/ Asahi/Guangzhou, June 16, 2022 -- Recently, Tencent cloud native database TDSQL-C released a new version, which has been comprehensively upgraded in cloud native architecture, basic hardware capabilities, self-developed kernel and other aspects. Core features such as level 2 cache based on Intel ® Outton ™ persistent memory, full link RDMA network, 7*24 continuous lock-free backup are covered. The performance test results show that the new version of TDSQL-C comprehensively outperforms other cloud native database products in the industry and achieves a 200% performance improvement compared with traditional cloud databases in full-cache scenarios and large data sets scenarios. In addition, TDSQL-C is 1/10 the price of commercial databases, making it the best choice for enterprise cloud.
TDSQL-C is a new-generation cloud native relational database developed by Tencent Cloud. It provides enterprises with extreme elasticity, high performance, high availability, high reliability and security database services, and can achieve high throughput of over one million QPS, massive distributed intelligent storage, and Serverless second scale.
The new version of TDSQL-C released this time is equipped with TXSQL, Tencent's self-developed database kernel. TXSQL has carried out a lot of optimization for cloud native database scenarios, including the optimization of dirt brushing mechanism, transaction conflict optimization and redo log brushing mechanism optimization, which greatly improves database performance and can better meet enterprises' ultimate pursuit of database performance.
Combined with new hardware features, TDSQL-C continuously optimizes product architecture and technical implementation. Through joint innovation with the Intel technology team, refactoring the secondary cache design with the latest generation Intel ® Xeon ® expandable processor and Intel ® Odeon ™ Persistent Memory (PMem) hardware features more than tripled read/write performance in the IO bound scenario. In addition, TDSQL-C works with Intel to optimize storage solution design in multiple directions, such as adding polling, algorithm optimization, lock elimination and other mechanisms, optimize storage engine, and introduce SPDK development kit provided by Intel to optimize IOPS and delay performance of NVMe solid-state disk.
The new version of TDSQL-C comprehensively upgrades the network architecture. It uses a full-link RDMA network and relies on features such as zero copy, kernel bypass, and no CPU interference to further optimize the system performance of the critical path between the storage layer and the computing layer, as well as multiple copies of the storage layer. The request latency is reduced by up to 80% and I/O performance is no longer a bottleneck.
In exploring the actual performance of the new version, testers performed performance tests against write only, read only, and mixed read and write scenarios using the benchmark tool sysbench and a fully simulated business production environment. The report shows that TDSQL-C is leading similar products in the market and can fully meet the requirements of enterprises for database performance. The test data in the report is only the results of a single node. TDSQL-C MySQL Edition supports second-level expansion, which can help enterprises achieve more than 1500% read performance.
In the scenarios of instant kill, recommendation, live video broadcast of e-commerce business, and hot data access of game business, it is usually necessary to put all the data into the cache, and there is no need to read or write the disk to update the cache during the query process. The test report shows that in the full cache scenario, the larger the instance size, the more obvious the performance advantage of TDSQL-C MySQL. After the 32 core specification, the performance of traditional cloud database MySQL in the read-only scenario has reached the bottleneck, but the TDSQL-C MySQL edition can further improve QPS with the increase of CPU. Compared with other cloud native databases, the performance of TDSQL-C MySQL edition can also be maintained. In terms of resource utilization, TDSQL-C MySQL edition can achieve more than 90% CPU utilization of compute nodes, and there is almost no jitter on request delay rtt.
Compared with the full cache scenario, the large data set scenario occurs more frequently in real business use. When massive data is used, disks and caches need to be read and written during query and update. The performance of TDSQL-C MySQL edition exceeds that of similar products in the market in read-only scenarios with large data sets and in general scenarios with mixed read and write. With the increase of specifications, the performance advantage becomes more and more obvious.
Cloud native database TDSQL-C MySQL version adopts the design concept of "storage and computing separation", supporting the maximum of over 400TB single library mass data storage. With the application of this feature, more and more enterprises no longer use the method of database and table to store historical archived data, resulting in the phenomenon of "super large table". With 4 billion records in a single table and 1TB of data storage space, TDSQL-C MySQL Edition still performs well. In mixed read/write scenarios, with the increase of database specifications, the performance of TDSQL-C MySQL version is 200% higher than that of traditional cloud database MySQL and has obvious advantages over other cloud native databases.
At present, TDSQL-C has strongly supported Tencent's massive ace business, such as wechat red envelope, wechat Pay, QQ music, Tencent Conference, etc. Internally, TDSQL-C has provided stable and reliable enterprise-level cloud database services for users in different industries, such as Pinduoduo, B Station, Weimeng, Xinjingtao, etc. With a full set of database operation and maintenance solutions such as backup, recovery, monitoring, rapid expansion and data transmission, IT helps enterprises simplify IT operation and maintenance work and focus more on business development.
As the pioneer of cloud native technology, Tencent cloud database kernel team spare no effort in the production and research co-construction and ecological integration, focusing on the market and customer needs, to better implement the technical concept, through in-depth cooperation with Intel and the open source community, continue to contribute technical strength, to bring new business value and technical advantages to all walks of life.
