vector instructions using a shuffle table | |
| US7480747B2 (en) | Method and apparatus to reduce latency and improve throughput of input/output data in a processor | |
| JP2015191657A (en) | Inter-architecture compatibility module for allowing code module of one architecture to use library module of another architecture | |
| JP2021090188A (en) | Streaming fabric interface | |
| US20210089317A1 (en) | Instruction processing apparatuses, processors, and processing methods | |
| JPH076151A (en) | Cpu core bus optimized for access of on-chip memory device | |
| Skillman et al. | A technical overview of cortex-m55 and ethos-u55: Arm’s most capable processors for endpoint ai | |
| CN112463329A (en) | Linux system network interface performance optimization method suitable for Loongson 3A | |
| WO2021168861A1 (en) | Multi-core processor, multi-core processor processing method and related device | |
| US20210089482A1 (en) | Processor and interrupt controller therein | |
| CN113268270A (en) | Acceleration method, system and device for paired hidden Markov models | |
| WO2021179411A1 (en) | Quantum computing-oriented data interaction device, method and apparatus and medium | |
| CN111338777B (en) | Low-delay high-stability autonomous platform interrupt response method and equipment | |
| JPH10283304A (en) | Method and system for processing interruption request | |
| US20220092010A1 (en) | Multi-host direct memory access system for integrated circuits | |
| US11467844B2 (en) | Storing multiple instructions in a single reordering buffer entry | |
| CN117597664A (en) | Dynamic capability discovery and execution for accelerators and devices in multi-tenant systems | |
| US11886910B2 (en) | Dynamic prioritization of system-on-chip interconnect traffic using information from an operating system and hardware | |
| WO2021061626A1 (en) | Instruction executing method and apparatus | |
| CN109542516A (en) | A kind of acceleration arm processor concurrent working system and its working method | |
| CN117312202B (en) | System on chip and data transmission method for system on chip | |
| US20230333921A1 (en) | Input/output (i/o) virtualization acceleration | |
| US12032925B1 (en) | Latency processing unit |
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: 20210309 |