Note:
This topic has been translated from a Chinese forum by GPT and might contain errors.Original topic: tidb的日吞吐和秒级查询
|
username: TiDBer_4Jz4pjD8
Could you please share with the community what kind of cluster configuration can achieve 1 billion queries with second-level read performance? Additionally, what configurations are generally associated with write performance?
There are too many factors affecting this kind of problem, it’s impossible to make an assessment off the top of one’s head. It’s best to run more stress tests based on the actual situation and then make calculations. The configuration requirements related to write performance, from highest to lowest (personal opinion): disk, network, CPU. For large-scale data writing, start with NVMe SSDs and 10 Gigabit network cards.
Is 1 billion queries per second? Per minute? Per hour?
What kind of queries? select 1? point queries? join queries?
Personally, I think disk performance and CPU are two major factors affecting write performance.
It is best to run it with actual business operations; practical testing is the most reliable. Each business system is different, with different table structures, SQL writing standards, and SQL optimization levels, so it’s impossible to give you an evaluation. For single table performance, just find 5 machines and test it practically; it should be quite easy.
I don’t quite understand what “billion queries per second read” means.
For TPCC data generation, generating 1 billion records for stress testing should be sufficient. This is highly related to your hardware and parameter settings.
The description is unclear. What is the time unit for 1 billion queries? Constructing queries, such as point queries, is not an issue at the second level. Write performance is directly related to the TiKV disk, followed by network, hotspots, etc.
|
username: 
|
username: 
|
username: 
|
username: 
|
username: 
|
username: