Abstract: With Huawei Kunpeng BoostKit enable virtualization suite (referred to as Kunpeng BoostKit virtualization), can accelerate the journey to cloud computing journey. The KAE acceleration engine makes data encryption more secure. The live broadcast will introduce the panorama of Kunpeng’s BoostKit acceleration library, based on the performance optimization practices of the BoostKit acceleration library in typical application scenarios such as WEB, big data, and distributed storage.
This article is shared from the HUAWEI cloud community " [Cloud-based co-creation] Kunpeng application enablement kit KAE acceleration engine to make data encryption more secure ", the author: I have tested IE and become a siege lion.
Preface and introduction:
Huawei Kunpeng BoostKit virtualization enablement kit introduction:
Nowadays, transforming from a company that uses digital technology to a digital company is a challenge facing all walks of life around the world. Application modernization is the core of digital transformation, helping companies attract customers, empower employees, optimize operations, and improve products. As an IT infrastructure for digital transformation, cloud computing technology has developed rapidly in recent years, especially with the rapid development of virtualization, cloud services, and containers. The process of digital transformation of enterprises has also benefited greatly from the development of cloud computing technology. Cloud computing innovation, also largely due to the rapid development of open source technology and ecology , to QEMU-KVM, OpenStack, Docker, Kubernetes as the representative of open source cloud computing technology, breaking the once closed inefficient "chimney" The computing power architecture is constantly enriching IT infrastructure, helping users' applications to evolve in a more agile and efficient direction, and making great strides towards digital transformation.
With the help of Huawei Kunpeng BoostKit virtualization enablement kit (Kunpeng BoostKit virtualization for short), the journey to cloud computing can be accelerated. The KAE acceleration engine makes data encryption more secure. The live broadcast will introduce the panorama of Kunpeng’s BoostKit acceleration library, based on the performance optimization practices of the BoostKit acceleration library in typical application scenarios such as WEB, big data, and distributed storage.
Acceleration library introduction:
The acceleration library can be combined with the computer architecture to improve the calculation efficiency of the basic software library by changing the data structure and algorithm of the software code; it can also use the diversity of computer chips, through software development, give full play to the high performance characteristics of the proprietary chip, and improve The computational efficiency of basic functions in a specific field.
Speed up library issues:
Rabbit reproduction problem: There is a pair of newborn rabbits. From the third month, they give birth to a pair of rabbits every month, and the newborn rabbits give birth to a pair of rabbits every month from the third month. According to this rule, and assuming that the rabbits have not died, how many rabbits will there be in 1 year? What about 10 years later?
It can be seen from the above problems that the efficiency of the algorithm will affect the performance, and there are many factors that affect the performance of the software. What Kunpeng acceleration library does is find out these factors that affect software performance and perform performance tuning . From this question, we derive the meaning of the acceleration library.
Why do acceleration libraries (1)
1. Acceleration library improves software computing efficiency
The calculation speed of the CPU is fast, do you need an acceleration library? Of course not!
In sports, the pursuit of all athletes is higher, faster, and farther, while the IT industry pursues faster and cheaper. Purely improving hardware computing power cannot achieve the purpose of quickly solving problems. It is necessary to rely on software algorithms to change the execution efficiency to truly increase the computing speed.
Therefore, if the user interface is faster and cheaper, it must be realized through the best software (acceleration library) to reflect the best performance of the hardware and provide the best service.
This is especially true when the current CPU development encounters a bottleneck.
Why do acceleration libraries (2)
2. Accelerate the library reconstruction of Moore's Law
What impact will the failure of Moore's Law bring?
Moore's Law is that with the development of technology, the computing power of the machine is gradually developed. enables the computer's computing power to maintain a rapid increase at a certain rate. However, in recent years, under the current technological background, the processing power of computers has encountered a development bottleneck, and the processing power of computers cannot continue to increase rapidly, which means that the cost-effectiveness for us has decreased.
We have found that even in the case of concurrency, the utilization of hardware resources is still improved, which requires us to reconstruct Moore's Law through software and combining some characteristics of hardware through software. Let it still grow rapidly at a linear rate.
So, is there a way to improve it?
◆Hardware diversity brings us the possibility of reconstructing Moore's Law;
◆The diversity of hardware brings us the possibility of reconstructing Moore's Law; the full use of computing power is also an important way.
Why do acceleration libraries (3)
- Acceleration library is one of the methods of system engineering
No component or module in the IT system exists independently, but cooperates with each other to form a whole, and software plays a huge role in it. for example:
◆In the era of single-core CPU, if there is a blocking task, the CPU will be in the waiting stage, and the computing power of the CPU will be wasted. Multithreading technology uses time slice division to give full play to the power of the CPU. This is a software technology;
◆In the era of multi-core CPU, how to make full use of multi-core by multi-threading also requires software scheduling for effective operation;
◆In the era of multiple CPUs, it is necessary to consider the coordination of CPU and memory through software, and effectively utilize resources such as multiple CPUs and memory;
◆Of course, multi-threading does not give full play to the capabilities of the CPU. It also needs to be able to ensure that the code implemented by the software, when the CPU is executed, the pipeline really flows.
1. Panoramic view of Kunpeng BoostKit acceleration library
1.1 The Best Solution of Kunpeng Acceleration Library
The acceleration library is actually more towards the bottom layer, between hardware resources and solutions . It belongs to the layer of basic software and is applied to various solutions, such as big data, distributed storage, database and so on. So as to support these solutions in government, operators, finance and other fields. It can be adapted to mainstream operating systems such as openEuler, CentOS, etc., by giving full play to the capabilities of the hardware, or embedded in the operating system, and as a function library alone to support the solution, improve the performance of the solution, and ultimately improve customers The price/performance ratio. That is to say, there is no need to increase the configuration of this hardware all the time, and always purchase additional hardware resources to improve the performance of the computer.
1.2 What is Kunpeng BoostKit?
Kunpeng application enablement kit BoostKit, releases multi-level performance advantages
BoostKit is a full-stack optimized enablement kit built on top of the Kunpeng whole machine (server) environment. The specific structure is shown in the figure below:
The hardware is the supply of computing power, and the performance of the software is a release of computing power. How to fully release the potential of hardware computing power and achieve the ultimate performance of application software is one of the main challenges facing the industry.
1.3 Eight Application Scenarios of Kunpeng Boostkit
The Kunpeng application enablement kit BoostKit releases multi-level performance advantages and provides eight scenarios of application enablement kits: big data, distributed storage, database, virtualization, ARM native, Web/CDN, NFV and HPC.
At present, Kunpeng BoostKit has been applied to the core business scenarios of key industries to help customers achieve significant performance acceleration effects. For example, in the application practice of Guangdong Government Cloud, we use the Kunpeng BoostKit enablement kit on the Kunpeng platform. Through the Huawei Cloud platform, the full-stack optimization of the domestic database, the query performance of the application support library is improved by more than 15%; in the bank In the practice of the core application system of Kunpeng, Kunpeng BoostKit is fully optimized for Kunpeng server + Galaxy Kylin server operating system V10, the domestic database is fully optimized, and the bill query performance is improved by 30%. It has been launched and served more than 100,000 users.
In the future, we will continue to work with customers and partners in various industries to accelerate business software in various industries based on Kunpeng BoostKit, and carry out continuous business plan innovation. Of course, Huawei Kunpeng BoostKit's innovation is not limited to these eight application scenarios. We can also discuss other application scenarios together.
The overall introduction of the Kunpeng acceleration library:
Kunpeng acceleration library: In-depth performance optimization of the software base library, building the performance competitiveness of commonly used software libraries on the Kunpeng platform, hardware acceleration + software acceleration library, and 18 acceleration libraries have been completed, which are based on the Kunpeng platform instruction optimization and software and hardware Combined acceleration technology.
Remarks: KAE (Kunpeng AcceleratorEngine) Kunpeng acceleration engine; HPRE (High Performance RSA Engine) high-performance RSA acceleration engine; SEC (Security Engine) hardware security acceleration engine.
Huawei Kunpeng Computing Acceleration Library is the collective name of Huawei's software collection after performance optimization of Kunpeng processors. Kunpeng BoostKit acceleration library provides deep optimization based on ARM instructions and acceleration libraries developed based on Kunpeng KAE (Kunpeng Hardware Acceleration Engine), covering Seven types of acceleration libraries, including system library, compression, encryption and decryption, media, math library, storage, and network, provide high-performance acceleration for application scenarios such as big data encryption and decryption, distributed storage compression, and video transcoding.
1. Compressed
Use Kunpeng hardware acceleration module or Kunpeng instructions to optimize the performance of the industry's mainstream open source compression libraries.
2. Encryption and decryption
Use Kunpeng hardware acceleration module and Kunpeng instructions to optimize the performance of the OpenSSL library, support the automatic coordination of hard acceleration and instruction acceleration, and use the encryption and decryption acceleration library without modifying the application logic.
3. System library
Based on the characteristics of the Kunpeng micro-architecture, the Kunpeng instructions are used to optimize the performance of the system's general basic library, and the instruction functions of the x86 platform are mapped to the public modules of the Kunpeng platform.
4. Media
Provide high-performance media primitive library and video codec library based on Kunpeng acceleration instructions.
5. Math library
Based on the characteristics of Kunpeng micro-architecture and Kunpeng acceleration instructions, it provides commonly used high-performance basic math libraries.
6. Storage
The use of cache disks with efficient prefetching algorithms improves system storage IO performance, thereby improving the overall performance of storage IO-intensive scenarios.
7. Network
Based on the characteristics of Kunpeng architecture, it provides TCP/IP and OVS network acceleration technology.
2. Introduction of Kunpeng BoosKit acceleration engine
2.1 Introduction of Kunpeng BoostKit KAE acceleration engine
Kunpeng acceleration engine is based on the hardware acceleration solution provided by Kunpeng 920 processor, including KAE encryption and decryption and KAEzip, respectively used to accelerate SSL/TLS applications and data compression, which can significantly reduce processor consumption, and improves processor efficiency. In addition, the acceleration engine shields the application layer from its internal implementation details, and users can quickly migrate existing services through OpenSSL and zlib standard interfaces.
Open source enablement for the community:
➢ Compatible with OpenSSL and Zlib standard interfaces, application layer call logic does not need to be adjusted, easy to use
➢Based on Huawei's self-developed UACCE accelerator kernel framework to avoid the path loss of traditional kernel calling methods
➢The driver is open source in Kunpeng community, users download the source code and compile and use by themselves
➢Support SR-IOV pass-through, support hardware acceleration in virtualization & container scenarios
2.2 Kunpeng BoostKit KAE acceleration engine enable
KAE enabling and calling process:
- The BMC subsystem manages the license of the KAE acceleration system. When the system is initialized, the BMC passes the license to the BIOS system.
- The BIOS subsystem parses the license information, and performs enable control on the chip acceleration subsystem according to the analysis results, and reports the accelerator ACPI table to the kernel.
- The upper application calls the Kunpeng acceleration engine based on the OpenSSL/zlib acceleration library, and calls the chip acceleration subsystem through register operations.
2.3 Technical specifications of Kunpeng BoostKit KAE
KAE encryption and decryption is the encryption and decryption module of the Kunpeng acceleration engine. It uses the Kunpeng hard acceleration engine RSA/SM3/SM4/DH/MD5/AES algorithm, combined with a lossless user mode driver framework, to provide high-performance symmetric encryption and decryption, asymmetric encryption and decryption algorithm capabilities , Compatible with OpenSSL 1.1.1a and later versions, support synchronous & asynchronous mechanism.
currently mainly supports the following algorithms:
• The digest algorithm SM3/MD5 supports asynchronous models.
• Symmetric encryption algorithm SM4, support asynchronous model, support CTR/XTS/CBC/ECB/OFB mode.
• Symmetric encryption algorithm AES, support asynchronous model, support ECB/CTR/XTS/CBC mode.
• Asymmetric algorithm RSA, support asynchronous model, support Key Sizes 1024/2048/3072/4096.
• Key agreement algorithm DH, support asynchronous model, support Key Sizes768/1024/1536/2048/3072/4096.
2.4 Kunpeng BoostKit KAEzip compression library big data application scenario
• A customer’s original big data platform architecture is based on open source HDP
• Use BoostKit KAEzip acceleration library to achieve Gzip compression acceleration: improve HDFS compression performance by 80%+
• Based on the open source HDP big data platform, the overall performance is improved by 30%+
Under the "new infrastructure", the rapid development of key technologies such as big data, the adjustment of key parameters and related components, to optimize operation efficiency. Focusing on challenges such as low efficiency of big data query and difficult performance optimization, it provides open source enablement and tuning of big data components, basic acceleration software packages such as IO intelligent prefetching, application acceleration software packages such as Spark algorithm acceleration library, and open source openLooKeng query engine, Improve the efficiency of big data analysis.
Learn more: https://www.hikunpeng.com/developer/boostkit/big-data
2.5 Kunpeng BoostKit KAE encryption and decryption web application scenario
Focusing on issues such as the HTTPS connection performance of web applications and the availability of web open source components, it provides an enabling tuning guide for open source web components. By enabling the Kunpeng RSA acceleration engine, it helps customers improve the security of the web site and achieve the ultimate user experience of HTTPS access.
Kunpeng BoostKit Web follows the open architecture standard. supports all open source web components, and provides good scene applicability according to the characteristics of specific web components. The Kunpeng BoostKit Web software stack is shown in the following figure:
Kunpeng BoostKit Web supports all open source web applications, has the characteristics of simple installation and deployment, and convenient system configuration. Combined with the multi-core architecture provided by Huawei Kunpeng 920 processor and Kunpeng acceleration engine, it can provide more excellent high concurrent processing capabilities. The main values are as follows:
1. Open ecology
Kunpeng BoostKit Web has good support for open source web applications and frameworks developed in commonly used languages such as C/C++/C#, Java, Python, Perl, PHP, Go, etc. It also supports Dongfangtong and Kingdee Tianyan, Zhongchuang Middleware and Commercial Web applications such as Powerland have also provided adaptation support. Currently, the transplantation and tuning of the following open source Web applications have been mainly completed, which have been opened in the Kunpeng community.
2. High performance
Based on the Kunpeng acceleration engine provided by the Huawei Kunpeng 920 processor, it can implement hardware offloading of the RSA2048 asymmetric encryption and decryption algorithm in HTTPS protocol processing, greatly reducing CPU resource usage, and HTTPS processing performance is doubled compared with software calculations, which can free up more CPU Used for business processing.
Based on the multi-core architecture and multi-core scheduling optimization algorithm of Huawei's Kunpeng 920 processor, the performance of Web applications with high concurrency, low latency, and computationally intensive characteristics has been significantly improved.
3. Easy to use:
Simple installation and deployment, performance optimization, and easy operation and maintenance:
Simplified installation: All web components provide migration installation instructions and one-click installation scripts, which are easy to install.
Simple performance optimization: All web components provide performance tuning guidance and one-click tuning scripts. Through tuning, the performance of the Kunpeng processor can be fully utilized.
4. Flexible deployment
Kunpeng BoostKit Web supports all installation and deployment methods in the industry: physical machine, virtual machine and container deployment, while meeting the flexible business needs of customers, while providing excellent computing capabilities.
Three, Kunpeng BoostKit soft acceleration library introduction
3.1 Technical specifications of Kunpeng BoostKit soft acceleration library
BoostKit can be divided into three categories:
• ARM-supported acceleration library: The open source community provides ARM-supported acceleration libraries, which are verified on the Kunpeng platform
• hard acceleration library (based on KAE): provides industry-leading performance acceleration capabilities, including ZIP (hard acceleration-compression), HPRE (hard acceleration-asymmetric encryption and decryption), SEC (hard acceleration-symmetric encryption and decryption)
• soft acceleration library: an acceleration library based on the deep optimization of Kunpeng instructions, and is open to the open source community
basic acceleration package:
Covers performance optimization methods for 4 subsystems, 7 types of acceleration libraries, and acceleration algorithms for 3 optimization directions.
Learn more: https://www.hikunpeng.com/zh/developer/boostkit/library
3.2 Performance acceleration effect of Hyperscan in typical scenarios
This is a high-performance regular expression matching library. A large number of efficient algorithms and SIMD instructions are used to achieve high-performance matching. Optimized based on the 5.2.0 version of the official website, using ARM Neon instructions to accelerate. Use multiple optimization methods such as data prefetching, branch prediction, restructuring, and loop unrolling to achieve acceleration.
application scenario:
[Scenario 1: Regular expression compilation presets]
Regular expression rules are configured and imported by applications that integrate Hyperscan. Considering performance issues, they are imported during the initialization phase, compiled, parsed, and cached.
[Scene 2: Block pattern matching]
It is suitable for matching a piece of ready-made complete data, and the result will be returned when the matching is over.
[Scenario 3: Streaming pattern matching]
Flow mode is a special matching mode designed for cross-message matching in network scenarios.
3.3 Kunpeng acceleration library-compression library (Snappy/ZSTD/GZIP) performance acceleration effect
Snappy: Google is a fast data compression and decompression library written in C++ language based on the idea of LZ77.
ZSTD: FaceBook open source to the compression library, while taking into account the compression ratio and performance of the compression library, the compression ratio is ahead of the traditional deflate algorithm.
gzip: basic compression software, used for lossless compression of files, based on the deflate algorithm.
Using Kunpeng 920's CRC and prefetch instructions can improve the performance of compression and decompression.
Keep the existing interfaces and functional specifications of snappy/ZSTD/gzip unchanged, improve the compression and decompression speed, and will not affect the compression and decompression results of each level.
3.4 Kunpeng BoostKit Math Library (Kunpeng Math Library)
KML is a high-performance mathematical calculation acceleration library based on the Huawei Kunpeng processor. It provides mathematical functions optimized based on the Kunpeng platform. Mathematics library (KML_MATH), fast Fourier transform library (KML_FFT) and other five sub-libraries, which can be integrated and used according to business needs.
While ensuring the accuracy of mathematical operations, KML_FFT is based on the Kunpeng architecture, through vectorization and algorithm improvements, deep optimization of the fast discrete Fourier transform, which greatly improves the performance of the fast Fourier transform interface function. Compared with the open source FFTW, KML_FFT has a performance improvement of about 1 times.
In a broad sense, the mathematical library refers to a mathematical function interface library that provides high-performance scalar and vector calculations. Common mathematical libraries include the Intel Math Kernel Library (hereinafter referred to as Intel MKL) and some open source mathematical libraries. Kunpeng Math Library (Kunpeng Math Library, hereinafter referred to as KML) provides high-performance mathematical functions optimized based on the Kunpeng platform. When a user migrates an application that uses math library functions to the Kunpeng computing platform, the program needs to be re-adapted and developed due to the difference in function definitions of different computing platforms or different math libraries. In response to this problem, we provide the Kunpeng Math Library migration guide.
The guide contains the steps for migrating some math libraries in Intel MKL to KML (see Intel MKL to KML migration), including MKL's VML to KML_VML, MKL's BLAS to KML_BLAS, etc. It also contains the steps for migrating some common libraries to KML. (See the migration of other commonly used math libraries to KML). Through the migration, users can obtain better computing performance on the Kunpeng computing platform than the general open source math library.
3.5 Kunpeng BoostKit HMPPS signal library
Hyper Media Performance Primitives, Huawei's self-developed high-performance media performance primitive library, including image processing HMPPI function library and signal processing HMPPS function library two sub-libraries, adapt to multiple operating systems, provide function API, applied to motion tracking, radar signals , Image analysis, video enhancement, AI acceleration, medical scanning, communication engineering and many other industries. At present, the signal library HMPPS has completed the development of 1000+ function interfaces.
There are about 1700 functions in the design plan. The development of 1000+ function interfaces currently includes the following aspects:
- Realize the functions of memory operation and data calibration. Memory operation includes functions such as copy and initialization;
- Realize vector operations, including arithmetic, conversion, statistics and window operations, etc.
- Provide vector median filter function
- Realize the function of fixed vector accuracy, including the function of finding the power of vector elements
- Realize the fast Fourier transform of the signal, etc.
3.6 Kunpeng BoostKit HW265 video encoding performance data
The HW265 video encoder is a Huawei self-developed H.265 video encoder that complies with the H.265/HEVC video coding standard and is accelerated by the NEON instruction of the Kunpeng processor. It is an open source free software and function library used to encode movies that comply with the High Efficiency Video Coding (HEVC/H.265) standard.
• The HW265 8bit standard version achieves 1 to 5 times the performance advantage compared to the X86 open source X265.
• HW265 8bit high-definition low-code version realizes an average of 50% bit rate savings.
• Extreme Compression High Quality
3fps, suitable for popular movies/short videos and other scenes
• universal code Balance
25fs, suitable for OTT on-demand scenes
• fast encoding Speed
60fps, suitable for shows, games, online education and other scenes
• HW265 provides a video encoding library, which can be integrated into the FFmpeg open source transcoding framework.
• Keeping the x265_upper interface and functional specifications unchanged, improves the video encoding speed to without changing the original framework and logic of each software layer.
• Use arm v8 Neon vector assembly instructions to optimize the underlying operator of x265, improves the overall performance .
Optimize the X265 transcoding core operator, and increase the average performance by 13%+
• The code is open source and has given back to the open source community
Fourth, Kunpeng BoostKit IO smart prefetch introduction
4.1 Introduction to BoostKit IO Intelligent Prefetching Solution
Intelligent prefetching technology refers to the use of small-capacity high-speed storage media as the cache disk, and the IO data that is predicted to be accessed may be placed in the cache disk in advance, and the data will be directly obtained from the cache next time, which can significantly improve the overall system Storage IO performance. Kunpeng BoostKit distributed storage uses IO intelligent prefetch technology, innovatively uses cache disks and efficient prefetch algorithms to improve system storage IO performance, thereby improving the overall performance of storage IO-intensive scenarios, and increasing storage performance by 20%.
As shown in the smart prefetch use case view, Huawei Bcache runs in the Linux kernel mode as a virtual block device driver, and the prefetch engine acache_ client runs in the user mode as an independent process.
Huawei Bcache hosts the cache disk and the back-end main storage disk, and provides a unified-virtual storage disk as an access entry for the upper-level application.
After Huawei Bcache and acache client are installed on the user's target system, Bcache (the cache disk is paired with the back-end disk) needs to be configured. Then, the disk storage path directly accessed by the upper-level application can be adapted to the Bcache device path.
The application itself does not require any code modification or compilation, and has no perception of intelligent prefetching.
Project background:
In a large number of scenarios in solutions such as distributed storage and big data, access to the performance (bandwidth delay, IOPS) of the I0 memory (hard disk, SSD) has a significant impact on the overall performance.
The intelligent prefetch acceleration solution, innovatively uses a small-capacity cache disk and an efficient prefetch algorithm to improve the system storage 10 performance, thereby improving the overall performance of the above scenarios.
Design:
This solution uses the overall architecture of Linux kernel virtual block device driver Huawei Bcache plus user-mode intelligent prefetch engine. The driver layer is loosely coupled with the prefetch engine, and the application has no perception of the overall framework of intelligent prefetching.
4.2 Intelligent Prefetching Deployment and Typical Benefits of I0 in Big Data Scenarios
deployment method:
The underlying disk device accessed by HDFS in the big data scenario is adapted to the Bcache device, and the cache disk can be selected as Ramdisk and NVME SSD.
Hardware configuration recommendation:
Taking the native configuration of 12 SATA HDD disks for each node as an example, Ramdisk recommends a configuration of 12GB (each HDD is configured with 1GB Ramdisk as a cache disk), and the SSD capacity is recommended to be configured with a 1.6TB NVME SSD.
performance benefits:
The above performance income data adopts the basic hardware configuration of Kunpeng 920 48 core*2 + 384GB Dram, and the hard disk adopts Seagate 8T ST8000NM0055-1 RM112-177.
The intelligent pre-fetch repository is enhanced based on the Linux native Bcache. The intelligent pre-fetch algorithm engine analyzes and predicts user IO request information, automatically optimizes the storage IO configuration based on the prediction results, and guides Bcache to more intelligently perform data pre-fetching , Which can achieve:
• The storage IO configuration-free performance is optimal in each application scenario of the big data platform.
• The application of upper-level big data components is unaware.
• Improve data hit rate and enhance disk read performance.
• The application performance of Spark and HBase in storage IO-intensive scenarios can be improved by 10% to 20%.
The intelligent prefetch storage library is suitable for disk IO-intensive applications in big data. Because the data disk needs to be formatted and bound to the cache disk when used, it is only suitable for use when creating a new cluster.
specific applicable application scenarios of
• HBase sequential reading scenario (25GE networking).
• Spark disk IO-intensive scenarios (not applicable to Spark SQL using Parquet/ORC format).
4.3 IO intelligent prefetch deployment and typical benefits in distributed storage scenarios
deployment method and hardware configuration recommendations:
For a balanced server in a distributed storage scenario, the original recommended configuration has used NVME SSD as the Bcache solution for the cache disk. The intelligent pre-fetch storage 10 acceleration library only needs to be deployed to the target system, and there is no need to modify the hardware and software configuration.
platform typical configuration information:
Server: TaiShan vs x86
CPU: 2*Kunpeng 920 3210
RAM: 8 * 16G DDR4
System Disk: 2 * 480G SATA SSD
Cache disk: 1 * 3.2TB NVMe SSD
Data Disk: 12* 8TB SATA HDD
Network: 2 25GE, 2 GE
V. Kunpeng BoostKit acceleration library information acquisition
5.1 Kunpeng BoostKit acceleration library acquisition method
1. Getting to know BoostKit
BoostKit overall introduction:
https://www.hikunpeng.com/developer/boostkit
2. Understand the overall capability of the program
The overall introduction of BoostKit big data:
https://www.hikunpeng.com/developer/boostkit/big-data
3. Advanced knowledge & growth
BoostKit documentation:
https://support.huaweicloud.com/wtsnew-kunpengbds/index.html
BoostKit big data growth map: https://support.huaweicloud.com/kunpengbds/kunpengbds.html
4. Practice exchange platform:
https://bbs.huaweicloud.com/forum/thread-143349-1-1.html
While enabling open source, Kunpeng BoostKit continues to contribute to open source, returning many key performance optimization results to the open source community. Currently, Kunpeng BoostKit has 190+ key Patch packages and 14 key feature examples adopted by the open source community.
Kunpeng BoostKit submitted an important optimization patch to the C language library Glibc version 2.31, which increased the performance of commonly used functions such as memcpy, memcmp, memchr, strcpy, strlen on Kunpeng and ARM platforms by more than 20%, and allowed many C language to be programmed Language software benefits.
Kunpeng BoostKit provides MariaDB 10.6 with key optimization improvements such as lock optimization, NUMA Aware, Global Counter, etc., to achieve a performance improvement of more than 40% for the MariaDB database on Kunpeng and ARM platforms.
5.2 Developer Contribution Method
Github link: https://github.com/kunpengcompute/Kunpeng
If you have any questions or comments, please also create an issue on Github.
If you want to contact the Kunpeng computing team directly, you can also send an email to kunpengcompute@huawei.com.
Summarize
Facing the era of diverse computing, Huawei is fully opening up Kunpeng's full-stack capabilities and sharing diverse computing tool suites: Kunpeng Application Enablement Kit Kunpeng BoostKit and Kunpeng Development Kit Kunpeng DevKit, accelerating industrial innovation, enabling minimalist development, and working with partners to build Kunpeng Computing industry ecology. This series of courses is mainly aimed at Kunpeng developers and ISV partners to help you quickly understand the best capabilities and practices of the eight scenarios supported by the BoostKit Kunpeng application enablement kit, Kunpeng's full R&D workflow tool kit Kunpeng DevKit and Kunpeng basic software open source And other related content, together with global developers to light up a new era of diverse computing.
Click to follow, and learn about Huawei Cloud's fresh technology for the first time~
**粗体** _斜体_ [链接](http://example.com) `代码` - 列表 > 引用。你还可以使用@来通知其他用户。