Note:
This topic has been translated from a Chinese forum by GPT and might contain errors.Original topic: pd为什么需要ssd
|
username: TiDBer_hwEZA4rV
Why does PD need SSD? Is it just to prevent metadata migration overload in case of node failure?
|
username: TiDBer_hwEZA4rV
|
username: PD also has a large number of updates and write-to-disk operations.
|
username: The image is not visible. Please provide the text you need translated.
|
username: Not only does it need to write to disk during failover, but PD also needs to frequently read and persist data during normal operation. It has an integrated etcd, and although the data volume is not large, insufficient IO capability can affect PD performance, such as TSO allocation, TiCDC metadata access, and scheduling processing.
|
username: 
|
username: PD requests for TSO need to be written to disk, and a single I/O operation can last several minutes, but it still needs to be written to disk. Region information storage also needs to be written to disk. I feel that PD’s demand for SSD is not as strong as TiKV’s, but if available, it should be given. If not, prioritizing TiKV should be fine as well. This is just my personal opinion and has not been tested.
|
username: The main issues are still the reading of TSO and the synchronization of raft logs in the PD cluster.
|
username: PD is the brain; if PD runs efficiently, the entire cluster can operate stably. Otherwise, if the brain is sluggish, it will also affect the operational efficiency of other components.
|
username: PD is a critical node in the cluster, and all cluster metadata is stored in PD. During normal operation, there will be continuous read and write operations, so the IO consumption is relatively high, and SSDs are required. It is possible to run without SSDs, but the cluster performance and various processing efficiencies will be significantly compromised.
|
username: This topic will be automatically closed 60 days after the last reply. No new replies are allowed.