Since April 11, we have officially launched a series of live broadcasts with the theme of the future of database. After the first three live broadcasts, we have gradually described the future database in our eyes. Here are some text and video reviews from the third issue of “smart database, how to reduce your mental burden”~
What kind of world do we live in?
Traffic peaks that may occur at any time
Many business emergencies, may cause more than usual, or even dozens of times the impact of traffic. These traffic impacts every layer of the back-end architecture,But the database is the last lifeline and the most difficult to save。
In order to resist the traffic burst at any time and ensure that the business is not affected by the traffic. Physical resources must be purchased to support “expected” business peaks.
Cloud is changing everything
The common phenomenon in the industry is that the utilization rate of physical resources is only about 10% at ordinary times, and the most essential change brought about by the cloud is from “buying houses” to “renting houses”. Only the physical resources that meet the daily needs need need need need need need need need to be prepared. In case of traffic explosion, the “renting” method can be used to meet the needs in a short time to ensure normal operation. It’s also at the heart of the severless concept – let the data layer have itIntelligent scheduling capability。
The server less in the market mainly faces the computing layer, while the storage layer still relies on the traditional database. In fact, serverless should not assume that the underlying database runs on several machines. It should schedule and accurately calculate the cost according to workload and business flexibility. The precondition for the database to realize serverless is the scheduling ability. When the traffic and load change dramatically, the database can adjust its topology according to the business form. It is necessary to achieve such an effect. Not all databases have such a scheduling capability, and it is precisely with such preconditions that tidb can realize serverless at the database level.
In fact, we have been preparing for this direction since the beginning of tidb design. For example, if you want the database to have such flexible scheduling, you must have the ability of dynamic fragmentation, and tikv’s fragmentation strategy is designed for this purpose. Fast flexible scheduling also has preconditions, and the size of tikv slices is also designed for this purpose. Also in accuracy, independence and generality, the architecture of each part is designed.
In tidb 4.0, the concept car is on the road.
Of course, we can have different flexible scheduling methods for different scenarios
- Load based split balancing and replica adjustment
The number of replicas can be increased or decreased according to the real-time situation under different loads.
- Automatic node expansion
In the face of large-scale fluctuation of traffic, the number of nodes is automatically increased or decreased to ensure smooth and smooth operation of business.
- Automatic cold and hot separation, storage medium separation (WIP)
Some data of ordinary cluster are often accessed, and some data are not accessed for a long time. When elastic scheduling exists, the storage media will be adjusted automatically to reduce the cost.
- Autonomous hotspot isolation (WIP)
Flexible scheduling is used to segment the hot and cold nodes.
In general, tidb 4.0 can be summarized in two aspects: real time HTAP & serverless. In other words, tidb 4.0 is a self-help database.
Full video link: https://www.bilibili.com/vide…
Elite video links: https://www.bilibili.com/vide…
At 20:00 p.m. on Thursday, the final issue of the live broadcast will come to an end. The theme of this issue is “what?” we have collected interesting questions that you are very interested in
- After watching tidb’s previous attempts on serverless, I feel that DBA will be eliminated by the times in the near future. What suggestions do teachers have for DBA students to meet the needs of the times?
- In the scenario of e-commerce inventory deduction and concurrent processing (in extreme cases, seckill), what level of processing capacity can be achieved? Is there any problem with transaction control processing, which can be aimed at small and medium-sized e-commerce enterprises, and the concurrency is not very large?
- What are the most important optimization points in the implementation of raft by tidb?
- About ha and dual live database, as well as the introduction of visual data flow?
In the live room, we will answer these questions one by one, and welcome you to interact with us.
If you are interested in this live broadcast, click here, add tidb robot as a friend and reply [new features] to enter the live communication group
Welcome to pingcap official website for technical documents and Blogs: https://pingcap.com
If you have any questions about the use of tidb, you can also log in https://Asktug.com Search or post