[
https://issues.apache.org/jira/browse/HDDS-11230?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17868738#comment-17868738
]
Wei-Chiu Chuang commented on HDDS-11230:
----------------------------------------
It looks like we have one block cache per column family (table). It would make
sense to make the block cache for certain tables larger (keyTable, fileTable)
Or they can also share one big block cache.
> Make RocksDB cache configurable in Ozone Manager
> ------------------------------------------------
>
> Key: HDDS-11230
> URL: https://issues.apache.org/jira/browse/HDDS-11230
> Project: Apache Ozone
> Issue Type: Improvement
> Components: OM
> Reporter: Duong
> Priority: Major
>
> Ozone Manager's performance heavily depends on RocksDB performance. RocksDB
> cache plays an important role in RocksDB's read performance, especially when
> a client has to host OM's metadata in not-so-fast hardware.
> Today, OM RocksDB cache is fixed at 256MB (see DBProfile), this is
> unreasonable because the cache size should depend on the client use-case,
> i.e. how big is their active key space. For example, if a client always has
> 1M of hot keys daily, and each key metadata is 1kb, having a cache size of
> 1GB makes a lot of sense.
> Ref:
> * [https://github.com/facebook/rocksdb/wiki/Block-Cache]
> * [https://github.com/facebook/rocksdb/wiki/Performance-Benchmarks]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
