This article shares with you the latest status of GaussDB’s business progress and product capability upgrades.
1. Huawei Cloud GaussDB is covering more industries than finance
Starting in 2019, through continuous efforts within Huawei, we have launched the independently innovative GaussDB database that integrates multiple technologies, and have successively completed the replacement of Huawei's internal core systems, which includes three aspects.
First, 6PB of data has been completely replaced and brought online on the terminal cloud. The distributed nodes have a scale of 6,000 nodes, and the resource utilization rate has increased by 30%.
Second, during the ERP replacement, more than 600 sets of business libraries were replaced, and after experiencing ten times the test of sudden traffic, business efficiency was improved tenfold, achieving< a i=1>Zero failures, zero delays and zero adjustments. Let me briefly introduce the characteristics of Huawei's order system. At the end of every month, every quarter, and every year, its traffic is 5-10 times the usual traffic. So when Huawei launched its ERP, we had to go through 20 times the traffic test and stress test before it could go online.
Third, more than 300,000 sets of operator equipment have been shipped in total.
It can also be found from the revenue structure that GaussDB has covered more key industries from the initial financial industry, and the proportion of non-financial companies has doubled. More and more customers choose GaussDB as their digital transformation partner.
2. Polish and mature in rich practical scenarios
The database is a full-scenario software, so the scenario is actually the grindstone of the database. Next, we will further explain that the several businesses just mentioned and the financial business to be discussed later are extremely important for the training of the database.
Huawei's MetaERP system is a typical scenario that heavily uses traditional commercial databases. It is also a representative of a very typical application in the manufacturing industry. At Huawei, ERP is our production system, which supports hundreds of billions of Huawei orders every year. , orders shipped to more than 170 countries. From the perspective of Huawei's own business continuity, Huawei's ERP replacement is equivalent to forcibly crossing the Dadu River on the Long March.
As one of the most complex ERP systems with database application scenarios in the world, Huawei's MetaERP faces several challenges:
-
First of all, there are nearly 700 million lines of SQL scripts that need to be modified;
-
Secondly, during peak business periods, business traffic may reach 5-10 times. Before the actual launch, we conducted a stress test of 20 times the traffic;
-
Finally, there is the transformation of large tables. There are large tables with a maximum of 16 billion rows, and there are more than a dozen tables with more than 10 billion rows. The migration of these large tables is very challenging. So how to solve it?
First, for the replacement of SQL scripts, UGO tools achieve nearly 100% automated structure migration. Today, to replace a database, it is unsustainable if dozens of people are invested in transforming an application to achieve replacement.
Second, data migration, 320 billion rows of data were migrated within 35 hours. In other words, today we no longer have to worry about the consistency and completeness of data migration.
In addition, through intelligent cost estimation, efficient index concurrency control and other algorithms, it successfully passed the stress test of 20 times the maximum peak traffic in business history.
Let me tell you a little story here. After the ERP switch was successful, an operator from Huawei Chengdu ran the asset accounting task for the first time and it only took 3 minutes. But the previous historical experience of this operator required more than 2 hours. Therefore, he thought that the operation might have failed this time, and he immediately issued an early warning according to the operating specifications. After actual confirmation, the result is accurate and it is a "beautiful misunderstanding".
As for Huawei Terminal Cloud Service, this represents a typical application of an emerging type of cloud-native enterprise that was born in the cloud and grew up in the cloud. What are its main challenges? The first is the cost issue, how to improve resource utilization, especially in the case of massive data and distribution, the cost problem caused by a large number of deployed nodes and the cost challenge of converting traditional computer rooms to the cloud; secondly, the massive business belt The large number of distributed demands that are coming require very strong scalability and elastic scaling capabilities; finally, data diversification, because there are both relational data and non-relational data.
GaussDB's native distributed architecture makes the load more balanced. The maximum number of single cluster nodes that have been online exceeds 200, and it supports multiple ecosystems, achieving a substantial improvement in resource utilization and operational efficiency.
There are also financial core systems that have the highest database requirements. Almost all financial CIOs or CTOs know that migration is not easy to do. A large number of applications in the financial industry rely heavily on traditional database interfaces, and there is a lot of uncertainty. Customers all hope There is a set of quantifiable and step-by-step implementation plans.
From chips, servers, storage, networks to databases and operating systems, GaussDB is currently the only domestic brand in China that can achieve software and hardware collaboration and full-stack independent innovation. It is highly compatible with the syntax of traditional databases and has a one-stop smooth migration of data and applications. solution to make migration easier. At the same time, the native distributed architecture based on the majority protocol greatly improves the availability of the system.
3. Facing deeper and broader scenarios, we think about how to better meet customer demands
As GaussDB matures, GaussDB will develop in two directions in the future. The first is to go deep into the financial industry and solve the dual demands of financial customers for digital transformation and sustainable development at once; the second is to move from finance to more key information infrastructure industries such as government affairs, energy, transportation, etc. This is also a new mission.
While doing a thorough job in finance, we found that customers used to only focus on the adaptability of the interface, but now they are gradually beginning to pay attention to the usability and stability of the replaced application. As we all know, most of the reliability of the host comes from the hardware. Without the support of this kind of dedicated hardware, it is very important to ensure the overall availability index of the host through the cooperation of software and hardware. Financial core business is different from the Internet's pursuit of performance peak limits. What is more needed is a certain jitter-free performance so that the delay of each transaction can be controlled. Finally, in the process of customers replacing a few applications to hundreds of applications, how to make the migration a quantifiable implementation process is the biggest challenge facing every financial CIO.
In recent years, the scale of data in related industries has become larger and larger. Due to weak scientific and technological strength and relatively small investment in technical personnel, the database needs to be simple and easy to use. It is best to directly reuse mature technologies that have been implemented in the financial industry. Reduce total cost of ownership and achieve rapid advancement.
4. GaussDB always takes security, trustworthiness and high quality as its primary goals
In order to meet customer demands in more scenarios, a new generation of distributed database GaussDB was released, which has the comprehensive capabilities of "five highs and two easy". Next, I will explain and share with you some of the key new capabilities.
Before talking about specific product capability upgrades and changes, I would like to share with you something that has not changed in GaussDB, which is the pursuit of security, trustworthiness, and high quality.
From product definition, code implementation, open source governance to operation and maintenance, GaussDB is based on Huawei's software engineering full-stack trust framework, achieving trust from results to process trust, and ensuring that the entire tool chain, including the development of GaussDB software, is True independent innovation within the company.
At the same time, a 9-layer test protection network has been built to reduce low-probability and complex interaction problems through a full-link in-depth interactive test platform. Currently, there are 200,000 test cases that have eliminated most basic problems, and this year will continue The number has increased to 800,000, and more than 10 scenario-based protection networks for financial customers have been built to reduce scenario-based financial problems and achieve true high quality.
5. Focus on customer business pain points and upgrade the key capabilities of GaussDB products
In terms of upgrading product capabilities, what everyone is most concerned about isusability. The new version of GaussDB supports the Paxos consistency protocol, which better improves RTO. Under high-load business, DN failure switching can be stable within 10 seconds; last year, it launched the country's first dual-cluster strong consistency solution with ICBC to achieve It has completely isolated cluster-level faults with RPO=0, dual AZ active-active. This year, it further supports logical replication of dual clusters, and a new lossless and transparent switching solution for applications, so that large version upgrades of applications can be upgraded without any downtime, and active-standby switchover of applications can be done easily. Perception, truly realizing 24/7 uninterrupted mainframe business.
In terms ofperformance, the GaussDB storage engine can maintain system high performance without jitter even if the business is frequently updated for a long time. This is due to the joint creation of GaussDB and China Merchants Bank, which reconstructed the entire underlying storage engine and adopted an in-place update model. The biggest difference between it and the common engine based on continuous appending is that the traditional model is based on the underlying There will be very large jitter when garbage collection and memory are digitized. After actual testing, under heavy pressure, performance jitter can still be controlled at 3%, while storage space utilization increased by 17%.
In terms ofintelligence, the most troublesome thing for DBAs is how to quickly sense when the system is in a sub-healthy state. Problems can be identified and analyzed in a timely manner to facilitate further operations. As the first AI-Native database in China, GaussDB provides a full-process database intelligent experience from application development to operation and maintenance. It has a brand-new SQL Audit tool that completes automatic SQL audit during the development and verification stage to reduce the occurrence of sub-health. Once a sub-health state occurs, GaussDB can quickly sense the problem, identify real-time slow SQL, and further analyze the time-consuming points of the slow SQL to automatically diagnose whether it is blocked, so that operation and maintenance personnel can judge and kill it.
In some serious overload situations, the overload circuit breaker capability can automatically kill overloaded sessions to avoid slowing down the entire system due to individual slow SQL statements. In addition, DBMind's various operation and maintenance functions such as slow SQL root cause analysis, index recommendation, and anomaly detection also make DBAs more comfortable.
In addition to the hard-core technologies mentioned above, the new version of GaussDB has also done a lot of work on cost-effectiveness and smooth migration that customers are concerned about.
The first is the pressure on storage resources brought by massive data. Compression is one way, but a simple high compression ratio is not the goal we are pursuing. We are more concerned about how to allow businesses to use compression as painlessly as possible, and the performance will not be significantly reduced. It is best to control it at 5%, or even more Low performance impact, this is truly useful compression. We can both reduce resource usage and be less intrusive to the application.
Secondly, the new version will also support the multi-tenant capability of the kernel, helping users to quickly migrate to different tenant resources, making application use more flexible and resource utilization higher.
Finally, GaussDB has a one-stop migration solution that turns the originally uncertain migration work into a deterministic thing.
-
The first one is our UGO, which can scan and evaluate all applications of the existing system, telling us which ones are compatible, which ones are not compatible, and how to transform them. Now we have achieved 95% automation.
-
The second is DRS, which can achieve online zero-disruption migration and ensure zero data loss through data comparison.
-
The third one is traffic playback. Working with UGO, you can truly capture the traffic on the source database and play it back on the new database to avoid the common problem of incomplete coverage that everyone now encounters. I believe that only through the above-mentioned engineered and implementable solutions can we truly achieve large-scale replacement of domestic databases.
The development of databases, in addition to product innovation, is inseparable from the cooperation between industry, academia and research. Financial customers are an important starting point and foothold for the database, playing a key role in the development of GaussDB. We hope that more financial customers can open up more typical business scenarios of their own, design their own multi-location and multi-center solutions based on distributed architecture, form best practices, and accelerate the digital transformation of the industry.
That’s it for today’s sharing. Everyone is welcome to discuss and communicate together.