Pika 3.5 参数优化手册 #1970
AlexStocks started this conversation in General
Pika 3.5 参数优化手册 #1970
AlexStocks
Sep 11, 2023 · 1 comment
Return to top
Discussion options
Quote reply
edited
[
AlexStocks
](https://github.com/AlexStocks)[Sep 11, 2023](https://github.com/OpenAtomFoundation/pika/discussions/1970#discussion-5614484)
Maintainer
-
- 1 不要使用压缩，snappy::internaluncompress 占用较高；
- 2 关闭 binlog；
- 3 调高 write-buffer-size；
- 4 Cluster 模式下把 slot 数目调整为 128，不要使用 1024；
- 5 调大 compaction 线程数目；
- 6 缓存 使用场景下，可以动态关闭 WAL；
|
Beta Was this translation helpful? Give feedback.
1 You must be logged in to vote
All reactions
Comment options
Quote reply
edited by Mixficsol
[
AlexStocks
](https://github.com/AlexStocks)[Oct 23, 2023](https://github.com/OpenAtomFoundation/pika/discussions/1970#discussioncomment-7357996)
Maintainer Author
-
from #2008 (comment): based on v3.5.1， you can optimize some parameters as follows. 1 Adjust max-background-flushes and max-background-compaction to 4 to increase the number of threads for compaction and flushdb. Adjust max-background-jobs to 8.
2 Adjust the block-cache to be 60% of the memory allocated to the Pika instance.
3 Uncomment share-block-cache and set it to yes.
Adjust thread-num to 80% of the CPU cores of the Pika instance.
5 Adjust thread-pool-size to CPU cores * 1.5.
6 Set rate-limiter-bandwidth to 2000Mib/s for gigabit Ethernet and 10000Mib/s for 10-gigabit Ethernet in Chinese: max-background-flushes 和 max-backgroud-compaction可以调整到 4，提高 compaction 和 flushdb 的线程数， max-backgroumd-jobs调整到 8block-cache的大小可以调整为分给 Pika 实例内存的 60%share-block-cache解除注释，改为yesthread-num调整为 Pika 实例的 CPU 核数的 80%thread-pool-size调整为 CPU 核数 * 1.5rate-limiter-bandwidth：千兆网卡2000Mib/s 万兆网卡：10000Mib/s
|